Unsynchronized ScriptableObject.getSlotToSet contains references/modifications
to the slots array which is no go under multithreading. The attached patch
replaces references to slots by references to its local copy and moves code
to allocate the initial array to synchronized addSlot.
The patch also replace throwing of RuntimeException in case of broken code by
if (Context.check && badCondition) Context.codeBug();
Regards, Igor
also has a significant regression introduced in it. The default compiler
not only works, but also is noticably faster. Ant takes care of the
selection of the compiler automatically based on the JDK level, so the
following patch should make things better all around.
Patch from Igor:
The 2 attached patches allow to avoid wrapping of array indexes to Double object
when Interpreter knows that the index is an integer number. It speed up array
benchmark by 5-10%
array_access.diff adds to ScriptRuntime getStrIdElem and setStrIdElem to get/set
properties which known to be strings plus it modifies NativeArray to use these methods.
interpreter.diff contains the Interpreter modifications to call get/setElem for
integer or string properties when the property type is known for sure.
errors, etc.''):
We now ReportStatementTooLarge only if
- a jump offset overflows 32 bits, signed;
- there are 2**32 or more span dependencies in a script;
- a backpatch chain link is more than (2**30 - 1) bytecodes long;
- a source note's distance from the last note, or from script main entry
point, is > 0x7fffff bytes.
Narrative of the patch, by file:
- js.c
The js_SrcNoteName array of const char * is now a js_SrcNoteSpec array of
"specifiers", structs that include a const char *name member. Also, due to
span-dependent jumps at the ends of basic blocks where the decompiler knows
the basic block length, but not the jump format, we need an offset operand
for SRC_COND, SRC_IF_ELSE, and SRC_WHILE (to tell the distance from the
branch bytecode after the condition expression to the span-dependent jump).
- jsarena.[ch]
JS arenas are used mainly for last-in-first-out allocation with _en masse_
release to the malloc pool (or, optionally, to a private freelist). But
the code generator needs to allocate and grow (by doubling, to avoid O(n^2)
growth) allocations that hold bytecode, source notes, and span-dependency
records. This exception to LIFO allocation works by claiming an entire
arena from the pool and realloc'ing it, as soon as the allocation size
reaches the pool's default arena size. Call such an allocation a "large
single allocation".
This patch adds a new arena API, JS_ArenaFreeAllocation, which can be used
to free a large single allocation. If called with an allocation that's not
a large single allocation, it will nevertheless attempt to retract the arena
containing that allocation, if the allocation is last within its arena.
Thus JS_ArenaFreeAllocation adds a non-LIFO "free" special case to match the
non-LIFO "grow" special case already implemented under JS_ARENA_GROW for
large single allocations.
The code generator still benefits via this extension to arenas, over purely
manual malloc/realloc/free, by virtue of _en masse_ free (JS_ARENA_RELEASE
after code generation has completed, successfully or not).
To avoid searching for the previous arena, in order to update its next
member upon reallocation of the arena containing a large single allocation,
the oversized arena has a back-pointer to that next member stored (but not
as allocable space within the arena) in a (JSArena **) footer at its end.
- jscntxt.c
I've observed for many scripts that the bytes of source notes and bytecode
are of comparable lengths, but only now am I fixing the default arena size
for cx->notePool to match the size for cx->codePool (1024 instead of 256).
- jsemit.c
Span-dependent instructions in JS bytecode consist of the jump (JOF_JUMP)
and switch (JOF_LOOKUPSWITCH, JOF_TABLESWITCH) format opcodes, subdivided
into unconditional (gotos and gosubs), and conditional jumps or branches
(which pop a value, test it, and jump depending on its value). Most jumps
have just one immediate operand, a signed offset from the jump opcode's pc
to the target bytecode. The lookup and table switch opcodes may contain
many jump offsets.
This patch adds "X" counterparts to the opcodes/formats (X is suffixed, btw,
to prefer JSOP_ORX and thereby to avoid colliding on the JSOP_XOR name for
the extended form of the JSOP_OR branch opcode). The unextended or short
formats have 16-bit signed immediate offset operands, the extended or long
formats have 32-bit signed immediates. The span-dependency problem consists
of selecting as few long instructions as possible, or about as few -- since
jumps can span other jumps, extending one jump may cause another to need to
be extended.
