ruby/vm_backtrace.c

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

1834 строки
48 KiB
C
Исходник Обычный вид История

/**********************************************************************
vm_backtrace.c -
$Author: ko1 $
created at: Sun Jun 03 00:14:20 2012
Copyright (C) 1993-2012 Yukihiro Matsumoto
**********************************************************************/
#include "eval_intern.h"
#include "internal.h"
#include "internal/class.h"
#include "internal/error.h"
#include "internal/vm.h"
#include "iseq.h"
#include "ruby/debug.h"
#include "ruby/encoding.h"
#include "vm_core.h"
static VALUE rb_cBacktrace;
static VALUE rb_cBacktraceLocation;
static VALUE
id2str(ID id)
{
VALUE str = rb_id2str(id);
if (!str) return Qnil;
return str;
}
#define rb_id2str(id) id2str(id)
Improve performance of partial backtraces Previously, backtrace_each fully populated the rb_backtrace_t with all backtrace frames, even if caller only requested a partial backtrace (e.g. Kernel#caller_locations(1, 1)). This changes backtrace_each to only add the requested frames to the rb_backtrace_t. To do this, backtrace_each needs to be passed the starting frame and number of frames values passed to Kernel#caller or #caller_locations. backtrace_each works from the top of the stack to the bottom, where the bottom is the current frame. Due to how the location for cfuncs is tracked using the location of the previous iseq, we need to store an extra frame for the previous iseq if we are limiting the backtrace and final backtrace frame (the first one stored) would be a cfunc and not an iseq. To limit the amount of work in this case, while scanning until the start of the requested backtrace, for each iseq, store the cfp. If the first backtrace frame we care about is a cfunc, use the stored cfp to find the related iseq. Use a function pointer to handle the storage of the cfp in the iteration arg, and also store the location of the extra frame in the iteration arg. backtrace_each needs to return int instead of void in order to signal when a starting frame larger than backtrace size is given, as caller and caller_locations needs to return nil and not the empty array in these cases. To handle cases where a range is provided with a negative end, and the backtrace size is needed to calculate the result to pass to rb_range_beg_len, add a backtrace_size static function to calculate the size, which copies the logic from backtrace_each. As backtrace_each only adds the backtrace lines requested, backtrace_to_*_ary can be simplified to always operate on the entire backtrace. Previously, caller_locations(1,1) was about 6.2 times slower for an 800 deep callstack compared to an empty callstack. With this new approach, it is only 1.3 times slower. It will always be somewhat slower as it still needs to scan the cfps from the top of the stack until it finds the first requested backtrace frame. This initializes the backtrace memory to zero. I do not think this is necessary, as from my analysis, nothing during the setting of the backtrace entries can cause a garbage collection, but it seems the safest approach, and it's unlikely the performance decrease is significant. This removes the rb_backtrace_t backtrace_base member. backtrace and backtrace_base were initialized to the same value, and neither is modified, so it doesn't make sense to have two pointers. This also removes LOCATION_TYPE_IFUNC from vm_backtrace.c, as the value is never set. Fixes [Bug #17031]
2020-08-28 01:17:36 +03:00
#define BACKTRACE_START 0
#define ALL_BACKTRACE_LINES -1
inline static int
calc_pos(const rb_iseq_t *iseq, const VALUE *pc, int *lineno, int *node_id)
{
VM_ASSERT(iseq);
if (pc == NULL) {
if (ISEQ_BODY(iseq)->type == ISEQ_TYPE_TOP) {
VM_ASSERT(! ISEQ_BODY(iseq)->local_table);
VM_ASSERT(! ISEQ_BODY(iseq)->local_table_size);
return 0;
}
if (lineno) *lineno = ISEQ_BODY(iseq)->location.first_lineno;
#ifdef USE_ISEQ_NODE_ID
if (node_id) *node_id = -1;
#endif
return 1;
}
else {
VM_ASSERT(ISEQ_BODY(iseq));
VM_ASSERT(ISEQ_BODY(iseq)->iseq_encoded);
VM_ASSERT(ISEQ_BODY(iseq)->iseq_size);
ptrdiff_t n = pc - ISEQ_BODY(iseq)->iseq_encoded;
VM_ASSERT(n <= ISEQ_BODY(iseq)->iseq_size);
VM_ASSERT(n >= 0);
ASSUME(n >= 0);
size_t pos = n; /* no overflow */
if (LIKELY(pos)) {
/* use pos-1 because PC points next instruction at the beginning of instruction */
pos--;
}
#if VMDEBUG && defined(HAVE_BUILTIN___BUILTIN_TRAP)
else {
/* SDR() is not possible; that causes infinite loop. */
rb_print_backtrace();
__builtin_trap();
}
#endif
if (lineno) *lineno = rb_iseq_line_no(iseq, pos);
#ifdef USE_ISEQ_NODE_ID
if (node_id) *node_id = rb_iseq_node_id(iseq, pos);
#endif
return 1;
}
}
inline static int
calc_lineno(const rb_iseq_t *iseq, const VALUE *pc)
{
int lineno;
if (calc_pos(iseq, pc, &lineno, NULL)) return lineno;
return 0;
}
#ifdef USE_ISEQ_NODE_ID
inline static int
calc_node_id(const rb_iseq_t *iseq, const VALUE *pc)
{
int node_id;
if (calc_pos(iseq, pc, NULL, &node_id)) return node_id;
return -1;
}
#endif
int
rb_vm_get_sourceline(const rb_control_frame_t *cfp)
{
if (VM_FRAME_RUBYFRAME_P(cfp) && cfp->iseq) {
const rb_iseq_t *iseq = cfp->iseq;
int line = calc_lineno(iseq, cfp->pc);
if (line != 0) {
return line;
}
else {
return ISEQ_BODY(iseq)->location.first_lineno;
}
}
else {
return 0;
}
}
typedef struct rb_backtrace_location_struct {
enum LOCATION_TYPE {
LOCATION_TYPE_ISEQ = 1,
LOCATION_TYPE_CFUNC,
} type;
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
const rb_iseq_t *iseq;
const VALUE *pc;
ID mid;
} rb_backtrace_location_t;
struct valued_frame_info {
rb_backtrace_location_t *loc;
VALUE btobj;
};
static void
location_mark(void *ptr)
{
struct valued_frame_info *vfi = (struct valued_frame_info *)ptr;
rb_gc_mark(vfi->btobj);
}
static void
location_mark_entry(rb_backtrace_location_t *fi)
{
switch (fi->type) {
case LOCATION_TYPE_ISEQ:
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
rb_gc_mark_movable((VALUE)fi->iseq);
break;
case LOCATION_TYPE_CFUNC:
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
if (fi->iseq) {
rb_gc_mark_movable((VALUE)fi->iseq);
}
break;
default:
break;
}
}
static size_t
location_memsize(const void *ptr)
{
/* rb_backtrace_location_t *fi = (rb_backtrace_location_t *)ptr; */
return sizeof(rb_backtrace_location_t);
}
static const rb_data_type_t location_data_type = {
"frame_info",
{location_mark, RUBY_TYPED_DEFAULT_FREE, location_memsize,},
0, 0, RUBY_TYPED_FREE_IMMEDIATELY | RUBY_TYPED_WB_PROTECTED
};
int
rb_frame_info_p(VALUE obj)
{
return rb_typeddata_is_kind_of(obj, &location_data_type);
}
static inline rb_backtrace_location_t *
location_ptr(VALUE locobj)
{
struct valued_frame_info *vloc;
GetCoreDataFromValue(locobj, struct valued_frame_info, vloc);
return vloc->loc;
}
static int
location_lineno(rb_backtrace_location_t *loc)
{
switch (loc->type) {
case LOCATION_TYPE_ISEQ:
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
return calc_lineno(loc->iseq, loc->pc);
case LOCATION_TYPE_CFUNC:
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
if (loc->iseq && loc->pc) {
return calc_lineno(loc->iseq, loc->pc);
}
return 0;
default:
rb_bug("location_lineno: unreachable");
UNREACHABLE;
}
}
/*
* Returns the line number of this frame.
*
* For example, using +caller_locations.rb+ from Thread::Backtrace::Location
*
* loc = c(0..1).first
* loc.lineno #=> 2
*/
static VALUE
location_lineno_m(VALUE self)
{
return INT2FIX(location_lineno(location_ptr(self)));
}
static VALUE
location_label(rb_backtrace_location_t *loc)
{
switch (loc->type) {
case LOCATION_TYPE_ISEQ:
return ISEQ_BODY(loc->iseq)->location.label;
case LOCATION_TYPE_CFUNC:
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
return rb_id2str(loc->mid);
default:
rb_bug("location_label: unreachable");
UNREACHABLE;
}
}
/*
* Returns the label of this frame.
*
* Usually consists of method, class, module, etc names with decoration.
*
* Consider the following example:
*
* def foo
* puts caller_locations(0).first.label
*
* 1.times do
* puts caller_locations(0).first.label
*
* 1.times do
* puts caller_locations(0).first.label
* end
*
* end
* end
*
* The result of calling +foo+ is this:
*
* label: foo
* label: block in foo
* label: block (2 levels) in foo
*
*/
static VALUE
location_label_m(VALUE self)
{
return location_label(location_ptr(self));
}
static VALUE
location_base_label(rb_backtrace_location_t *loc)
{
switch (loc->type) {
case LOCATION_TYPE_ISEQ:
return ISEQ_BODY(loc->iseq)->location.base_label;
case LOCATION_TYPE_CFUNC:
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
return rb_id2str(loc->mid);
default:
rb_bug("location_base_label: unreachable");
UNREACHABLE;
}
}
/*
* Returns the base label of this frame.
*
* Usually same as #label, without decoration.
*/
static VALUE
location_base_label_m(VALUE self)
{
return location_base_label(location_ptr(self));
}
static const rb_iseq_t *
location_iseq(rb_backtrace_location_t *loc)
{
switch (loc->type) {
case LOCATION_TYPE_ISEQ:
return loc->iseq;
case LOCATION_TYPE_CFUNC:
return loc->iseq;
default:
rb_bug("location_iseq: unreachable");
UNREACHABLE;
}
}
/*
* Returns the file name of this frame. This will generally be an absolute
* path, unless the frame is in the main script, in which case it will be the
* script location passed on the command line.
*
* For example, using +caller_locations.rb+ from Thread::Backtrace::Location
*
* loc = c(0..1).first
* loc.path #=> caller_locations.rb
*/
static VALUE
location_path_m(VALUE self)
{
const rb_iseq_t *iseq = location_iseq(location_ptr(self));
return iseq ? rb_iseq_path(iseq) : Qnil;
}
#ifdef USE_ISEQ_NODE_ID
static int
location_node_id(rb_backtrace_location_t *loc)
{
switch (loc->type) {
case LOCATION_TYPE_ISEQ:
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
return calc_node_id(loc->iseq, loc->pc);
case LOCATION_TYPE_CFUNC:
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
if (loc->iseq && loc->pc) {
return calc_node_id(loc->iseq, loc->pc);
}
return -1;
default:
rb_bug("location_node_id: unreachable");
UNREACHABLE;
}
}
#endif
int
rb_get_node_id_from_frame_info(VALUE obj)
{
#ifdef USE_ISEQ_NODE_ID
rb_backtrace_location_t *loc = location_ptr(obj);
return location_node_id(loc);
#else
return -1;
#endif
}
const rb_iseq_t *
rb_get_iseq_from_frame_info(VALUE obj)
{
rb_backtrace_location_t *loc = location_ptr(obj);
const rb_iseq_t *iseq = location_iseq(loc);
return iseq;
}
static VALUE
location_realpath(rb_backtrace_location_t *loc)
{
switch (loc->type) {
case LOCATION_TYPE_ISEQ:
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
return rb_iseq_realpath(loc->iseq);
case LOCATION_TYPE_CFUNC:
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
if (loc->iseq) {
return rb_iseq_realpath(loc->iseq);
}
return Qnil;
default:
rb_bug("location_realpath: unreachable");
UNREACHABLE;
}
}
/*
* Returns the full file path of this frame.
*
* Same as #path, except that it will return absolute path
* even if the frame is in the main script.
*/
static VALUE
location_absolute_path_m(VALUE self)
{
return location_realpath(location_ptr(self));
}
static VALUE
location_format(VALUE file, int lineno, VALUE name)
{
VALUE s = rb_enc_sprintf(rb_enc_compatible(file, name), "%s", RSTRING_PTR(file));
if (lineno != 0) {
rb_str_catf(s, ":%d", lineno);
}
rb_str_cat_cstr(s, ":in ");
if (NIL_P(name)) {
rb_str_cat_cstr(s, "unknown method");
}
else {
rb_str_catf(s, "`%s'", RSTRING_PTR(name));
}
return s;
}
static VALUE
location_to_str(rb_backtrace_location_t *loc)
{
VALUE file, name;
int lineno;
switch (loc->type) {
case LOCATION_TYPE_ISEQ:
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
file = rb_iseq_path(loc->iseq);
name = ISEQ_BODY(loc->iseq)->location.label;
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
lineno = calc_lineno(loc->iseq, loc->pc);
break;
case LOCATION_TYPE_CFUNC:
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
if (loc->iseq && loc->pc) {
file = rb_iseq_path(loc->iseq);
lineno = calc_lineno(loc->iseq, loc->pc);
}
else {
file = GET_VM()->progname;
lineno = 0;
}
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
name = rb_id2str(loc->mid);
break;
default:
rb_bug("location_to_str: unreachable");
}
return location_format(file, lineno, name);
}
/*
* Returns a Kernel#caller style string representing this frame.
*/
static VALUE
location_to_str_m(VALUE self)
{
return location_to_str(location_ptr(self));
}
/*
* Returns the same as calling +inspect+ on the string representation of
* #to_str
*/
static VALUE
location_inspect_m(VALUE self)
{
return rb_str_inspect(location_to_str(location_ptr(self)));
}
typedef struct rb_backtrace_struct {
rb_backtrace_location_t *backtrace;
int backtrace_size;
VALUE strary;
VALUE locary;
} rb_backtrace_t;
static void
backtrace_mark(void *ptr)
{
rb_backtrace_t *bt = (rb_backtrace_t *)ptr;
size_t i, s = bt->backtrace_size;
for (i=0; i<s; i++) {
location_mark_entry(&bt->backtrace[i]);
}
rb_gc_mark_movable(bt->strary);
rb_gc_mark_movable(bt->locary);
}
static void
backtrace_free(void *ptr)
{
rb_backtrace_t *bt = (rb_backtrace_t *)ptr;
Improve performance of partial backtraces Previously, backtrace_each fully populated the rb_backtrace_t with all backtrace frames, even if caller only requested a partial backtrace (e.g. Kernel#caller_locations(1, 1)). This changes backtrace_each to only add the requested frames to the rb_backtrace_t. To do this, backtrace_each needs to be passed the starting frame and number of frames values passed to Kernel#caller or #caller_locations. backtrace_each works from the top of the stack to the bottom, where the bottom is the current frame. Due to how the location for cfuncs is tracked using the location of the previous iseq, we need to store an extra frame for the previous iseq if we are limiting the backtrace and final backtrace frame (the first one stored) would be a cfunc and not an iseq. To limit the amount of work in this case, while scanning until the start of the requested backtrace, for each iseq, store the cfp. If the first backtrace frame we care about is a cfunc, use the stored cfp to find the related iseq. Use a function pointer to handle the storage of the cfp in the iteration arg, and also store the location of the extra frame in the iteration arg. backtrace_each needs to return int instead of void in order to signal when a starting frame larger than backtrace size is given, as caller and caller_locations needs to return nil and not the empty array in these cases. To handle cases where a range is provided with a negative end, and the backtrace size is needed to calculate the result to pass to rb_range_beg_len, add a backtrace_size static function to calculate the size, which copies the logic from backtrace_each. As backtrace_each only adds the backtrace lines requested, backtrace_to_*_ary can be simplified to always operate on the entire backtrace. Previously, caller_locations(1,1) was about 6.2 times slower for an 800 deep callstack compared to an empty callstack. With this new approach, it is only 1.3 times slower. It will always be somewhat slower as it still needs to scan the cfps from the top of the stack until it finds the first requested backtrace frame. This initializes the backtrace memory to zero. I do not think this is necessary, as from my analysis, nothing during the setting of the backtrace entries can cause a garbage collection, but it seems the safest approach, and it's unlikely the performance decrease is significant. This removes the rb_backtrace_t backtrace_base member. backtrace and backtrace_base were initialized to the same value, and neither is modified, so it doesn't make sense to have two pointers. This also removes LOCATION_TYPE_IFUNC from vm_backtrace.c, as the value is never set. Fixes [Bug #17031]
2020-08-28 01:17:36 +03:00
if (bt->backtrace) ruby_xfree(bt->backtrace);
ruby_xfree(bt);
}
static void
location_update_entry(rb_backtrace_location_t *fi)
{
switch (fi->type) {
case LOCATION_TYPE_ISEQ:
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
fi->iseq = (rb_iseq_t*)rb_gc_location((VALUE)fi->iseq);
break;
case LOCATION_TYPE_CFUNC:
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
if (fi->iseq) {
fi->iseq = (rb_iseq_t*)rb_gc_location((VALUE)fi->iseq);
}
break;
default:
break;
}
}
static void
backtrace_update(void *ptr)
{
rb_backtrace_t *bt = (rb_backtrace_t *)ptr;
size_t i, s = bt->backtrace_size;
for (i=0; i<s; i++) {
location_update_entry(&bt->backtrace[i]);
}
bt->strary = rb_gc_location(bt->strary);
bt->locary = rb_gc_location(bt->locary);
}
static size_t
backtrace_memsize(const void *ptr)
{
rb_backtrace_t *bt = (rb_backtrace_t *)ptr;
return sizeof(rb_backtrace_t) + sizeof(rb_backtrace_location_t) * bt->backtrace_size;
}
static const rb_data_type_t backtrace_data_type = {
"backtrace",
{backtrace_mark, backtrace_free, backtrace_memsize, backtrace_update},
0, 0, RUBY_TYPED_FREE_IMMEDIATELY | RUBY_TYPED_WB_PROTECTED
};
int
rb_backtrace_p(VALUE obj)
{
return rb_typeddata_is_kind_of(obj, &backtrace_data_type);
}
static VALUE
backtrace_alloc(VALUE klass)
{
rb_backtrace_t *bt;
VALUE obj = TypedData_Make_Struct(klass, rb_backtrace_t, &backtrace_data_type, bt);
return obj;
}
Improve performance of partial backtraces Previously, backtrace_each fully populated the rb_backtrace_t with all backtrace frames, even if caller only requested a partial backtrace (e.g. Kernel#caller_locations(1, 1)). This changes backtrace_each to only add the requested frames to the rb_backtrace_t. To do this, backtrace_each needs to be passed the starting frame and number of frames values passed to Kernel#caller or #caller_locations. backtrace_each works from the top of the stack to the bottom, where the bottom is the current frame. Due to how the location for cfuncs is tracked using the location of the previous iseq, we need to store an extra frame for the previous iseq if we are limiting the backtrace and final backtrace frame (the first one stored) would be a cfunc and not an iseq. To limit the amount of work in this case, while scanning until the start of the requested backtrace, for each iseq, store the cfp. If the first backtrace frame we care about is a cfunc, use the stored cfp to find the related iseq. Use a function pointer to handle the storage of the cfp in the iteration arg, and also store the location of the extra frame in the iteration arg. backtrace_each needs to return int instead of void in order to signal when a starting frame larger than backtrace size is given, as caller and caller_locations needs to return nil and not the empty array in these cases. To handle cases where a range is provided with a negative end, and the backtrace size is needed to calculate the result to pass to rb_range_beg_len, add a backtrace_size static function to calculate the size, which copies the logic from backtrace_each. As backtrace_each only adds the backtrace lines requested, backtrace_to_*_ary can be simplified to always operate on the entire backtrace. Previously, caller_locations(1,1) was about 6.2 times slower for an 800 deep callstack compared to an empty callstack. With this new approach, it is only 1.3 times slower. It will always be somewhat slower as it still needs to scan the cfps from the top of the stack until it finds the first requested backtrace frame. This initializes the backtrace memory to zero. I do not think this is necessary, as from my analysis, nothing during the setting of the backtrace entries can cause a garbage collection, but it seems the safest approach, and it's unlikely the performance decrease is significant. This removes the rb_backtrace_t backtrace_base member. backtrace and backtrace_base were initialized to the same value, and neither is modified, so it doesn't make sense to have two pointers. This also removes LOCATION_TYPE_IFUNC from vm_backtrace.c, as the value is never set. Fixes [Bug #17031]
2020-08-28 01:17:36 +03:00
static long
backtrace_size(const rb_execution_context_t *ec)
{
const rb_control_frame_t *last_cfp = ec->cfp;
const rb_control_frame_t *start_cfp = RUBY_VM_END_CONTROL_FRAME(ec);
if (start_cfp == NULL) {
return -1;
}
start_cfp =
RUBY_VM_NEXT_CONTROL_FRAME(
RUBY_VM_NEXT_CONTROL_FRAME(start_cfp)); /* skip top frames */
if (start_cfp < last_cfp) {
return 0;
}
return start_cfp - last_cfp + 1;
}
static bool
is_internal_location(const rb_control_frame_t *cfp)
{
static const char prefix[] = "<internal:";
const size_t prefix_len = sizeof(prefix) - 1;
VALUE file = rb_iseq_path(cfp->iseq);
return strncmp(prefix, RSTRING_PTR(file), prefix_len) == 0;
}
static void
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
bt_update_cfunc_loc(unsigned long cfunc_counter, rb_backtrace_location_t *cfunc_loc, const rb_iseq_t *iseq, const VALUE *pc)
{
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
for (; cfunc_counter > 0; cfunc_counter--, cfunc_loc--) {
cfunc_loc->iseq = iseq;
cfunc_loc->pc = pc;
}
}
static VALUE location_create(rb_backtrace_location_t *srcloc, void *btobj);
static void
bt_yield_loc(rb_backtrace_location_t *loc, long num_frames, VALUE btobj)
{
for (; num_frames > 0; num_frames--, loc++) {
rb_yield(location_create(loc, (void *)btobj));
}
}
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
static VALUE
rb_ec_partial_backtrace_object(const rb_execution_context_t *ec, long start_frame, long num_frames, int* start_too_large, bool skip_internal, bool do_yield)
{
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
const rb_control_frame_t *cfp = ec->cfp;
const rb_control_frame_t *end_cfp = RUBY_VM_END_CONTROL_FRAME(ec);
ptrdiff_t size;
rb_backtrace_t *bt;
VALUE btobj = backtrace_alloc(rb_cBacktrace);
rb_backtrace_location_t *loc = NULL;
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
unsigned long cfunc_counter = 0;
GetCoreDataFromValue(btobj, rb_backtrace_t, bt);
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
// In the case the thread vm_stack or cfp is not initialized, there is no backtrace.
if (end_cfp == NULL) {
num_frames = 0;
}
else {
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
end_cfp = RUBY_VM_NEXT_CONTROL_FRAME(end_cfp);
/*
* top frame (dummy) <- RUBY_VM_END_CONTROL_FRAME
* top frame (dummy) <- end_cfp
* top frame <- main script
* top frame
* ...
* 2nd frame <- lev:0
* current frame <- ec->cfp
*/
size = end_cfp - cfp + 1;
if (size < 0) {
num_frames = 0;
}
else if (num_frames < 0 || num_frames > size) {
num_frames = size;
}
}
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
bt->backtrace = ZALLOC_N(rb_backtrace_location_t, num_frames);
bt->backtrace_size = 0;
if (num_frames == 0) {
if (start_too_large) *start_too_large = 0;
return btobj;
Improve performance of partial backtraces Previously, backtrace_each fully populated the rb_backtrace_t with all backtrace frames, even if caller only requested a partial backtrace (e.g. Kernel#caller_locations(1, 1)). This changes backtrace_each to only add the requested frames to the rb_backtrace_t. To do this, backtrace_each needs to be passed the starting frame and number of frames values passed to Kernel#caller or #caller_locations. backtrace_each works from the top of the stack to the bottom, where the bottom is the current frame. Due to how the location for cfuncs is tracked using the location of the previous iseq, we need to store an extra frame for the previous iseq if we are limiting the backtrace and final backtrace frame (the first one stored) would be a cfunc and not an iseq. To limit the amount of work in this case, while scanning until the start of the requested backtrace, for each iseq, store the cfp. If the first backtrace frame we care about is a cfunc, use the stored cfp to find the related iseq. Use a function pointer to handle the storage of the cfp in the iteration arg, and also store the location of the extra frame in the iteration arg. backtrace_each needs to return int instead of void in order to signal when a starting frame larger than backtrace size is given, as caller and caller_locations needs to return nil and not the empty array in these cases. To handle cases where a range is provided with a negative end, and the backtrace size is needed to calculate the result to pass to rb_range_beg_len, add a backtrace_size static function to calculate the size, which copies the logic from backtrace_each. As backtrace_each only adds the backtrace lines requested, backtrace_to_*_ary can be simplified to always operate on the entire backtrace. Previously, caller_locations(1,1) was about 6.2 times slower for an 800 deep callstack compared to an empty callstack. With this new approach, it is only 1.3 times slower. It will always be somewhat slower as it still needs to scan the cfps from the top of the stack until it finds the first requested backtrace frame. This initializes the backtrace memory to zero. I do not think this is necessary, as from my analysis, nothing during the setting of the backtrace entries can cause a garbage collection, but it seems the safest approach, and it's unlikely the performance decrease is significant. This removes the rb_backtrace_t backtrace_base member. backtrace and backtrace_base were initialized to the same value, and neither is modified, so it doesn't make sense to have two pointers. This also removes LOCATION_TYPE_IFUNC from vm_backtrace.c, as the value is never set. Fixes [Bug #17031]
2020-08-28 01:17:36 +03:00
}
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
for (; cfp != end_cfp && (bt->backtrace_size < num_frames); cfp = RUBY_VM_PREVIOUS_CONTROL_FRAME(cfp)) {
if (cfp->iseq) {
if (cfp->pc) {
if (start_frame > 0) {
start_frame--;
}
else if (!skip_internal || !is_internal_location(cfp)) {
const rb_iseq_t *iseq = cfp->iseq;
const VALUE *pc = cfp->pc;
loc = &bt->backtrace[bt->backtrace_size++];
loc->type = LOCATION_TYPE_ISEQ;
RB_OBJ_WRITE(btobj, &loc->iseq, iseq);
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
loc->pc = pc;
bt_update_cfunc_loc(cfunc_counter, loc-1, iseq, pc);
if (do_yield) {
bt_yield_loc(loc - cfunc_counter, cfunc_counter+1, btobj);
}
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
cfunc_counter = 0;
}
}
}
else {
VM_ASSERT(RUBYVM_CFUNC_FRAME_P(cfp));
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
if (start_frame > 0) {
start_frame--;
}
else {
loc = &bt->backtrace[bt->backtrace_size++];
loc->type = LOCATION_TYPE_CFUNC;
loc->iseq = NULL;
loc->pc = NULL;
loc->mid = rb_vm_frame_method_entry(cfp)->def->original_id;
cfunc_counter++;
}
}
Improve performance of partial backtraces Previously, backtrace_each fully populated the rb_backtrace_t with all backtrace frames, even if caller only requested a partial backtrace (e.g. Kernel#caller_locations(1, 1)). This changes backtrace_each to only add the requested frames to the rb_backtrace_t. To do this, backtrace_each needs to be passed the starting frame and number of frames values passed to Kernel#caller or #caller_locations. backtrace_each works from the top of the stack to the bottom, where the bottom is the current frame. Due to how the location for cfuncs is tracked using the location of the previous iseq, we need to store an extra frame for the previous iseq if we are limiting the backtrace and final backtrace frame (the first one stored) would be a cfunc and not an iseq. To limit the amount of work in this case, while scanning until the start of the requested backtrace, for each iseq, store the cfp. If the first backtrace frame we care about is a cfunc, use the stored cfp to find the related iseq. Use a function pointer to handle the storage of the cfp in the iteration arg, and also store the location of the extra frame in the iteration arg. backtrace_each needs to return int instead of void in order to signal when a starting frame larger than backtrace size is given, as caller and caller_locations needs to return nil and not the empty array in these cases. To handle cases where a range is provided with a negative end, and the backtrace size is needed to calculate the result to pass to rb_range_beg_len, add a backtrace_size static function to calculate the size, which copies the logic from backtrace_each. As backtrace_each only adds the backtrace lines requested, backtrace_to_*_ary can be simplified to always operate on the entire backtrace. Previously, caller_locations(1,1) was about 6.2 times slower for an 800 deep callstack compared to an empty callstack. With this new approach, it is only 1.3 times slower. It will always be somewhat slower as it still needs to scan the cfps from the top of the stack until it finds the first requested backtrace frame. This initializes the backtrace memory to zero. I do not think this is necessary, as from my analysis, nothing during the setting of the backtrace entries can cause a garbage collection, but it seems the safest approach, and it's unlikely the performance decrease is significant. This removes the rb_backtrace_t backtrace_base member. backtrace and backtrace_base were initialized to the same value, and neither is modified, so it doesn't make sense to have two pointers. This also removes LOCATION_TYPE_IFUNC from vm_backtrace.c, as the value is never set. Fixes [Bug #17031]
2020-08-28 01:17:36 +03:00
}
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
if (cfunc_counter > 0) {
for (; cfp != end_cfp; cfp = RUBY_VM_PREVIOUS_CONTROL_FRAME(cfp)) {
if (cfp->iseq && cfp->pc && (!skip_internal || !is_internal_location(cfp))) {
bt_update_cfunc_loc(cfunc_counter, loc, cfp->iseq, cfp->pc);
RB_OBJ_WRITTEN(btobj, Qundef, cfp->iseq);
if (do_yield) {
bt_yield_loc(loc - cfunc_counter, cfunc_counter, btobj);
}
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
break;
}
}
}
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
if (start_too_large) *start_too_large = (start_frame > 0 ? -1 : 0);
return btobj;
}