Most JS scripts are short, so need no extended jumps. We optimize for this
case by generating short jumps until we know a long jump is needed. After
that point, we keep generating short jumps, but each jump's 16-bit immediate
offset operand is actually an unsigned index into cg->spanDeps, an array of
JSSpanDep structs. Each struct tells the top offset in the script of the
opcode, the "before" offset of the jump (which will be the same as top for
simplex jumps, but which will index further into the bytecode array for a
non-initial jump offset in a lookup or table switch), the after "offset"
adjusted during span-dependent instruction selection (initially the same
value as the "before" offset), and the jump target (more below).
Since we generate cg->spanDeps lazily, from within js_SetJumpOffset, we must
ensure that all bytecode generated so far can be inspected to discover where
the jump offset immediate operands lie within CG_CODE(cg). But the bonus is
that we generate span-dependency records sorted by their offsets, so we can
binary-search when trying to find a JSSpanDep for a given bytecode offset,
or the nearest JSSpanDep at or above a given pc.
To avoid limiting scripts to 64K jumps, if the cg->spanDeps index overflows
65534, we store SPANDEP_INDEX_HUGE in the jump's immediate operand. This
tells us that we need to binary-search for the cg->spanDeps entry by the
jump opcode's bytecode offset (sd->before).
Jump targets need to be maintained in a data structure that lets us look
up an already-known target by its address (jumps may have a common target),
and that also lets us update the addresses (script-relative, a.k.a. absolute
offsets) of targets that come after a jump target (for when a jump below
that target needs to be extended). We use an AVL tree, implemented using
recursion, but with some tricky optimizations to its height-balancing code
(see http://www.enteract.com/~bradapp/ftp/src/libs/C++/AvlTrees.html).
A final wrinkle: backpatch chains are linked by jump-to-jump offsets with
positive sign, even though they link "backward" (i.e., toward lower bytecode
address). We don't want to waste space and search time in the AVL tree for
such temporary backpatch deltas, so we use a single-bit wildcard scheme to
tag true JSJumpTarget pointers and encode untagged, signed (positive) deltas
in JSSpanDep.target pointers, depending on whether the JSSpanDep has a known
target, or is still awaiting backpatching.
Note that backpatch chains would present a problem for BuildSpanDepTable,
which inspects bytecode to build cg->spanDeps on demand, when the first
short jump offset overflows. To solve this temporary problem, we emit a
proxy bytecode (JSOP_BACKPATCH; JSOP_BACKPATCH_PUSH for jumps that push a
result on the interpreter's stack, namely JSOP_GOSUB; or JSOP_BACKPATCH_POP
for branch ops) whose nuses/ndefs counts help keep the stack balanced, but
whose opcode format distinguishes its backpatch delta immediate operand from
a normal jump offset.
The cg->spanDeps array and JSJumpTarget structs are allocated from the
cx->tempPool arena-pool. This created a LIFO vs. non-LIFO conflict: there
were two places under the TOK_SWITCH case in js_EmitTree that used tempPool
to allocate and release a chunk of memory, during whose lifetime JSSpanDep
and/or JSJumpTarget structs might also be allocated from tempPool -- the
ensuing release would prove disastrous. These bitmap and table temporaries
are now allocated from the malloc heap.
- jsinterp.c
Straightforward cloning and JUMP => JUMPX mutating of the jump and switch
format bytecode cases.
- jsobj.c
Silence warnings about %p used without (void *) casts.
- jsopcode.c
Massive and scary decompiler whackage to cope with extended jumps, using
source note offsets to help find jumps whose format (short or long) can't
be discovered from properties of prior instructions in the script.
One cute hack here: long || and && expressions are broken up to wrap before
the 80th column, with the operator at the end of each non-terminal line.
- jsopcode.h, jsopcode.tbl
The new extended jump opcodes, formats, and fundamental parameterization
macros. Also, more comments.
- jsparse.c
Random and probably only aesthetic fix to avoid decorating a foo[i]++ or
--foo[i] parse tree node with JSOP_SETCALL, wrongly (only foo(i)++ or
--foo(i), or the other post- or prefix form operator, should have such an
opcode decoration on its parse tree).