Improve performance of partial backtraces Previously, backtrace_each fully populated the rb_backtrace_t with all backtrace frames, even if caller only requested a partial backtrace (e.g. Kernel#caller_locations(1, 1)). This changes backtrace_each to only add the requested frames to the rb_backtrace_t. To do this, backtrace_each needs to be passed the starting frame and number of frames values passed to Kernel#caller or #caller_locations. backtrace_each works from the top of the stack to the bottom, where the bottom is the current frame. Due to how the location for cfuncs is tracked using the location of the previous iseq, we need to store an extra frame for the previous iseq if we are limiting the backtrace and final backtrace frame (the first one stored) would be a cfunc and not an iseq. To limit the amount of work in this case, while scanning until the start of the requested backtrace, for each iseq, store the cfp. If the first backtrace frame we care about is a cfunc, use the stored cfp to find the related iseq. Use a function pointer to handle the storage of the cfp in the iteration arg, and also store the location of the extra frame in the iteration arg. backtrace_each needs to return int instead of void in order to signal when a starting frame larger than backtrace size is given, as caller and caller_locations needs to return nil and not the empty array in these cases. To handle cases where a range is provided with a negative end, and the backtrace size is needed to calculate the result to pass to rb_range_beg_len, add a backtrace_size static function to calculate the size, which copies the logic from backtrace_each. As backtrace_each only adds the backtrace lines requested, backtrace_to_*_ary can be simplified to always operate on the entire backtrace. Previously, caller_locations(1,1) was about 6.2 times slower for an 800 deep callstack compared to an empty callstack. With this new approach, it is only 1.3 times slower. It will always be somewhat slower as it still needs to scan the cfps from the top of the stack until it finds the first requested backtrace frame. This initializes the backtrace memory to zero. I do not think this is necessary, as from my analysis, nothing during the setting of the backtrace entries can cause a garbage collection, but it seems the safest approach, and it's unlikely the performance decrease is significant. This removes the rb_backtrace_t backtrace_base member. backtrace and backtrace_base were initialized to the same value, and neither is modified, so it doesn't make sense to have two pointers. This also removes LOCATION_TYPE_IFUNC from vm_backtrace.c, as the value is never set. Fixes [Bug #17031]
2020-08-28 01:17:36 +03:00
MJIT_FUNC_EXPORTED VALUE
rb_ec_backtrace_object(const rb_execution_context_t *ec)
{
return rb_ec_partial_backtrace_object(ec, BACKTRACE_START, ALL_BACKTRACE_LINES, NULL, FALSE, FALSE);
Improve performance of partial backtraces Previously, backtrace_each fully populated the rb_backtrace_t with all backtrace frames, even if caller only requested a partial backtrace (e.g. Kernel#caller_locations(1, 1)). This changes backtrace_each to only add the requested frames to the rb_backtrace_t. To do this, backtrace_each needs to be passed the starting frame and number of frames values passed to Kernel#caller or #caller_locations. backtrace_each works from the top of the stack to the bottom, where the bottom is the current frame. Due to how the location for cfuncs is tracked using the location of the previous iseq, we need to store an extra frame for the previous iseq if we are limiting the backtrace and final backtrace frame (the first one stored) would be a cfunc and not an iseq. To limit the amount of work in this case, while scanning until the start of the requested backtrace, for each iseq, store the cfp. If the first backtrace frame we care about is a cfunc, use the stored cfp to find the related iseq. Use a function pointer to handle the storage of the cfp in the iteration arg, and also store the location of the extra frame in the iteration arg. backtrace_each needs to return int instead of void in order to signal when a starting frame larger than backtrace size is given, as caller and caller_locations needs to return nil and not the empty array in these cases. To handle cases where a range is provided with a negative end, and the backtrace size is needed to calculate the result to pass to rb_range_beg_len, add a backtrace_size static function to calculate the size, which copies the logic from backtrace_each. As backtrace_each only adds the backtrace lines requested, backtrace_to_*_ary can be simplified to always operate on the entire backtrace. Previously, caller_locations(1,1) was about 6.2 times slower for an 800 deep callstack compared to an empty callstack. With this new approach, it is only 1.3 times slower. It will always be somewhat slower as it still needs to scan the cfps from the top of the stack until it finds the first requested backtrace frame. This initializes the backtrace memory to zero. I do not think this is necessary, as from my analysis, nothing during the setting of the backtrace entries can cause a garbage collection, but it seems the safest approach, and it's unlikely the performance decrease is significant. This removes the rb_backtrace_t backtrace_base member. backtrace and backtrace_base were initialized to the same value, and neither is modified, so it doesn't make sense to have two pointers. This also removes LOCATION_TYPE_IFUNC from vm_backtrace.c, as the value is never set. Fixes [Bug #17031]
2020-08-28 01:17:36 +03:00
}
static VALUE
Improve performance of partial backtraces Previously, backtrace_each fully populated the rb_backtrace_t with all backtrace frames, even if caller only requested a partial backtrace (e.g. Kernel#caller_locations(1, 1)). This changes backtrace_each to only add the requested frames to the rb_backtrace_t. To do this, backtrace_each needs to be passed the starting frame and number of frames values passed to Kernel#caller or #caller_locations. backtrace_each works from the top of the stack to the bottom, where the bottom is the current frame. Due to how the location for cfuncs is tracked using the location of the previous iseq, we need to store an extra frame for the previous iseq if we are limiting the backtrace and final backtrace frame (the first one stored) would be a cfunc and not an iseq. To limit the amount of work in this case, while scanning until the start of the requested backtrace, for each iseq, store the cfp. If the first backtrace frame we care about is a cfunc, use the stored cfp to find the related iseq. Use a function pointer to handle the storage of the cfp in the iteration arg, and also store the location of the extra frame in the iteration arg. backtrace_each needs to return int instead of void in order to signal when a starting frame larger than backtrace size is given, as caller and caller_locations needs to return nil and not the empty array in these cases. To handle cases where a range is provided with a negative end, and the backtrace size is needed to calculate the result to pass to rb_range_beg_len, add a backtrace_size static function to calculate the size, which copies the logic from backtrace_each. As backtrace_each only adds the backtrace lines requested, backtrace_to_*_ary can be simplified to always operate on the entire backtrace. Previously, caller_locations(1,1) was about 6.2 times slower for an 800 deep callstack compared to an empty callstack. With this new approach, it is only 1.3 times slower. It will always be somewhat slower as it still needs to scan the cfps from the top of the stack until it finds the first requested backtrace frame. This initializes the backtrace memory to zero. I do not think this is necessary, as from my analysis, nothing during the setting of the backtrace entries can cause a garbage collection, but it seems the safest approach, and it's unlikely the performance decrease is significant. This removes the rb_backtrace_t backtrace_base member. backtrace and backtrace_base were initialized to the same value, and neither is modified, so it doesn't make sense to have two pointers. This also removes LOCATION_TYPE_IFUNC from vm_backtrace.c, as the value is never set. Fixes [Bug #17031]
2020-08-28 01:17:36 +03:00
backtrace_collect(rb_backtrace_t *bt, VALUE (*func)(rb_backtrace_location_t *, void *arg), void *arg)
{
VALUE btary;
int i;
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
btary = rb_ary_new2(bt->backtrace_size);
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
for (i=0; i<bt->backtrace_size; i++) {
rb_backtrace_location_t *loc = &bt->backtrace[i];
rb_ary_push(btary, func(loc, arg));
}
return btary;
}
static VALUE
location_to_str_dmyarg(rb_backtrace_location_t *loc, void *dmy)
{
return location_to_str(loc);
}
static VALUE
Improve performance of partial backtraces Previously, backtrace_each fully populated the rb_backtrace_t with all backtrace frames, even if caller only requested a partial backtrace (e.g. Kernel#caller_locations(1, 1)). This changes backtrace_each to only add the requested frames to the rb_backtrace_t. To do this, backtrace_each needs to be passed the starting frame and number of frames values passed to Kernel#caller or #caller_locations. backtrace_each works from the top of the stack to the bottom, where the bottom is the current frame. Due to how the location for cfuncs is tracked using the location of the previous iseq, we need to store an extra frame for the previous iseq if we are limiting the backtrace and final backtrace frame (the first one stored) would be a cfunc and not an iseq. To limit the amount of work in this case, while scanning until the start of the requested backtrace, for each iseq, store the cfp. If the first backtrace frame we care about is a cfunc, use the stored cfp to find the related iseq. Use a function pointer to handle the storage of the cfp in the iteration arg, and also store the location of the extra frame in the iteration arg. backtrace_each needs to return int instead of void in order to signal when a starting frame larger than backtrace size is given, as caller and caller_locations needs to return nil and not the empty array in these cases. To handle cases where a range is provided with a negative end, and the backtrace size is needed to calculate the result to pass to rb_range_beg_len, add a backtrace_size static function to calculate the size, which copies the logic from backtrace_each. As backtrace_each only adds the backtrace lines requested, backtrace_to_*_ary can be simplified to always operate on the entire backtrace. Previously, caller_locations(1,1) was about 6.2 times slower for an 800 deep callstack compared to an empty callstack. With this new approach, it is only 1.3 times slower. It will always be somewhat slower as it still needs to scan the cfps from the top of the stack until it finds the first requested backtrace frame. This initializes the backtrace memory to zero. I do not think this is necessary, as from my analysis, nothing during the setting of the backtrace entries can cause a garbage collection, but it seems the safest approach, and it's unlikely the performance decrease is significant. This removes the rb_backtrace_t backtrace_base member. backtrace and backtrace_base were initialized to the same value, and neither is modified, so it doesn't make sense to have two pointers. This also removes LOCATION_TYPE_IFUNC from vm_backtrace.c, as the value is never set. Fixes [Bug #17031]
2020-08-28 01:17:36 +03:00
backtrace_to_str_ary(VALUE self)
{
VALUE r;
Improve performance of partial backtraces Previously, backtrace_each fully populated the rb_backtrace_t with all backtrace frames, even if caller only requested a partial backtrace (e.g. Kernel#caller_locations(1, 1)). This changes backtrace_each to only add the requested frames to the rb_backtrace_t. To do this, backtrace_each needs to be passed the starting frame and number of frames values passed to Kernel#caller or #caller_locations. backtrace_each works from the top of the stack to the bottom, where the bottom is the current frame. Due to how the location for cfuncs is tracked using the location of the previous iseq, we need to store an extra frame for the previous iseq if we are limiting the backtrace and final backtrace frame (the first one stored) would be a cfunc and not an iseq. To limit the amount of work in this case, while scanning until the start of the requested backtrace, for each iseq, store the cfp. If the first backtrace frame we care about is a cfunc, use the stored cfp to find the related iseq. Use a function pointer to handle the storage of the cfp in the iteration arg, and also store the location of the extra frame in the iteration arg. backtrace_each needs to return int instead of void in order to signal when a starting frame larger than backtrace size is given, as caller and caller_locations needs to return nil and not the empty array in these cases. To handle cases where a range is provided with a negative end, and the backtrace size is needed to calculate the result to pass to rb_range_beg_len, add a backtrace_size static function to calculate the size, which copies the logic from backtrace_each. As backtrace_each only adds the backtrace lines requested, backtrace_to_*_ary can be simplified to always operate on the entire backtrace. Previously, caller_locations(1,1) was about 6.2 times slower for an 800 deep callstack compared to an empty callstack. With this new approach, it is only 1.3 times slower. It will always be somewhat slower as it still needs to scan the cfps from the top of the stack until it finds the first requested backtrace frame. This initializes the backtrace memory to zero. I do not think this is necessary, as from my analysis, nothing during the setting of the backtrace entries can cause a garbage collection, but it seems the safest approach, and it's unlikely the performance decrease is significant. This removes the rb_backtrace_t backtrace_base member. backtrace and backtrace_base were initialized to the same value, and neither is modified, so it doesn't make sense to have two pointers. This also removes LOCATION_TYPE_IFUNC from vm_backtrace.c, as the value is never set. Fixes [Bug #17031]
2020-08-28 01:17:36 +03:00
rb_backtrace_t *bt;
GetCoreDataFromValue(self, rb_backtrace_t, bt);
Improve performance of partial backtraces Previously, backtrace_each fully populated the rb_backtrace_t with all backtrace frames, even if caller only requested a partial backtrace (e.g. Kernel#caller_locations(1, 1)). This changes backtrace_each to only add the requested frames to the rb_backtrace_t. To do this, backtrace_each needs to be passed the starting frame and number of frames values passed to Kernel#caller or #caller_locations. backtrace_each works from the top of the stack to the bottom, where the bottom is the current frame. Due to how the location for cfuncs is tracked using the location of the previous iseq, we need to store an extra frame for the previous iseq if we are limiting the backtrace and final backtrace frame (the first one stored) would be a cfunc and not an iseq. To limit the amount of work in this case, while scanning until the start of the requested backtrace, for each iseq, store the cfp. If the first backtrace frame we care about is a cfunc, use the stored cfp to find the related iseq. Use a function pointer to handle the storage of the cfp in the iteration arg, and also store the location of the extra frame in the iteration arg. backtrace_each needs to return int instead of void in order to signal when a starting frame larger than backtrace size is given, as caller and caller_locations needs to return nil and not the empty array in these cases. To handle cases where a range is provided with a negative end, and the backtrace size is needed to calculate the result to pass to rb_range_beg_len, add a backtrace_size static function to calculate the size, which copies the logic from backtrace_each. As backtrace_each only adds the backtrace lines requested, backtrace_to_*_ary can be simplified to always operate on the entire backtrace. Previously, caller_locations(1,1) was about 6.2 times slower for an 800 deep callstack compared to an empty callstack. With this new approach, it is only 1.3 times slower. It will always be somewhat slower as it still needs to scan the cfps from the top of the stack until it finds the first requested backtrace frame. This initializes the backtrace memory to zero. I do not think this is necessary, as from my analysis, nothing during the setting of the backtrace entries can cause a garbage collection, but it seems the safest approach, and it's unlikely the performance decrease is significant. This removes the rb_backtrace_t backtrace_base member. backtrace and backtrace_base were initialized to the same value, and neither is modified, so it doesn't make sense to have two pointers. This also removes LOCATION_TYPE_IFUNC from vm_backtrace.c, as the value is never set. Fixes [Bug #17031]
2020-08-28 01:17:36 +03:00
r = backtrace_collect(bt, location_to_str_dmyarg, 0);
RB_GC_GUARD(self);
return r;
}
VALUE
rb_backtrace_to_str_ary(VALUE self)
{
rb_backtrace_t *bt;
GetCoreDataFromValue(self, rb_backtrace_t, bt);
if (!bt->strary) {
RB_OBJ_WRITE(self, &bt->strary, backtrace_to_str_ary(self));
}
return bt->strary;
}
mjit_compile.c: merge initial JIT compiler which has been developed by Takashi Kokubun <takashikkbn@gmail> as YARV-MJIT. Many of its bugs are fixed by wanabe <s.wanabe@gmail.com>. This JIT compiler is designed to be a safe migration path to introduce JIT compiler to MRI. So this commit does not include any bytecode changes or dynamic instruction modifications, which are done in original MJIT. This commit even strips off some aggressive optimizations from YARV-MJIT, and thus it's slower than YARV-MJIT too. But it's still fairly faster than Ruby 2.5 in some benchmarks (attached below). Note that this JIT compiler passes `make test`, `make test-all`, `make test-spec` without JIT, and even with JIT. Not only it's perfectly safe with JIT disabled because it does not replace VM instructions unlike MJIT, but also with JIT enabled it stably runs Ruby applications including Rails applications. I'm expecting this version as just "initial" JIT compiler. I have many optimization ideas which are skipped for initial merging, and you may easily replace this JIT compiler with a faster one by just replacing mjit_compile.c. `mjit_compile` interface is designed for the purpose. common.mk: update dependencies for mjit_compile.c. internal.h: declare `rb_vm_insn_addr2insn` for MJIT. vm.c: exclude some definitions if `-DMJIT_HEADER` is provided to compiler. This avoids to include some functions which take a long time to compile, e.g. vm_exec_core. Some of the purpose is achieved in transform_mjit_header.rb (see `IGNORED_FUNCTIONS`) but others are manually resolved for now. Load mjit_helper.h for MJIT header. mjit_helper.h: New. This is a file used only by JIT-ed code. I'll refactor `mjit_call_cfunc` later. vm_eval.c: add some #ifdef switches to skip compiling some functions like Init_vm_eval. win32/mkexports.rb: export thread/ec functions, which are used by MJIT. include/ruby/defines.h: add MJIT_FUNC_EXPORTED macro alis to clarify that a function is exported only for MJIT. array.c: export a function used by MJIT. bignum.c: ditto. class.c: ditto. compile.c: ditto. error.c: ditto. gc.c: ditto. hash.c: ditto. iseq.c: ditto. numeric.c: ditto. object.c: ditto. proc.c: ditto. re.c: ditto. st.c: ditto. string.c: ditto. thread.c: ditto. variable.c: ditto. vm_backtrace.c: ditto. vm_insnhelper.c: ditto. vm_method.c: ditto. I would like to improve maintainability of function exports, but I believe this way is acceptable as initial merging if we clarify the new exports are for MJIT (so that we can use them as TODO list to fix) and add unit tests to detect unresolved symbols. I'll add unit tests of JIT compilations in succeeding commits. Author: Takashi Kokubun <takashikkbn@gmail.com> Contributor: wanabe <s.wanabe@gmail.com> Part of [Feature #14235] --- * Known issues * Code generated by gcc is faster than clang. The benchmark may be worse in macOS. Following benchmark result is provided by gcc w/ Linux. * Performance is decreased when Google Chrome is running * JIT can work on MinGW, but it doesn't improve performance at least in short running benchmark. * Currently it doesn't perform well with Rails. We'll try to fix this before release. --- * Benchmark reslts Benchmarked with: Intel 4.0GHz i7-4790K with 16GB memory under x86-64 Ubuntu 8 Cores - 2.0.0-p0: Ruby 2.0.0-p0 - r62186: Ruby trunk (early 2.6.0), before MJIT changes - JIT off: On this commit, but without `--jit` option - JIT on: On this commit, and with `--jit` option ** Optcarrot fps Benchmark: https://github.com/mame/optcarrot | |2.0.0-p0 |r62186 |JIT off |JIT on | |:--------|:--------|:--------|:--------|:--------| |fps |37.32 |51.46 |51.31 |58.88 | |vs 2.0.0 |1.00x |1.38x |1.37x |1.58x | ** MJIT benchmarks Benchmark: https://github.com/benchmark-driver/mjit-benchmarks (Original: https://github.com/vnmakarov/ruby/tree/rtl_mjit_branch/MJIT-benchmarks) | |2.0.0-p0 |r62186 |JIT off |JIT on | |:----------|:--------|:--------|:--------|:--------| |aread |1.00 |1.09 |1.07 |2.19 | |aref |1.00 |1.13 |1.11 |2.22 | |aset |1.00 |1.50 |1.45 |2.64 | |awrite |1.00 |1.17 |1.13 |2.20 | |call |1.00 |1.29 |1.26 |2.02 | |const2 |1.00 |1.10 |1.10 |2.19 | |const |1.00 |1.11 |1.10 |2.19 | |fannk |1.00 |1.04 |1.02 |1.00 | |fib |1.00 |1.32 |1.31 |1.84 | |ivread |1.00 |1.13 |1.12 |2.43 | |ivwrite |1.00 |1.23 |1.21 |2.40 | |mandelbrot |1.00 |1.13 |1.16 |1.28 | |meteor |1.00 |2.97 |2.92 |3.17 | |nbody |1.00 |1.17 |1.15 |1.49 | |nest-ntimes|1.00 |1.22 |1.20 |1.39 | |nest-while |1.00 |1.10 |1.10 |1.37 | |norm |1.00 |1.18 |1.16 |1.24 | |nsvb |1.00 |1.16 |1.16 |1.17 | |red-black |1.00 |1.02 |0.99 |1.12 | |sieve |1.00 |1.30 |1.28 |1.62 | |trees |1.00 |1.14 |1.13 |1.19 | |while |1.00 |1.12 |1.11 |2.41 | ** Discourse's script/bench.rb Benchmark: https://github.com/discourse/discourse/blob/v1.8.7/script/bench.rb NOTE: Rails performance was somehow a little degraded with JIT for now. We should fix this. (At least I know opt_aref is performing badly in JIT and I have an idea to fix it. Please wait for the fix.) *** JIT off Your Results: (note for timings- percentile is first, duration is second in millisecs) categories_admin: 50: 17 75: 18 90: 22 99: 29 home_admin: 50: 21 75: 21 90: 27 99: 40 topic_admin: 50: 17 75: 18 90: 22 99: 32 categories: 50: 35 75: 41 90: 43 99: 77 home: 50: 39 75: 46 90: 49 99: 95 topic: 50: 46 75: 52 90: 56 99: 101 *** JIT on Your Results: (note for timings- percentile is first, duration is second in millisecs) categories_admin: 50: 19 75: 21 90: 25 99: 33 home_admin: 50: 24 75: 26 90: 30 99: 35 topic_admin: 50: 19 75: 20 90: 25 99: 30 categories: 50: 40 75: 44 90: 48 99: 76 home: 50: 42 75: 48 90: 51 99: 89 topic: 50: 49 75: 55 90: 58 99: 99 git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@62197 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2018-02-04 14:22:28 +03:00
MJIT_FUNC_EXPORTED void
rb_backtrace_use_iseq_first_lineno_for_last_location(VALUE self)
{
const rb_backtrace_t *bt;
rb_backtrace_location_t *loc;
GetCoreDataFromValue(self, rb_backtrace_t, bt);
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
VM_ASSERT(bt->backtrace_size > 0);
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
loc = &bt->backtrace[0];
VM_ASSERT(loc->type == LOCATION_TYPE_ISEQ);
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
loc->pc = NULL; // means location.first_lineno
}
static VALUE
location_create(rb_backtrace_location_t *srcloc, void *btobj)
{
VALUE obj;
struct valued_frame_info *vloc;
obj = TypedData_Make_Struct(rb_cBacktraceLocation, struct valued_frame_info, &location_data_type, vloc);
vloc->loc = srcloc;
RB_OBJ_WRITE(obj, &vloc->btobj, (VALUE)btobj);
return obj;
}
static VALUE
Improve performance of partial backtraces Previously, backtrace_each fully populated the rb_backtrace_t with all backtrace frames, even if caller only requested a partial backtrace (e.g. Kernel#caller_locations(1, 1)). This changes backtrace_each to only add the requested frames to the rb_backtrace_t. To do this, backtrace_each needs to be passed the starting frame and number of frames values passed to Kernel#caller or #caller_locations. backtrace_each works from the top of the stack to the bottom, where the bottom is the current frame. Due to how the location for cfuncs is tracked using the location of the previous iseq, we need to store an extra frame for the previous iseq if we are limiting the backtrace and final backtrace frame (the first one stored) would be a cfunc and not an iseq. To limit the amount of work in this case, while scanning until the start of the requested backtrace, for each iseq, store the cfp. If the first backtrace frame we care about is a cfunc, use the stored cfp to find the related iseq. Use a function pointer to handle the storage of the cfp in the iteration arg, and also store the location of the extra frame in the iteration arg. backtrace_each needs to return int instead of void in order to signal when a starting frame larger than backtrace size is given, as caller and caller_locations needs to return nil and not the empty array in these cases. To handle cases where a range is provided with a negative end, and the backtrace size is needed to calculate the result to pass to rb_range_beg_len, add a backtrace_size static function to calculate the size, which copies the logic from backtrace_each. As backtrace_each only adds the backtrace lines requested, backtrace_to_*_ary can be simplified to always operate on the entire backtrace. Previously, caller_locations(1,1) was about 6.2 times slower for an 800 deep callstack compared to an empty callstack. With this new approach, it is only 1.3 times slower. It will always be somewhat slower as it still needs to scan the cfps from the top of the stack until it finds the first requested backtrace frame. This initializes the backtrace memory to zero. I do not think this is necessary, as from my analysis, nothing during the setting of the backtrace entries can cause a garbage collection, but it seems the safest approach, and it's unlikely the performance decrease is significant. This removes the rb_backtrace_t backtrace_base member. backtrace and backtrace_base were initialized to the same value, and neither is modified, so it doesn't make sense to have two pointers. This also removes LOCATION_TYPE_IFUNC from vm_backtrace.c, as the value is never set. Fixes [Bug #17031]
2020-08-28 01:17:36 +03:00
backtrace_to_location_ary(VALUE self)
{
VALUE r;
Improve performance of partial backtraces Previously, backtrace_each fully populated the rb_backtrace_t with all backtrace frames, even if caller only requested a partial backtrace (e.g. Kernel#caller_locations(1, 1)). This changes backtrace_each to only add the requested frames to the rb_backtrace_t. To do this, backtrace_each needs to be passed the starting frame and number of frames values passed to Kernel#caller or #caller_locations. backtrace_each works from the top of the stack to the bottom, where the bottom is the current frame. Due to how the location for cfuncs is tracked using the location of the previous iseq, we need to store an extra frame for the previous iseq if we are limiting the backtrace and final backtrace frame (the first one stored) would be a cfunc and not an iseq. To limit the amount of work in this case, while scanning until the start of the requested backtrace, for each iseq, store the cfp. If the first backtrace frame we care about is a cfunc, use the stored cfp to find the related iseq. Use a function pointer to handle the storage of the cfp in the iteration arg, and also store the location of the extra frame in the iteration arg. backtrace_each needs to return int instead of void in order to signal when a starting frame larger than backtrace size is given, as caller and caller_locations needs to return nil and not the empty array in these cases. To handle cases where a range is provided with a negative end, and the backtrace size is needed to calculate the result to pass to rb_range_beg_len, add a backtrace_size static function to calculate the size, which copies the logic from backtrace_each. As backtrace_each only adds the backtrace lines requested, backtrace_to_*_ary can be simplified to always operate on the entire backtrace. Previously, caller_locations(1,1) was about 6.2 times slower for an 800 deep callstack compared to an empty callstack. With this new approach, it is only 1.3 times slower. It will always be somewhat slower as it still needs to scan the cfps from the top of the stack until it finds the first requested backtrace frame. This initializes the backtrace memory to zero. I do not think this is necessary, as from my analysis, nothing during the setting of the backtrace entries can cause a garbage collection, but it seems the safest approach, and it's unlikely the performance decrease is significant. This removes the rb_backtrace_t backtrace_base member. backtrace and backtrace_base were initialized to the same value, and neither is modified, so it doesn't make sense to have two pointers. This also removes LOCATION_TYPE_IFUNC from vm_backtrace.c, as the value is never set. Fixes [Bug #17031]
2020-08-28 01:17:36 +03:00
rb_backtrace_t *bt;
GetCoreDataFromValue(self, rb_backtrace_t, bt);
Improve performance of partial backtraces Previously, backtrace_each fully populated the rb_backtrace_t with all backtrace frames, even if caller only requested a partial backtrace (e.g. Kernel#caller_locations(1, 1)). This changes backtrace_each to only add the requested frames to the rb_backtrace_t. To do this, backtrace_each needs to be passed the starting frame and number of frames values passed to Kernel#caller or #caller_locations. backtrace_each works from the top of the stack to the bottom, where the bottom is the current frame. Due to how the location for cfuncs is tracked using the location of the previous iseq, we need to store an extra frame for the previous iseq if we are limiting the backtrace and final backtrace frame (the first one stored) would be a cfunc and not an iseq. To limit the amount of work in this case, while scanning until the start of the requested backtrace, for each iseq, store the cfp. If the first backtrace frame we care about is a cfunc, use the stored cfp to find the related iseq. Use a function pointer to handle the storage of the cfp in the iteration arg, and also store the location of the extra frame in the iteration arg. backtrace_each needs to return int instead of void in order to signal when a starting frame larger than backtrace size is given, as caller and caller_locations needs to return nil and not the empty array in these cases. To handle cases where a range is provided with a negative end, and the backtrace size is needed to calculate the result to pass to rb_range_beg_len, add a backtrace_size static function to calculate the size, which copies the logic from backtrace_each. As backtrace_each only adds the backtrace lines requested, backtrace_to_*_ary can be simplified to always operate on the entire backtrace. Previously, caller_locations(1,1) was about 6.2 times slower for an 800 deep callstack compared to an empty callstack. With this new approach, it is only 1.3 times slower. It will always be somewhat slower as it still needs to scan the cfps from the top of the stack until it finds the first requested backtrace frame. This initializes the backtrace memory to zero. I do not think this is necessary, as from my analysis, nothing during the setting of the backtrace entries can cause a garbage collection, but it seems the safest approach, and it's unlikely the performance decrease is significant. This removes the rb_backtrace_t backtrace_base member. backtrace and backtrace_base were initialized to the same value, and neither is modified, so it doesn't make sense to have two pointers. This also removes LOCATION_TYPE_IFUNC from vm_backtrace.c, as the value is never set. Fixes [Bug #17031]
2020-08-28 01:17:36 +03:00
r = backtrace_collect(bt, location_create, (void *)self);
RB_GC_GUARD(self);
return r;
}
VALUE
rb_backtrace_to_location_ary(VALUE self)
{
rb_backtrace_t *bt;
GetCoreDataFromValue(self, rb_backtrace_t, bt);
if (!bt->locary) {
RB_OBJ_WRITE(self, &bt->locary, backtrace_to_location_ary(self));
}
return bt->locary;
}
static VALUE
backtrace_dump_data(VALUE self)
{
VALUE str = rb_backtrace_to_str_ary(self);
return str;
}
static VALUE
backtrace_load_data(VALUE self, VALUE str)
{
rb_backtrace_t *bt;
GetCoreDataFromValue(self, rb_backtrace_t, bt);
RB_OBJ_WRITE(self, &bt->strary, str);
return self;
}
2021-12-20 10:07:25 +03:00
/*
2022-05-11 23:32:33 +03:00
* call-seq: Thread::Backtrace::limit -> integer
2021-12-20 10:07:25 +03:00
*
* Returns maximum backtrace length set by <tt>--backtrace-limit</tt>
* command-line option. The default is <tt>-1</tt> which means unlimited
2021-12-20 10:07:25 +03:00
* backtraces. If the value is zero or positive, the error backtraces,
* produced by Exception#full_message, are abbreviated and the extra lines
* are replaced by <tt>... 3 levels... </tt>
*
* $ ruby -r net/http -e "p Thread::Backtrace.limit; Net::HTTP.get(URI('http://wrong.address'))"
* - 1
* .../lib/ruby/3.1.0/socket.rb:227:in `getaddrinfo': Failed to open TCP connection to wrong.address:80 (getaddrinfo: Name or service not known) (SocketError)
* from .../lib/ruby/3.1.0/socket.rb:227:in `foreach'
* from .../lib/ruby/3.1.0/socket.rb:632:in `tcp'
* from .../lib/ruby/3.1.0/net/http.rb:998:in `connect'
* from .../lib/ruby/3.1.0/net/http.rb:976:in `do_start'
* from .../lib/ruby/3.1.0/net/http.rb:965:in `start'
* from .../lib/ruby/3.1.0/net/http.rb:627:in `start'
* from .../lib/ruby/3.1.0/net/http.rb:503:in `get_response'
* from .../lib/ruby/3.1.0/net/http.rb:474:in `get'
* .../lib/ruby/3.1.0/socket.rb:227:in `getaddrinfo': getaddrinfo: Name or service not known (SocketError)
* from .../lib/ruby/3.1.0/socket.rb:227:in `foreach'
* from .../lib/ruby/3.1.0/socket.rb:632:in `tcp'
* from .../lib/ruby/3.1.0/net/http.rb:998:in `connect'
* from .../lib/ruby/3.1.0/net/http.rb:976:in `do_start'
* from .../lib/ruby/3.1.0/net/http.rb:965:in `start'
* from .../lib/ruby/3.1.0/net/http.rb:627:in `start'
* from .../lib/ruby/3.1.0/net/http.rb:503:in `get_response'
* from .../lib/ruby/3.1.0/net/http.rb:474:in `get'
* from -e:1:in `<main>'
*
* $ ruby --backtrace-limit 2 -r net/http -e "p Thread::Backtrace.limit; Net::HTTP.get(URI('http://wrong.address'))"
* 2
* .../lib/ruby/3.1.0/socket.rb:227:in `getaddrinfo': Failed to open TCP connection to wrong.address:80 (getaddrinfo: Name or service not known) (SocketError)
* from .../lib/ruby/3.1.0/socket.rb:227:in `foreach'
* from .../lib/ruby/3.1.0/socket.rb:632:in `tcp'
* ... 7 levels...