- jsscript.h
Random macro naming sanity: use trailing _ rather than leading _ for macro
local variables in order to avoid invading the standard C global namespace.
We have a tool that looks for a scary noop case of assigning an instance field
to itself. this usually comes from a constructor that assigns a argument to a
instance field with the same name and then later the argument changes name. we
ran our tool on all of our classes we have in our classpath here and found this
problem in your code.
rhino1_5R2/src/org/mozilla/javascript/regexp/NativeRegExp.java line 159 it has:
this.flags = flags;
This seems to be a bad cut and paste from the CompilerState constructor on line
2155. or has some initialization that used to work been lost?
There is a bug in JavaScriptException which prevents it from being used with
out a Rhino Context. When the getMessage() method is invoked on it, the
exception goes to the ScriptRuntime to toString the value. If you have
already exited your context, the runtime will throw an error. The solution
is to simply remove the overridden getMessage method from
JavaScriptException. JavaScriptException's constructor calls the Exception
constructor with the toString'ed value. The default implementation of
getMessage will return the exception message.
Jeff
I'm having problems getting inner class objects with Rhino.
I create a Hashmap, which is an implementation of Map. Map.Entry is an
inner interface of Map with key-value pairs. If I have a Map object,
"property", I should be able to get the key element with the expression
"property.key".
When I look at the "property" class name that Rhino returns I get:
"java.util.HashMap$Entry". I don't believe Rhino has a notion of the
inner Map.Entry object. The expression "property" succeeds. The
expression "property.key", which should retrieve the Map.Entry
keyValue(), fails with a "unexpected IllegalAccessException accessing
Java field".
I'm including a simple example that illustrates the problem. I hope you
can shed some light on this. Thanks!
Justyna
< Justyna.Horwat@Sun.com >
----
import java.io.*;
import java.util.*;
import org.mozilla.javascript.*;
public class MapTest {
public static void main(String argv[]) {
Test test = new Test();
test.testMap();
}
}
class Test {
Map map;
Set set;
Iterator it;
Map.Entry entry;
public void testMap() {
System.out.println("testMap");
map = new HashMap();
populate();
set = map.entrySet();
it = set.iterator();
// let's see if Map is populated correctly
while (it.hasNext()) {
entry = (Map.Entry) it.next();
System.out.println("entry: " + entry.getClass().getName());
System.out.println("key: " + entry.getKey());
System.out.println("value: " + entry.getValue());
}
evaluate();
}
void populate() {
map.put("firstKey", "firstValue");
map.put("secondKey", "secondValue");
map.put("thirdKey", "thirdValue");
map.put("fourthKey", "fourthValue");
}
public void evaluate() {
Context cx = Context.enter();
Scriptable scope = cx.initStandardObjects(null);
set = map.entrySet();
it = set.iterator();
while (it.hasNext()) {
entry = (Map.Entry) it.next();
scope.put("property", scope, cx.toObject(entry,scope));
}
Object eval = null;
try {
// attempt to get Map.Entry key value using Rhino
eval = cx.evaluateString(scope, "property.key", "", 0,
null);
// Unwrap scoped object
if (eval instanceof Wrapper)
eval = ((Wrapper) eval).unwrap();
} catch (JavaScriptException jse) {
System.out.println("EXCEPTION: " + jse.getMessage());
}
// DELETE
System.out.println("RHINO result: " + eval + ":");
System.out.println("RHINO class: " + eval.getClass().getName());
}
}
with JSRESOLVE_ASSIGNING, wrongly), plus a few miscellaneous bugfixes.
- Combine the JSStackFrame members constructing, special, overrides, and
reserved into a uint32 flags member.
- Separate JOF_ASSIGNING from the JOF_SET bytecode format flag, and impute
JSRESOLVE_ASSIGNING from the presence of JOF_ASSIGNING among the current
opcode's format flags. To handle the for-in loop opcodes, which do more
than simply assign -- in particular, they do property lookups whose resolve
hook outcalls should not be flagged with JSRESOLVE_ASSIGNING -- a new frame
flag, JSFRAME_ASSIGNING, has been added.
- Fix interpreter version selection to respect JS_SetVersion, whose effect on
cx->version is "sticky".