* .../lib/ruby/3.1.0/socket.rb:227:in `getaddrinfo': getaddrinfo: Name or service not known (SocketError)
* from .../lib/ruby/3.1.0/socket.rb:227:in `foreach'
* from .../lib/ruby/3.1.0/socket.rb:632:in `tcp'
* ... 7 levels...
*
* $ ruby --backtrace-limit 0 -r net/http -e "p Thread::Backtrace.limit; Net::HTTP.get(URI('http://wrong.address'))"
* 0
* .../lib/ruby/3.1.0/socket.rb:227:in `getaddrinfo': Failed to open TCP connection to wrong.address:80 (getaddrinfo: Name or service not known) (SocketError)
* ... 9 levels...
* .../lib/ruby/3.1.0/socket.rb:227:in `getaddrinfo': getaddrinfo: Name or service not known (SocketError)
* ... 9 levels...
*
*/
static VALUE
backtrace_limit(VALUE self)
{
return LONG2NUM(rb_backtrace_length_limit);
}
VALUE
rb_ec_backtrace_str_ary(const rb_execution_context_t *ec, long lev, long n)
{
return rb_backtrace_to_str_ary(rb_ec_partial_backtrace_object(ec, lev, n, NULL, FALSE, FALSE));
}
VALUE
rb_ec_backtrace_location_ary(const rb_execution_context_t *ec, long lev, long n, bool skip_internal)
{
return rb_backtrace_to_location_ary(rb_ec_partial_backtrace_object(ec, lev, n, NULL, skip_internal, FALSE));
}
/* make old style backtrace directly */
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
static void
backtrace_each(const rb_execution_context_t *ec,
void (*init)(void *arg, size_t size),
void (*iter_iseq)(void *arg, const rb_control_frame_t *cfp),
void (*iter_cfunc)(void *arg, const rb_control_frame_t *cfp, ID mid),
void *arg)
{
const rb_control_frame_t *last_cfp = ec->cfp;
const rb_control_frame_t *start_cfp = RUBY_VM_END_CONTROL_FRAME(ec);
const rb_control_frame_t *cfp;
ptrdiff_t size, i;
// In the case the thread vm_stack or cfp is not initialized, there is no backtrace.
if (start_cfp == NULL) {
init(arg, 0);
return;
}
/* <- start_cfp (end control frame)
* top frame (dummy)
* top frame (dummy)
* top frame <- start_cfp
* top frame
* ...
* 2nd frame <- lev:0
* current frame <- ec->cfp
*/
start_cfp =
RUBY_VM_NEXT_CONTROL_FRAME(
RUBY_VM_NEXT_CONTROL_FRAME(start_cfp)); /* skip top frames */
if (start_cfp < last_cfp) {
size = 0;
}
else {
size = start_cfp - last_cfp + 1;
}
init(arg, size);
/* SDR(); */
for (i=0, cfp = start_cfp; i<size; i++, cfp = RUBY_VM_NEXT_CONTROL_FRAME(cfp)) {
/* fprintf(stderr, "cfp: %d\n", (rb_control_frame_t *)(ec->vm_stack + ec->vm_stack_size) - cfp); */
if (cfp->iseq) {
if (cfp->pc) {
iter_iseq(arg, cfp);
}
}
else {
VM_ASSERT(RUBYVM_CFUNC_FRAME_P(cfp));
Make backtrace generation work outward from current frame This fixes multiple bugs found in the partial backtrace optimization added in 3b24b7914c16930bfadc89d6aff6326a51c54295. These bugs occurs when passing a start argument to caller where the start argument lands on a iseq frame without a pc. Before this commit, the following code results in the same line being printed twice, both for the #each method. ```ruby def a; [1].group_by { b } end def b; puts(caller(2, 1).first, caller(3, 1).first) end a ``` After this commit and in Ruby 2.7, the lines are different, with the first line being for each and the second for group_by. Before this commit, the following code can either segfault or result in an infinite loop: ```ruby def foo caller_locations(2, 1).inspect # segfault caller_locations(2, 1)[0].path # infinite loop end 1.times.map { 1.times.map { foo } } ``` After this commit, this code works correctly. This commit completely refactors the backtrace handling. Instead of processing the backtrace from the outermost frame working in, process it from the innermost frame working out. This is much faster for partial backtraces, since you only access the control frames you need to in order to construct the backtrace. To handle cfunc frames in the new design, they start out with no location information. We increment a counter for each cfunc frame added. When an iseq frame with pc is accessed, after adding the iseq backtrace location, we use the location for the iseq backtrace location for all of the directly preceding cfunc backtrace locations. If the last backtrace line is a cfunc frame, we continue scanning for iseq frames until the end control frame, and use the location information from the first one for the trailing cfunc frames in the backtrace. As only rb_ec_partial_backtrace_object uses the new backtrace implementation, remove all of the function pointers and inline the functions. This makes the process easier to understand. Restore the Ruby 2.7 implementation of backtrace_each and use it for all the other functions that called backtrace_each other than rb_ec_partial_backtrace_object. All other cases requested the entire backtrace, so there is no advantage of using the new algorithm for those. Additionally, there are implicit assumptions in the other code that the backtrace processing works inward instead of outward. Remove the cfunc/iseq union in rb_backtrace_location_t, and remove the prev_loc member for cfunc. Both cfunc and iseq types can now have iseq and pc entries, so the location information can be accessed the same way for each. This avoids the need for a extra backtrace location entry to store an iseq backtrace location if the final entry in the backtrace is a cfunc. This is also what fixes the segfault and infinite loop issues in the above bugs. Here's Ruby pseudocode for the new algorithm, where start and length are the arguments to caller or caller_locations: ```ruby end_cf = VM.end_control_frame.next cf = VM.start_control_frame size = VM.num_control_frames - 2 bt = [] cfunc_counter = 0 if length.nil? || length > size length = size end while cf != end_cf && bt.size != length if cf.iseq? if cf.instruction_pointer? if start > 0 start -= 1 else bt << cf.iseq_backtrace_entry cf_counter.times do |i| bt[-1 - i].loc = cf.loc end cfunc_counter = 0 end end elsif cf.cfunc? if start > 0 start -= 1 else bt << cf.cfunc_backtrace_entry cfunc_counter += 1 end end cf = cf.prev end if cfunc_counter > 0 while cf != end_cf if (cf.iseq? && cf.instruction_pointer?) cf_counter.times do |i| bt[-i].loc = cf.loc end end cf = cf.prev end end ``` With the following benchmark, which uses a call depth of around 100 (common in many Ruby applications): ```ruby class T def test(depth, &block) if depth == 0 yield self else test(depth - 1, &block) end end def array Array.new end def first caller_locations(1, 1) end def full caller_locations end end t = T.new t.test((ARGV.first || 100).to_i) do Benchmark.ips do |x| x.report ('caller_loc(1, 1)') {t.first} x.report ('caller_loc') {t.full} x.report ('Array.new') {t.array} x.compare! end end ``` Results before commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 281.159k (_ 0.7%) i/s - 1.426M in 5.073055s caller_loc 15.836k (_ 2.1%) i/s - 79.450k in 5.019426s Array.new 1.852M (_ 2.5%) i/s - 9.296M in 5.022511s Comparison: Array.new: 1852297.5 i/s caller_loc(1, 1): 281159.1 i/s - 6.59x (_ 0.00) slower caller_loc: 15835.9 i/s - 116.97x (_ 0.00) slower ``` Results after commit: ``` Calculating ------------------------------------- caller_loc(1, 1) 562.286k (_ 0.8%) i/s - 2.858M in 5.083249s caller_loc 16.402k (_ 1.0%) i/s - 83.200k in 5.072963s Array.new 1.853M (_ 0.1%) i/s - 9.278M in 5.007523s Comparison: Array.new: 1852776.5 i/s caller_loc(1, 1): 562285.6 i/s - 3.30x (_ 0.00) slower caller_loc: 16402.3 i/s - 112.96x (_ 0.00) slower ``` This shows that the speed of caller_locations(1, 1) has roughly doubled, and the speed of caller_locations with no arguments has improved slightly. So this new algorithm is significant faster, much simpler, and fixes bugs in the previous algorithm. Fixes [Bug #18053]
2021-07-22 02:44:56 +03:00
const rb_callable_method_entry_t *me = rb_vm_frame_method_entry(cfp);
ID mid = me->def->original_id;
iter_cfunc(arg, cfp, mid);
}
}
}
struct oldbt_arg {
VALUE filename;
int lineno;
void (*func)(void *data, VALUE file, int lineno, VALUE name);
void *data; /* result */
};
static void
oldbt_init(void *ptr, size_t dmy)
{
struct oldbt_arg *arg = (struct oldbt_arg *)ptr;
arg->filename = GET_VM()->progname;
arg->lineno = 0;
}
static void
oldbt_iter_iseq(void *ptr, const rb_control_frame_t *cfp)
{
const rb_iseq_t *iseq = cfp->iseq;
const VALUE *pc = cfp->pc;
struct oldbt_arg *arg = (struct oldbt_arg *)ptr;
VALUE file = arg->filename = rb_iseq_path(iseq);
VALUE name = ISEQ_BODY(iseq)->location.label;
int lineno = arg->lineno = calc_lineno(iseq, pc);
(arg->func)(arg->data, file, lineno, name);
}
static void
oldbt_iter_cfunc(void *ptr, const rb_control_frame_t *cfp, ID mid)
{
struct oldbt_arg *arg = (struct oldbt_arg *)ptr;
VALUE file = arg->filename;
VALUE name = rb_id2str(mid);
int lineno = arg->lineno;
(arg->func)(arg->data, file, lineno, name);
}
static void
oldbt_print(void *data, VALUE file, int lineno, VALUE name)
{
FILE *fp = (FILE *)data;
if (NIL_P(name)) {
fprintf(fp, "\tfrom %s:%d:in unknown method\n",
RSTRING_PTR(file), lineno);
}
else {
fprintf(fp, "\tfrom %s:%d:in `%s'\n",
RSTRING_PTR(file), lineno, RSTRING_PTR(name));
}
}
static void
vm_backtrace_print(FILE *fp)
{
struct oldbt_arg arg;
arg.func = oldbt_print;
arg.data = (void *)fp;
backtrace_each(GET_EC(),
oldbt_init,
oldbt_iter_iseq,
oldbt_iter_cfunc,
&arg);
}
static void
oldbt_bugreport(void *arg, VALUE file, int line, VALUE method)
{
const char *filename = NIL_P(file) ? "ruby" : RSTRING_PTR(file);
if (!*(int *)arg) {
fprintf(stderr, "-- Ruby level backtrace information "
"----------------------------------------\n");
*(int *)arg = 1;
}
if (NIL_P(method)) {
fprintf(stderr, "%s:%d:in unknown method\n", filename, line);
}
else {
fprintf(stderr, "%s:%d:in `%s'\n", filename, line, RSTRING_PTR(method));
}
}
void
rb_backtrace_print_as_bugreport(void)
{
struct oldbt_arg arg;
int i = 0;
arg.func = oldbt_bugreport;
arg.data = (int *)&i;
backtrace_each(GET_EC(),
oldbt_init,
oldbt_iter_iseq,
oldbt_iter_cfunc,
&arg);
}
void
rb_backtrace(void)
{
vm_backtrace_print(stderr);
}
struct print_to_arg {
VALUE (*iter)(VALUE recv, VALUE str);
VALUE output;
};
static void
oldbt_print_to(void *data, VALUE file, int lineno, VALUE name)
{
const struct print_to_arg *arg = data;
VALUE str = rb_sprintf("\tfrom %"PRIsVALUE":%d:in ", file, lineno);
if (NIL_P(name)) {
rb_str_cat2(str, "unknown method\n");
}
else {
rb_str_catf(str, " `%"PRIsVALUE"'\n", name);
}
(*arg->iter)(arg->output, str);
}
void
rb_backtrace_each(VALUE (*iter)(VALUE recv, VALUE str), VALUE output)
{
struct oldbt_arg arg;
struct print_to_arg parg;
parg.iter = iter;
parg.output = output;
arg.func = oldbt_print_to;
arg.data = &parg;
backtrace_each(GET_EC(),
oldbt_init,
oldbt_iter_iseq,
oldbt_iter_cfunc,
&arg);
}
VALUE
rb_make_backtrace(void)
{
Improve performance of partial backtraces Previously, backtrace_each fully populated the rb_backtrace_t with all backtrace frames, even if caller only requested a partial backtrace (e.g. Kernel#caller_locations(1, 1)). This changes backtrace_each to only add the requested frames to the rb_backtrace_t. To do this, backtrace_each needs to be passed the starting frame and number of frames values passed to Kernel#caller or #caller_locations. backtrace_each works from the top of the stack to the bottom, where the bottom is the current frame. Due to how the location for cfuncs is tracked using the location of the previous iseq, we need to store an extra frame for the previous iseq if we are limiting the backtrace and final backtrace frame (the first one stored) would be a cfunc and not an iseq. To limit the amount of work in this case, while scanning until the start of the requested backtrace, for each iseq, store the cfp. If the first backtrace frame we care about is a cfunc, use the stored cfp to find the related iseq. Use a function pointer to handle the storage of the cfp in the iteration arg, and also store the location of the extra frame in the iteration arg. backtrace_each needs to return int instead of void in order to signal when a starting frame larger than backtrace size is given, as caller and caller_locations needs to return nil and not the empty array in these cases. To handle cases where a range is provided with a negative end, and the backtrace size is needed to calculate the result to pass to rb_range_beg_len, add a backtrace_size static function to calculate the size, which copies the logic from backtrace_each. As backtrace_each only adds the backtrace lines requested, backtrace_to_*_ary can be simplified to always operate on the entire backtrace. Previously, caller_locations(1,1) was about 6.2 times slower for an 800 deep callstack compared to an empty callstack. With this new approach, it is only 1.3 times slower. It will always be somewhat slower as it still needs to scan the cfps from the top of the stack until it finds the first requested backtrace frame. This initializes the backtrace memory to zero. I do not think this is necessary, as from my analysis, nothing during the setting of the backtrace entries can cause a garbage collection, but it seems the safest approach, and it's unlikely the performance decrease is significant. This removes the rb_backtrace_t backtrace_base member. backtrace and backtrace_base were initialized to the same value, and neither is modified, so it doesn't make sense to have two pointers. This also removes LOCATION_TYPE_IFUNC from vm_backtrace.c, as the value is never set. Fixes [Bug #17031]
2020-08-28 01:17:36 +03:00
return rb_ec_backtrace_str_ary(GET_EC(), BACKTRACE_START, ALL_BACKTRACE_LINES);
}
static VALUE