- Fix js_DecompileValueGenerator to deal with JSOP_ENUMELEM -- it never had,
as this testcase shows (it crashes without this patch):
version(120);
eval("function fe(s) { for (it[s] in this); }");
try { fe('rdonly'); } catch (e) { print(e); }
My optimization for PreorderNodeIterator has a bug that would cause an attempt
to access stack[-1] in
currentParent = (current == null) ? null : stack[stackTop - 1];
when current refers to a start node sibling. This is not visible in Rhino because
currently PreorderNodeIterator is always started from nodes with node.next == null.
iter.diff fixes that plus it removes currentParent field because it is always
available as stack[stackTop - 1] and code to access its value are executed less
frequently than the lines to update it in nextNode
Regarsd, Igor
As profiler data show, the execution time of the nextNode and replaceCurrent
methods in PreorderNodeIterator contribute quite significantly to the total
time to run Context.compileReader.
replaceCurrent is slow because it calls Node.replaceChild which have to
iterate through all previous siblings to find the nearest to the current.
But it is easy to avoid this search by caching the previous sibling of the
current while iterating over the node tree in nextNode.
nextNode slowness is attributed to the usage of java.lang.Stack which is
expensive due to its synchronized methods. In the attched patch I replaced
it by the explicit array management.
It allows to cut Context.compileReader time by 5%-30% when processing
20K-3MB sources assembled form JS files in the test suite.
Note form omj/Parser.java:
* OPT source info collection is a potential performance bottleneck;
* Source wraps a java.lang.StringBuffer, which is synchronized. It
* might be faster to implement Source with its own char buffer and
* toString method.
It is indeed a bottleneck under JDK 1.1. When I replaced StringBuffer
by a char array (see the attached patch), execution time of
Context.compileReader decreased by 15%: to test I combined a few test
cases to get a 3MB JS source and then measured time to process it by
Context.compileReader in the interpreter mode.
Under JDK 1.3 the difference is less then 1%, but still using the explicit
string buffer saves memory. When converting StringBuffer to String Sun JDK
shares the internal char array in StringBuffer with new String, but in the
Parser case typically the capacity of this buffer is bigger then the actual
string length, so this unused space in source strings is wasted in the
interpreter mode that keeps these strings in InterpreterData.
Regards, Igor
========
I implemented that member expression as function name syntactic sugar to
support scripts using this MS extension. This is only available when
Context.hasFeature(Context.FEATURE_MEMBER_EXPR_AS_FUNCTION_NAME)
returns true to allow the deviation from the standard only when required.
The patch effectively transforms 'function <memberExpr>(...)...' to
'<memberExpr> = function(...)...' when <memberExpr> is not simple
identifier. I am not sure that MS implementation does exactly this
but hopefully it is sufficiently general to cover MS cases.
(The patch assumes that source_change.patch is already applied)
Regards, Igor
I implemented that member expression as function name syntactic sugar to support
scripts using this MS extension. This is only available when
Context.hasFeature(Context.FEATURE_MEMBER_EXPR_AS_FUNCTION_NAME)
returns true to allow the deviation from the standard only when required.
The patch effectively transforms 'function <memberExpr>(...)...' to
'<memberExpr> = function(...)...' when <memberExpr> is not simple identifier.
I am not sure that MS implementation does exactly this but hopefully it is
sufficiently general to cover MS cases.
(The patch assumes that source_change.patch is already applied)
Regards, Igor
Currently omj/TokenStream and omj/optimizer/Optimizer.java both contain
code to convert number value to a wrapper object of smallest size. The
attached patch moves this wrapping to Node constructor to avoid code
duplication and eliminate special treatment of exact integers in
Optimizer.java.
The constant folding code in omj/optimizer/Optimizer.java currently always
replaces x * 1, x - 0 by simply x which does not force the toNumber convertion,
which is visible, for example, via typeof. For example, when running at
optimization level 2, the following
function f() {
return "0" * 1;
}
print(typeof(f()));
prints "string" instead of expected "number".
The const_fold.patch fixes this via replacing x*1 by (+x) to force number convertion.