ec_backtrace_to_ary(const rb_execution_context_t *ec, int argc, const VALUE *argv, int lev_default, int lev_plus, int to_str)
{
VALUE level, vn;
long lev, n;
Improve performance of partial backtraces Previously, backtrace_each fully populated the rb_backtrace_t with all backtrace frames, even if caller only requested a partial backtrace (e.g. Kernel#caller_locations(1, 1)). This changes backtrace_each to only add the requested frames to the rb_backtrace_t. To do this, backtrace_each needs to be passed the starting frame and number of frames values passed to Kernel#caller or #caller_locations. backtrace_each works from the top of the stack to the bottom, where the bottom is the current frame. Due to how the location for cfuncs is tracked using the location of the previous iseq, we need to store an extra frame for the previous iseq if we are limiting the backtrace and final backtrace frame (the first one stored) would be a cfunc and not an iseq. To limit the amount of work in this case, while scanning until the start of the requested backtrace, for each iseq, store the cfp. If the first backtrace frame we care about is a cfunc, use the stored cfp to find the related iseq. Use a function pointer to handle the storage of the cfp in the iteration arg, and also store the location of the extra frame in the iteration arg. backtrace_each needs to return int instead of void in order to signal when a starting frame larger than backtrace size is given, as caller and caller_locations needs to return nil and not the empty array in these cases. To handle cases where a range is provided with a negative end, and the backtrace size is needed to calculate the result to pass to rb_range_beg_len, add a backtrace_size static function to calculate the size, which copies the logic from backtrace_each. As backtrace_each only adds the backtrace lines requested, backtrace_to_*_ary can be simplified to always operate on the entire backtrace. Previously, caller_locations(1,1) was about 6.2 times slower for an 800 deep callstack compared to an empty callstack. With this new approach, it is only 1.3 times slower. It will always be somewhat slower as it still needs to scan the cfps from the top of the stack until it finds the first requested backtrace frame. This initializes the backtrace memory to zero. I do not think this is necessary, as from my analysis, nothing during the setting of the backtrace entries can cause a garbage collection, but it seems the safest approach, and it's unlikely the performance decrease is significant. This removes the rb_backtrace_t backtrace_base member. backtrace and backtrace_base were initialized to the same value, and neither is modified, so it doesn't make sense to have two pointers. This also removes LOCATION_TYPE_IFUNC from vm_backtrace.c, as the value is never set. Fixes [Bug #17031]
2020-08-28 01:17:36 +03:00
VALUE btval;
VALUE r;
Improve performance of partial backtraces Previously, backtrace_each fully populated the rb_backtrace_t with all backtrace frames, even if caller only requested a partial backtrace (e.g. Kernel#caller_locations(1, 1)). This changes backtrace_each to only add the requested frames to the rb_backtrace_t. To do this, backtrace_each needs to be passed the starting frame and number of frames values passed to Kernel#caller or #caller_locations. backtrace_each works from the top of the stack to the bottom, where the bottom is the current frame. Due to how the location for cfuncs is tracked using the location of the previous iseq, we need to store an extra frame for the previous iseq if we are limiting the backtrace and final backtrace frame (the first one stored) would be a cfunc and not an iseq. To limit the amount of work in this case, while scanning until the start of the requested backtrace, for each iseq, store the cfp. If the first backtrace frame we care about is a cfunc, use the stored cfp to find the related iseq. Use a function pointer to handle the storage of the cfp in the iteration arg, and also store the location of the extra frame in the iteration arg. backtrace_each needs to return int instead of void in order to signal when a starting frame larger than backtrace size is given, as caller and caller_locations needs to return nil and not the empty array in these cases. To handle cases where a range is provided with a negative end, and the backtrace size is needed to calculate the result to pass to rb_range_beg_len, add a backtrace_size static function to calculate the size, which copies the logic from backtrace_each. As backtrace_each only adds the backtrace lines requested, backtrace_to_*_ary can be simplified to always operate on the entire backtrace. Previously, caller_locations(1,1) was about 6.2 times slower for an 800 deep callstack compared to an empty callstack. With this new approach, it is only 1.3 times slower. It will always be somewhat slower as it still needs to scan the cfps from the top of the stack until it finds the first requested backtrace frame. This initializes the backtrace memory to zero. I do not think this is necessary, as from my analysis, nothing during the setting of the backtrace entries can cause a garbage collection, but it seems the safest approach, and it's unlikely the performance decrease is significant. This removes the rb_backtrace_t backtrace_base member. backtrace and backtrace_base were initialized to the same value, and neither is modified, so it doesn't make sense to have two pointers. This also removes LOCATION_TYPE_IFUNC from vm_backtrace.c, as the value is never set. Fixes [Bug #17031]
2020-08-28 01:17:36 +03:00
int too_large;
rb_scan_args(argc, argv, "02", &level, &vn);
if (argc == 2 && NIL_P(vn)) argc--;
switch (argc) {
case 0:
lev = lev_default + lev_plus;
Improve performance of partial backtraces Previously, backtrace_each fully populated the rb_backtrace_t with all backtrace frames, even if caller only requested a partial backtrace (e.g. Kernel#caller_locations(1, 1)). This changes backtrace_each to only add the requested frames to the rb_backtrace_t. To do this, backtrace_each needs to be passed the starting frame and number of frames values passed to Kernel#caller or #caller_locations. backtrace_each works from the top of the stack to the bottom, where the bottom is the current frame. Due to how the location for cfuncs is tracked using the location of the previous iseq, we need to store an extra frame for the previous iseq if we are limiting the backtrace and final backtrace frame (the first one stored) would be a cfunc and not an iseq. To limit the amount of work in this case, while scanning until the start of the requested backtrace, for each iseq, store the cfp. If the first backtrace frame we care about is a cfunc, use the stored cfp to find the related iseq. Use a function pointer to handle the storage of the cfp in the iteration arg, and also store the location of the extra frame in the iteration arg. backtrace_each needs to return int instead of void in order to signal when a starting frame larger than backtrace size is given, as caller and caller_locations needs to return nil and not the empty array in these cases. To handle cases where a range is provided with a negative end, and the backtrace size is needed to calculate the result to pass to rb_range_beg_len, add a backtrace_size static function to calculate the size, which copies the logic from backtrace_each. As backtrace_each only adds the backtrace lines requested, backtrace_to_*_ary can be simplified to always operate on the entire backtrace. Previously, caller_locations(1,1) was about 6.2 times slower for an 800 deep callstack compared to an empty callstack. With this new approach, it is only 1.3 times slower. It will always be somewhat slower as it still needs to scan the cfps from the top of the stack until it finds the first requested backtrace frame. This initializes the backtrace memory to zero. I do not think this is necessary, as from my analysis, nothing during the setting of the backtrace entries can cause a garbage collection, but it seems the safest approach, and it's unlikely the performance decrease is significant. This removes the rb_backtrace_t backtrace_base member. backtrace and backtrace_base were initialized to the same value, and neither is modified, so it doesn't make sense to have two pointers. This also removes LOCATION_TYPE_IFUNC from vm_backtrace.c, as the value is never set. Fixes [Bug #17031]
2020-08-28 01:17:36 +03:00
n = ALL_BACKTRACE_LINES;
break;
case 1:
{
Improve performance of partial backtraces Previously, backtrace_each fully populated the rb_backtrace_t with all backtrace frames, even if caller only requested a partial backtrace (e.g. Kernel#caller_locations(1, 1)). This changes backtrace_each to only add the requested frames to the rb_backtrace_t. To do this, backtrace_each needs to be passed the starting frame and number of frames values passed to Kernel#caller or #caller_locations. backtrace_each works from the top of the stack to the bottom, where the bottom is the current frame. Due to how the location for cfuncs is tracked using the location of the previous iseq, we need to store an extra frame for the previous iseq if we are limiting the backtrace and final backtrace frame (the first one stored) would be a cfunc and not an iseq. To limit the amount of work in this case, while scanning until the start of the requested backtrace, for each iseq, store the cfp. If the first backtrace frame we care about is a cfunc, use the stored cfp to find the related iseq. Use a function pointer to handle the storage of the cfp in the iteration arg, and also store the location of the extra frame in the iteration arg. backtrace_each needs to return int instead of void in order to signal when a starting frame larger than backtrace size is given, as caller and caller_locations needs to return nil and not the empty array in these cases. To handle cases where a range is provided with a negative end, and the backtrace size is needed to calculate the result to pass to rb_range_beg_len, add a backtrace_size static function to calculate the size, which copies the logic from backtrace_each. As backtrace_each only adds the backtrace lines requested, backtrace_to_*_ary can be simplified to always operate on the entire backtrace. Previously, caller_locations(1,1) was about 6.2 times slower for an 800 deep callstack compared to an empty callstack. With this new approach, it is only 1.3 times slower. It will always be somewhat slower as it still needs to scan the cfps from the top of the stack until it finds the first requested backtrace frame. This initializes the backtrace memory to zero. I do not think this is necessary, as from my analysis, nothing during the setting of the backtrace entries can cause a garbage collection, but it seems the safest approach, and it's unlikely the performance decrease is significant. This removes the rb_backtrace_t backtrace_base member. backtrace and backtrace_base were initialized to the same value, and neither is modified, so it doesn't make sense to have two pointers. This also removes LOCATION_TYPE_IFUNC from vm_backtrace.c, as the value is never set. Fixes [Bug #17031]
2020-08-28 01:17:36 +03:00
long beg, len, bt_size = backtrace_size(ec);
switch (rb_range_beg_len(level, &beg, &len, bt_size - lev_plus, 0)) {
case Qfalse:
lev = NUM2LONG(level);
if (lev < 0) {
rb_raise(rb_eArgError, "negative level (%ld)", lev);
}
lev += lev_plus;
Improve performance of partial backtraces Previously, backtrace_each fully populated the rb_backtrace_t with all backtrace frames, even if caller only requested a partial backtrace (e.g. Kernel#caller_locations(1, 1)). This changes backtrace_each to only add the requested frames to the rb_backtrace_t. To do this, backtrace_each needs to be passed the starting frame and number of frames values passed to Kernel#caller or #caller_locations. backtrace_each works from the top of the stack to the bottom, where the bottom is the current frame. Due to how the location for cfuncs is tracked using the location of the previous iseq, we need to store an extra frame for the previous iseq if we are limiting the backtrace and final backtrace frame (the first one stored) would be a cfunc and not an iseq. To limit the amount of work in this case, while scanning until the start of the requested backtrace, for each iseq, store the cfp. If the first backtrace frame we care about is a cfunc, use the stored cfp to find the related iseq. Use a function pointer to handle the storage of the cfp in the iteration arg, and also store the location of the extra frame in the iteration arg. backtrace_each needs to return int instead of void in order to signal when a starting frame larger than backtrace size is given, as caller and caller_locations needs to return nil and not the empty array in these cases. To handle cases where a range is provided with a negative end, and the backtrace size is needed to calculate the result to pass to rb_range_beg_len, add a backtrace_size static function to calculate the size, which copies the logic from backtrace_each. As backtrace_each only adds the backtrace lines requested, backtrace_to_*_ary can be simplified to always operate on the entire backtrace. Previously, caller_locations(1,1) was about 6.2 times slower for an 800 deep callstack compared to an empty callstack. With this new approach, it is only 1.3 times slower. It will always be somewhat slower as it still needs to scan the cfps from the top of the stack until it finds the first requested backtrace frame. This initializes the backtrace memory to zero. I do not think this is necessary, as from my analysis, nothing during the setting of the backtrace entries can cause a garbage collection, but it seems the safest approach, and it's unlikely the performance decrease is significant. This removes the rb_backtrace_t backtrace_base member. backtrace and backtrace_base were initialized to the same value, and neither is modified, so it doesn't make sense to have two pointers. This also removes LOCATION_TYPE_IFUNC from vm_backtrace.c, as the value is never set. Fixes [Bug #17031]
2020-08-28 01:17:36 +03:00
n = ALL_BACKTRACE_LINES;
break;
case Qnil:
return Qnil;
default:
lev = beg + lev_plus;
n = len;
break;
}
break;
}
case 2:
lev = NUM2LONG(level);
n = NUM2LONG(vn);
if (lev < 0) {
rb_raise(rb_eArgError, "negative level (%ld)", lev);
}
if (n < 0) {
rb_raise(rb_eArgError, "negative size (%ld)", n);
}
lev += lev_plus;
break;
default:
lev = n = 0; /* to avoid warning */
break;
}
if (n == 0) {
return rb_ary_new();
}
btval = rb_ec_partial_backtrace_object(ec, lev, n, &too_large, FALSE, FALSE);
Improve performance of partial backtraces Previously, backtrace_each fully populated the rb_backtrace_t with all backtrace frames, even if caller only requested a partial backtrace (e.g. Kernel#caller_locations(1, 1)). This changes backtrace_each to only add the requested frames to the rb_backtrace_t. To do this, backtrace_each needs to be passed the starting frame and number of frames values passed to Kernel#caller or #caller_locations. backtrace_each works from the top of the stack to the bottom, where the bottom is the current frame. Due to how the location for cfuncs is tracked using the location of the previous iseq, we need to store an extra frame for the previous iseq if we are limiting the backtrace and final backtrace frame (the first one stored) would be a cfunc and not an iseq. To limit the amount of work in this case, while scanning until the start of the requested backtrace, for each iseq, store the cfp. If the first backtrace frame we care about is a cfunc, use the stored cfp to find the related iseq. Use a function pointer to handle the storage of the cfp in the iteration arg, and also store the location of the extra frame in the iteration arg. backtrace_each needs to return int instead of void in order to signal when a starting frame larger than backtrace size is given, as caller and caller_locations needs to return nil and not the empty array in these cases. To handle cases where a range is provided with a negative end, and the backtrace size is needed to calculate the result to pass to rb_range_beg_len, add a backtrace_size static function to calculate the size, which copies the logic from backtrace_each. As backtrace_each only adds the backtrace lines requested, backtrace_to_*_ary can be simplified to always operate on the entire backtrace. Previously, caller_locations(1,1) was about 6.2 times slower for an 800 deep callstack compared to an empty callstack. With this new approach, it is only 1.3 times slower. It will always be somewhat slower as it still needs to scan the cfps from the top of the stack until it finds the first requested backtrace frame. This initializes the backtrace memory to zero. I do not think this is necessary, as from my analysis, nothing during the setting of the backtrace entries can cause a garbage collection, but it seems the safest approach, and it's unlikely the performance decrease is significant. This removes the rb_backtrace_t backtrace_base member. backtrace and backtrace_base were initialized to the same value, and neither is modified, so it doesn't make sense to have two pointers. This also removes LOCATION_TYPE_IFUNC from vm_backtrace.c, as the value is never set. Fixes [Bug #17031]
2020-08-28 01:17:36 +03:00
if (too_large) {
return Qnil;
}
if (to_str) {
Improve performance of partial backtraces Previously, backtrace_each fully populated the rb_backtrace_t with all backtrace frames, even if caller only requested a partial backtrace (e.g. Kernel#caller_locations(1, 1)). This changes backtrace_each to only add the requested frames to the rb_backtrace_t. To do this, backtrace_each needs to be passed the starting frame and number of frames values passed to Kernel#caller or #caller_locations. backtrace_each works from the top of the stack to the bottom, where the bottom is the current frame. Due to how the location for cfuncs is tracked using the location of the previous iseq, we need to store an extra frame for the previous iseq if we are limiting the backtrace and final backtrace frame (the first one stored) would be a cfunc and not an iseq. To limit the amount of work in this case, while scanning until the start of the requested backtrace, for each iseq, store the cfp. If the first backtrace frame we care about is a cfunc, use the stored cfp to find the related iseq. Use a function pointer to handle the storage of the cfp in the iteration arg, and also store the location of the extra frame in the iteration arg. backtrace_each needs to return int instead of void in order to signal when a starting frame larger than backtrace size is given, as caller and caller_locations needs to return nil and not the empty array in these cases. To handle cases where a range is provided with a negative end, and the backtrace size is needed to calculate the result to pass to rb_range_beg_len, add a backtrace_size static function to calculate the size, which copies the logic from backtrace_each. As backtrace_each only adds the backtrace lines requested, backtrace_to_*_ary can be simplified to always operate on the entire backtrace. Previously, caller_locations(1,1) was about 6.2 times slower for an 800 deep callstack compared to an empty callstack. With this new approach, it is only 1.3 times slower. It will always be somewhat slower as it still needs to scan the cfps from the top of the stack until it finds the first requested backtrace frame. This initializes the backtrace memory to zero. I do not think this is necessary, as from my analysis, nothing during the setting of the backtrace entries can cause a garbage collection, but it seems the safest approach, and it's unlikely the performance decrease is significant. This removes the rb_backtrace_t backtrace_base member. backtrace and backtrace_base were initialized to the same value, and neither is modified, so it doesn't make sense to have two pointers. This also removes LOCATION_TYPE_IFUNC from vm_backtrace.c, as the value is never set. Fixes [Bug #17031]
2020-08-28 01:17:36 +03:00
r = backtrace_to_str_ary(btval);
}
else {
Improve performance of partial backtraces Previously, backtrace_each fully populated the rb_backtrace_t with all backtrace frames, even if caller only requested a partial backtrace (e.g. Kernel#caller_locations(1, 1)). This changes backtrace_each to only add the requested frames to the rb_backtrace_t. To do this, backtrace_each needs to be passed the starting frame and number of frames values passed to Kernel#caller or #caller_locations. backtrace_each works from the top of the stack to the bottom, where the bottom is the current frame. Due to how the location for cfuncs is tracked using the location of the previous iseq, we need to store an extra frame for the previous iseq if we are limiting the backtrace and final backtrace frame (the first one stored) would be a cfunc and not an iseq. To limit the amount of work in this case, while scanning until the start of the requested backtrace, for each iseq, store the cfp. If the first backtrace frame we care about is a cfunc, use the stored cfp to find the related iseq. Use a function pointer to handle the storage of the cfp in the iteration arg, and also store the location of the extra frame in the iteration arg. backtrace_each needs to return int instead of void in order to signal when a starting frame larger than backtrace size is given, as caller and caller_locations needs to return nil and not the empty array in these cases. To handle cases where a range is provided with a negative end, and the backtrace size is needed to calculate the result to pass to rb_range_beg_len, add a backtrace_size static function to calculate the size, which copies the logic from backtrace_each. As backtrace_each only adds the backtrace lines requested, backtrace_to_*_ary can be simplified to always operate on the entire backtrace. Previously, caller_locations(1,1) was about 6.2 times slower for an 800 deep callstack compared to an empty callstack. With this new approach, it is only 1.3 times slower. It will always be somewhat slower as it still needs to scan the cfps from the top of the stack until it finds the first requested backtrace frame. This initializes the backtrace memory to zero. I do not think this is necessary, as from my analysis, nothing during the setting of the backtrace entries can cause a garbage collection, but it seems the safest approach, and it's unlikely the performance decrease is significant. This removes the rb_backtrace_t backtrace_base member. backtrace and backtrace_base were initialized to the same value, and neither is modified, so it doesn't make sense to have two pointers. This also removes LOCATION_TYPE_IFUNC from vm_backtrace.c, as the value is never set. Fixes [Bug #17031]
2020-08-28 01:17:36 +03:00
r = backtrace_to_location_ary(btval);
}
RB_GC_GUARD(btval);
return r;
}
static VALUE
thread_backtrace_to_ary(int argc, const VALUE *argv, VALUE thval, int to_str)
{
rb_thread_t *target_th = rb_thread_ptr(thval);
if (target_th->to_kill || target_th->status == THREAD_KILLED)
return Qnil;
return ec_backtrace_to_ary(target_th->ec, argc, argv, 0, 0, to_str);
}
VALUE
rb_vm_thread_backtrace(int argc, const VALUE *argv, VALUE thval)
{
return thread_backtrace_to_ary(argc, argv, thval, 1);
}
VALUE
rb_vm_thread_backtrace_locations(int argc, const VALUE *argv, VALUE thval)
{
return thread_backtrace_to_ary(argc, argv, thval, 0);
}
2022-08-06 04:13:20 +03:00
VALUE
rb_vm_backtrace(int argc, const VALUE * argv, struct rb_execution_context_struct * ec)
{
return ec_backtrace_to_ary(ec, argc, argv, 0, 0, 1);
}
2022-08-06 04:13:20 +03:00
VALUE
rb_vm_backtrace_locations(int argc, const VALUE * argv, struct rb_execution_context_struct * ec)
{
return ec_backtrace_to_ary(ec, argc, argv, 0, 0, 0);
}
/*
* call-seq:
* caller(start=1, length=nil) -> array or nil
* caller(range) -> array or nil
*
* Returns the current execution stack---an array containing strings in
* the form <code>file:line</code> or <code>file:line: in
* `method'</code>.
*
* The optional _start_ parameter determines the number of initial stack
* entries to omit from the top of the stack.
*
* A second optional +length+ parameter can be used to limit how many entries
* are returned from the stack.
*
* Returns +nil+ if _start_ is greater than the size of
* current execution stack.
*
* Optionally you can pass a range, which will return an array containing the
* entries within the specified range.
*
* def a(skip)
* caller(skip)
* end
* def b(skip)
* a(skip)
* end
* def c(skip)
* b(skip)
* end
* c(0) #=> ["prog:2:in `a'", "prog:5:in `b'", "prog:8:in `c'", "prog:10:in `<main>'"]
* c(1) #=> ["prog:5:in `b'", "prog:8:in `c'", "prog:11:in `<main>'"]
* c(2) #=> ["prog:8:in `c'", "prog:12:in `<main>'"]
* c(3) #=> ["prog:13:in `<main>'"]
* c(4) #=> []
* c(5) #=> nil
*/
static VALUE
rb_f_caller(int argc, VALUE *argv, VALUE _)
{
return ec_backtrace_to_ary(GET_EC(), argc, argv, 1, 1, 1);
}
/*
* call-seq:
* caller_locations(start=1, length=nil) -> array or nil
* caller_locations(range) -> array or nil
*
* Returns the current execution stack---an array containing
* backtrace location objects.
*
* See Thread::Backtrace::Location for more information.
*
* The optional _start_ parameter determines the number of initial stack
* entries to omit from the top of the stack.
*
* A second optional +length+ parameter can be used to limit how many entries
* are returned from the stack.
*
* Returns +nil+ if _start_ is greater than the size of
* current execution stack.
*
* Optionally you can pass a range, which will return an array containing the
* entries within the specified range.
*/
static VALUE
rb_f_caller_locations(int argc, VALUE *argv, VALUE _)
{
return ec_backtrace_to_ary(GET_EC(), argc, argv, 1, 1, 0);
}
/*
* call-seq:
* Thread.each_caller_location{ |loc| ... } -> nil
*
* Yields each frame of the current execution stack as a
* backtrace location object.
*/
static VALUE
each_caller_location(VALUE unused)
{
rb_ec_partial_backtrace_object(GET_EC(), 2, ALL_BACKTRACE_LINES, NULL, FALSE, TRUE);
return Qnil;
}
/* called from Init_vm() in vm.c */
void
Init_vm_backtrace(void)
{
2021-12-20 10:07:25 +03:00
/*
* An internal representation of the backtrace. The user will never interact with
* objects of this class directly, but class methods can be used to get backtrace
* settings of the current session.
*/
rb_cBacktrace = rb_define_class_under(rb_cThread, "Backtrace", rb_cObject);
rb_define_alloc_func(rb_cBacktrace, backtrace_alloc);
rb_undef_method(CLASS_OF(rb_cBacktrace), "new");
rb_marshal_define_compat(rb_cBacktrace, rb_cArray, backtrace_dump_data, backtrace_load_data);
rb_define_singleton_method(rb_cBacktrace, "limit", backtrace_limit, 0);
/*
* An object representation of a stack frame, initialized by
* Kernel#caller_locations.
*
* For example:
*
* # caller_locations.rb
* def a(skip)
* caller_locations(skip)
* end
* def b(skip)
* a(skip)
* end
* def c(skip)
* b(skip)
* end
*
* c(0..2).map do |call|
* puts call.to_s
* end
*
* Running <code>ruby caller_locations.rb</code> will produce:
*
* caller_locations.rb:2:in `a'
* caller_locations.rb:5:in `b'
* caller_locations.rb:8:in `c'
*
* Here's another example with a slightly different result:
*
* # foo.rb
* class Foo
* attr_accessor :locations
* def initialize(skip)
* @locations = caller_locations(skip)
* end
* end
*
* Foo.new(0..2).locations.map do |call|
* puts call.to_s
* end
*
* Now run <code>ruby foo.rb</code> and you should see:
*
* init.rb:4:in `initialize'
* init.rb:8:in `new'
* init.rb:8:in `<main>'
*/
rb_cBacktraceLocation = rb_define_class_under(rb_cBacktrace, "Location", rb_cObject);
rb_undef_alloc_func(rb_cBacktraceLocation);
rb_undef_method(CLASS_OF(rb_cBacktraceLocation), "new");
rb_define_method(rb_cBacktraceLocation, "lineno", location_lineno_m, 0);
rb_define_method(rb_cBacktraceLocation, "label", location_label_m, 0);
rb_define_method(rb_cBacktraceLocation, "base_label", location_base_label_m, 0);
rb_define_method(rb_cBacktraceLocation, "path", location_path_m, 0);
rb_define_method(rb_cBacktraceLocation, "absolute_path", location_absolute_path_m, 0);
rb_define_method(rb_cBacktraceLocation, "to_s", location_to_str_m, 0);
rb_define_method(rb_cBacktraceLocation, "inspect", location_inspect_m, 0);
rb_define_global_function("caller", rb_f_caller, -1);
rb_define_global_function("caller_locations", rb_f_caller_locations, -1);
rb_define_singleton_method(rb_cThread, "each_caller_location", each_caller_location, 0);
}
/* debugger API */
RUBY_SYMBOL_EXPORT_BEGIN
RUBY_SYMBOL_EXPORT_END
struct rb_debug_inspector_struct {
rb_execution_context_t *ec;
rb_control_frame_t *cfp;
VALUE backtrace;
VALUE contexts; /* [[klass, binding, iseq, cfp], ...] */
long backtrace_size;
};
enum {
CALLER_BINDING_SELF,
CALLER_BINDING_CLASS,
CALLER_BINDING_BINDING,
CALLER_BINDING_ISEQ,
CALLER_BINDING_CFP,
CALLER_BINDING_DEPTH,
};
struct collect_caller_bindings_data {
VALUE ary;
const rb_execution_context_t *ec;
};
static void
collect_caller_bindings_init(void *arg, size_t size)
{
/* */
}
static VALUE
get_klass(const rb_control_frame_t *cfp)
{
VALUE klass;
if (rb_vm_control_frame_id_and_class(cfp, 0, 0, &klass)) {
if (RB_TYPE_P(klass, T_ICLASS)) {
return RBASIC(klass)->klass;
}
else {
return klass;
}
}
else {
return Qnil;
}
}
static int
frame_depth(const rb_execution_context_t *ec, const rb_control_frame_t *cfp)
{
VM_ASSERT(RUBY_VM_END_CONTROL_FRAME(ec) >= cfp);
return (int)(RUBY_VM_END_CONTROL_FRAME(ec) - cfp);
}
static void
collect_caller_bindings_iseq(void *arg, const rb_control_frame_t *cfp)
{
struct collect_caller_bindings_data *data = (struct collect_caller_bindings_data *)arg;
VALUE frame = rb_ary_new2(6);
rb_ary_store(frame, CALLER_BINDING_SELF, cfp->self);
rb_ary_store(frame, CALLER_BINDING_CLASS, get_klass(cfp));
rb_ary_store(frame, CALLER_BINDING_BINDING, GC_GUARDED_PTR(cfp)); /* create later */
2015-07-22 01:52:59 +03:00
rb_ary_store(frame, CALLER_BINDING_ISEQ, cfp->iseq ? (VALUE)cfp->iseq : Qnil);
rb_ary_store(frame, CALLER_BINDING_CFP, GC_GUARDED_PTR(cfp));
rb_ary_store(frame, CALLER_BINDING_DEPTH, INT2FIX(frame_depth(data->ec, cfp)));
rb_ary_push(data->ary, frame);
}
static void
collect_caller_bindings_cfunc(void *arg, const rb_control_frame_t *cfp, ID mid)
{
struct collect_caller_bindings_data *data = (struct collect_caller_bindings_data *)arg;
VALUE frame = rb_ary_new2(6);
rb_ary_store(frame, CALLER_BINDING_SELF, cfp->self);
rb_ary_store(frame, CALLER_BINDING_CLASS, get_klass(cfp));
rb_ary_store(frame, CALLER_BINDING_BINDING, Qnil); /* not available */
rb_ary_store(frame, CALLER_BINDING_ISEQ, Qnil); /* not available */
rb_ary_store(frame, CALLER_BINDING_CFP, GC_GUARDED_PTR(cfp));
rb_ary_store(frame, CALLER_BINDING_DEPTH, INT2FIX(frame_depth(data->ec, cfp)));
rb_ary_push(data->ary, frame);
}
static VALUE
collect_caller_bindings(const rb_execution_context_t *ec)
{
int i;
VALUE result;
struct collect_caller_bindings_data data = {
rb_ary_new(), ec
};
backtrace_each(ec,
collect_caller_bindings_init,
collect_caller_bindings_iseq,
collect_caller_bindings_cfunc,
&data);
result = rb_ary_reverse(data.ary);
/* bindings should be created from top of frame */
for (i=0; i<RARRAY_LEN(result); i++) {
VALUE entry = rb_ary_entry(result, i);
VALUE cfp_val = rb_ary_entry(entry, CALLER_BINDING_BINDING);
if (!NIL_P(cfp_val)) {
rb_control_frame_t *cfp = GC_GUARDED_PTR_REF(cfp_val);
rb_ary_store(entry, CALLER_BINDING_BINDING, rb_vm_make_binding(ec, cfp));
}
}
return result;
}
/*
* Note that the passed `rb_debug_inspector_t' will be disabled
* after `rb_debug_inspector_open'.
*/
VALUE
rb_debug_inspector_open(rb_debug_inspector_func_t func, void *data)
{
rb_debug_inspector_t dbg_context;
rb_execution_context_t *ec = GET_EC();
enum ruby_tag_type state;
volatile VALUE MAYBE_UNUSED(result);
/* escape all env to heap */
rb_vm_stack_to_heap(ec);
dbg_context.ec = ec;
dbg_context.cfp = dbg_context.ec->cfp;
dbg_context.backtrace = rb_ec_backtrace_location_ary(ec, BACKTRACE_START, ALL_BACKTRACE_LINES, FALSE);
dbg_context.backtrace_size = RARRAY_LEN(dbg_context.backtrace);
dbg_context.contexts = collect_caller_bindings(ec);
EC_PUSH_TAG(ec);
if ((state = EC_EXEC_TAG()) == TAG_NONE) {
result = (*func)(&dbg_context, data);
}
EC_POP_TAG();
/* invalidate bindings? */
if (state) {
EC_JUMP_TAG(ec, state);
}
return result;
}
static VALUE
frame_get(const rb_debug_inspector_t *dc, long index)
{
if (index < 0 || index >= dc->backtrace_size) {
rb_raise(rb_eArgError, "no such frame");
}
return rb_ary_entry(dc->contexts, index);
}
VALUE
rb_debug_inspector_frame_self_get(const rb_debug_inspector_t *dc, long index)
{
VALUE frame = frame_get(dc, index);
return rb_ary_entry(frame, CALLER_BINDING_SELF);
}
VALUE
rb_debug_inspector_frame_class_get(const rb_debug_inspector_t *dc, long index)
{
VALUE frame = frame_get(dc, index);
return rb_ary_entry(frame, CALLER_BINDING_CLASS);
}
VALUE
rb_debug_inspector_frame_binding_get(const rb_debug_inspector_t *dc, long index)
{
VALUE frame = frame_get(dc, index);
return rb_ary_entry(frame, CALLER_BINDING_BINDING);
}
VALUE
rb_debug_inspector_frame_iseq_get(const rb_debug_inspector_t *dc, long index)
{
VALUE frame = frame_get(dc, index);
2015-07-22 01:52:59 +03:00
VALUE iseq = rb_ary_entry(frame, CALLER_BINDING_ISEQ);
return RTEST(iseq) ? rb_iseqw_new((rb_iseq_t *)iseq) : Qnil;
}
VALUE
rb_debug_inspector_frame_depth(const rb_debug_inspector_t *dc, long index)
{
VALUE frame = frame_get(dc, index);
return rb_ary_entry(frame, CALLER_BINDING_DEPTH);
}
VALUE
rb_debug_inspector_current_depth(void)
{
rb_execution_context_t *ec = GET_EC();
return INT2FIX(frame_depth(ec, ec->cfp));
}
VALUE
rb_debug_inspector_backtrace_locations(const rb_debug_inspector_t *dc)
{
return dc->backtrace;
}
int
rb_profile_frames(int start, int limit, VALUE *buff, int *lines)
{
int i;
const rb_execution_context_t *ec = GET_EC();
const rb_control_frame_t *cfp = ec->cfp, *end_cfp = RUBY_VM_END_CONTROL_FRAME(ec);
const rb_callable_method_entry_t *cme;
// If this function is called inside a thread after thread creation, but
// before the CFP has been created, just return 0. This can happen when
// sampling via signals. Threads can be interrupted randomly by the
// signal, including during the time after the thread has been created, but
// before the CFP has been allocated
if (!cfp) {
return 0;
}
// Skip dummy frame; see `rb_ec_partial_backtrace_object` for details
end_cfp = RUBY_VM_NEXT_CONTROL_FRAME(end_cfp);
for (i=0; i<limit && cfp != end_cfp;) {
if (VM_FRAME_RUBYFRAME_P(cfp) && cfp->pc != 0) {
if (start > 0) {
start--;
continue;
}
2022-07-21 19:23:58 +03:00
/* record frame info */
cme = rb_vm_frame_method_entry(cfp);
if (cme && cme->def->type == VM_METHOD_TYPE_ISEQ) {
buff[i] = (VALUE)cme;
}
else {
2015-07-22 01:52:59 +03:00
buff[i] = (VALUE)cfp->iseq;
}
2022-07-21 19:23:58 +03:00
if (lines) lines[i] = calc_lineno(cfp->iseq, cfp->pc);
2022-07-21 19:23:58 +03:00
i++;
}
else {
cme = rb_vm_frame_method_entry(cfp);
if (cme && cme->def->type == VM_METHOD_TYPE_CFUNC) {
buff[i] = (VALUE)cme;
if (lines) lines[i] = 0;
i++;
}
}
cfp = RUBY_VM_PREVIOUS_CONTROL_FRAME(cfp);
}
return i;
}
2015-07-22 01:52:59 +03:00
static const rb_iseq_t *
frame2iseq(VALUE frame)
{
2021-10-03 16:34:45 +03:00
if (NIL_P(frame)) return NULL;
if (RB_TYPE_P(frame, T_IMEMO)) {
2015-07-22 01:52:59 +03:00
switch (imemo_type(frame)) {
case imemo_iseq:
return (const rb_iseq_t *)frame;
case imemo_ment:
{
const rb_callable_method_entry_t *cme = (rb_callable_method_entry_t *)frame;
2015-07-22 01:52:59 +03:00
switch (cme->def->type) {
case VM_METHOD_TYPE_ISEQ:
return cme->def->body.iseq.iseqptr;
default:
return NULL;
}
}
default:
2015-07-22 01:52:59 +03:00
break;
}
}
rb_bug("frame2iseq: unreachable");
}
VALUE
rb_profile_frame_path(VALUE frame)
{
2015-07-22 01:52:59 +03:00
const rb_iseq_t *iseq = frame2iseq(frame);
return iseq ? rb_iseq_path(iseq) : Qnil;
}
static const rb_callable_method_entry_t *
cframe(VALUE frame)
{
2021-10-03 16:34:45 +03:00
if (NIL_P(frame)) return NULL;
if (RB_TYPE_P(frame, T_IMEMO)) {
switch (imemo_type(frame)) {
case imemo_ment:
{
const rb_callable_method_entry_t *cme = (rb_callable_method_entry_t *)frame;
switch (cme->def->type) {
case VM_METHOD_TYPE_CFUNC:
return cme;
default:
return NULL;
}
}
default:
return NULL;
}
}
return NULL;
}
VALUE
rb_profile_frame_absolute_path(VALUE frame)
{
if (cframe(frame)) {
static VALUE cfunc_str = Qfalse;
if (!cfunc_str) {
cfunc_str = rb_str_new_literal("<cfunc>");
rb_gc_register_mark_object(cfunc_str);
}
return cfunc_str;
}
2015-07-22 01:52:59 +03:00
const rb_iseq_t *iseq = frame2iseq(frame);
return iseq ? rb_iseq_realpath(iseq) : Qnil;
}
VALUE
rb_profile_frame_label(VALUE frame)
{
2015-07-22 01:52:59 +03:00
const rb_iseq_t *iseq = frame2iseq(frame);
return iseq ? rb_iseq_label(iseq) : Qnil;
}
VALUE
rb_profile_frame_base_label(VALUE frame)
{
2015-07-22 01:52:59 +03:00
const rb_iseq_t *iseq = frame2iseq(frame);
return iseq ? rb_iseq_base_label(iseq) : Qnil;
}
VALUE
rb_profile_frame_first_lineno(VALUE frame)
{
2015-07-22 01:52:59 +03:00
const rb_iseq_t *iseq = frame2iseq(frame);
return iseq ? rb_iseq_first_lineno(iseq) : Qnil;
}
static VALUE
frame2klass(VALUE frame)
{
2021-10-03 16:34:45 +03:00
if (NIL_P(frame)) return Qnil;
if (RB_TYPE_P(frame, T_IMEMO)) {
const rb_callable_method_entry_t *cme = (rb_callable_method_entry_t *)frame;
if (imemo_type(frame) == imemo_ment) {
return cme->defined_class;
}
}
return Qnil;
}
VALUE
rb_profile_frame_classpath(VALUE frame)
{
VALUE klass = frame2klass(frame);
if (klass && !NIL_P(klass)) {
if (RB_TYPE_P(klass, T_ICLASS)) {
klass = RBASIC(klass)->klass;
}
else if (FL_TEST(klass, FL_SINGLETON)) {
klass = RCLASS_ATTACHED_OBJECT(klass);
if (!RB_TYPE_P(klass, T_CLASS) && !RB_TYPE_P(klass, T_MODULE))
return rb_sprintf("#<%s:%p>", rb_class2name(rb_obj_class(klass)), (void*)klass);
}
return rb_class_path(klass);
}
else {
return Qnil;
}
}
VALUE
rb_profile_frame_singleton_method_p(VALUE frame)
{
VALUE klass = frame2klass(frame);
2021-08-02 06:06:44 +03:00
return RBOOL(klass && !NIL_P(klass) && FL_TEST(klass, FL_SINGLETON));
}
VALUE
rb_profile_frame_method_name(VALUE frame)
{
const rb_callable_method_entry_t *cme = cframe(frame);
if (cme) {
ID mid = cme->def->original_id;
return id2str(mid);
}
2015-07-22 01:52:59 +03:00
const rb_iseq_t *iseq = frame2iseq(frame);
return iseq ? rb_iseq_method_name(iseq) : Qnil;
}
static VALUE
qualified_method_name(VALUE frame, VALUE method_name)
{
if (method_name != Qnil) {
VALUE classpath = rb_profile_frame_classpath(frame);
VALUE singleton_p = rb_profile_frame_singleton_method_p(frame);
if (classpath != Qnil) {
return rb_sprintf("%"PRIsVALUE"%s%"PRIsVALUE,
classpath, singleton_p == Qtrue ? "." : "#", method_name);
}
else {
return method_name;
}
}
else {
return Qnil;
}
}
VALUE
rb_profile_frame_qualified_method_name(VALUE frame)
{
VALUE method_name = rb_profile_frame_method_name(frame);
return qualified_method_name(frame, method_name);
}
VALUE
rb_profile_frame_full_label(VALUE frame)
{
const rb_callable_method_entry_t *cme = cframe(frame);
if (cme) {
ID mid = cme->def->original_id;
VALUE method_name = id2str(mid);
return qualified_method_name(frame, method_name);
}
VALUE label = rb_profile_frame_label(frame);
VALUE base_label = rb_profile_frame_base_label(frame);
VALUE qualified_method_name = rb_profile_frame_qualified_method_name(frame);
if (NIL_P(qualified_method_name) || base_label == qualified_method_name) {
return label;
}
else {
long label_length = RSTRING_LEN(label);
long base_label_length = RSTRING_LEN(base_label);
int prefix_len = rb_long2int(label_length - base_label_length);
return rb_sprintf("%.*s%"PRIsVALUE, prefix_len, RSTRING_PTR(label), qualified_method_name);
}
}