It assumes that the patch with number wrapping changes is in place.
convert number value to a wrapper object of smallest size. The attached patch
moves this wrapping to Node constructor to avoid code duplication and eliminate
special treatment of exact integers in Optimizer.java.
Currently omj/optimizer/Codegen.java uses special classes ConstantList
and ConstantDude to store the list of static constants in the generated
class. It seems that using a simple double[] array with a constant
counter and checking via "(int)number == number" for constant types not
only eliminates these 2 classes but makes the whole code simple, see
the attached patch.
The patch also modifies nodeIsConvertToObjectOfNumber to return not a
Number, but the number node itself that is used to extract double
value directly via Node.getDouble() call. I changed it to allow to
store values of number literals in nodes without using wrapper object.
Replacing usage of ShallowNodeIterator to loop throw node children by
explicit calls to Node.getFirstChild()/ Node.getNextSibling()) with
comments when the node children list is modified while iterating
through it.
It avoids creation of ShallowNodeIterator objects and eliminates the
need to have ShallowNodeIterator class.
Currently Rhino source has quite a few places with code like (String)node.getDatum()
or ((Number)node.getDatum()).doubleValue(). The patch changes this usage to call
node.getString() or node.getDouble().
It also adds new constructors to Node to accept int or double values in addition to
Object datum to replace new Node(token, new Integer(x)) by Node(token, x) etc. It
may allow in future not to create a wrapper object for int or double datum to speed
up parsing.
Currently in the interpreter mode all number literals are stored in
InterpreterData.itsICode as an index to InterpreterData.itsNumberTable
which holds the actual value.
For integers that fit 2 or 4 bytes this is an overkill and the attached
patch stores integers in InterpreterData.itsICode inline after special
TokenStream.INTNUMBER or TokenStream.SHORTNUMBERS tokens.
The changes made benchmarks to run 1.5% faster. It also saves memory
because InterpreterData.itsNumberTable is allocated only for non-integers
that present only in a small number of scripts.
In principle, it may be possible to store all numbers inline as well, but
unfortunately re-assembling of 8 bytes from InterpreterData.itsICode array
into double is rather slow operation and is not worth the hassles.
Regards, Igor
Hi, Norris!
Currently ScriptableObject.put does not check lastAccess cache during its search for
slots. When I added this check (see the attached patch) it speeded up the benchmark
suite by about 1.5% and in particular for setProp_bench.js the win was about 8%.
I think that even on multiprocessor machines it would not introduces any additional
issues like accessing the old value in the processor cache because the put method
accesses existing properties via unsynchronized getSlot, and the check for lastAccess
is on pair with that.
Trgards, Igor
When handling an Exception the Context tries to get the current script
and line number from the Java Stacktrace. To get the indication of which
entry in the trace might be an ECMA script, the file extension ".js" is
assumed.
For our integration we use the standard extension ".ecma" which collides
with the above assumption. But we don't force this extension, we just
have a convention. We name these files ".ecma" as they are not plain
ECMA but JSP-like ECMA. That is instead of using Java as the programming
language we use ECMA. In this respect they would be ".esp".
Patch fixes issue of not ignoring UNICODE format characters in match
and peek methods, adds explicit assertions checks for code assumptions
and makes handling of ASCII '\r', '\n' and UNICODE U+2028, U+2029 line
ends uniform.
It was rather tricky to fix format character issue and I spend some
time figuring out what TokenStream assumes about LineBuffer that
breaks my initial thoughts on the patch in cases like very long
sequences of format characters that do not fit in the buffer. I
fixed that but it made the code rather unclear so I put explicit
checks for assumptions/preconditions to help with debugging.
I added Context.check flag to turn on/off these checks and
Context.codeBug to throw an exception in case of check violations,
and also modified UintMap to use them instead of the private
flags there.
It would be nice to add some tests about format characters to the test
suite with checks similar to "eval('1 =\u200C= 1') == true" and
"eval('.\u200C1') == 0.1".
Hi, Norris!
I have found few problems with NativeArraj.java.
1. jsSet_length requires that the new length value should be an instance of Number. But according to Ecma 15.4.5.1, item 12-13, an error should be thrown only if ToUint32(length_value) != ToNumber(length_value). Here is a simple test that demonstrates it:
Array(5).length = new Number(1)
It currenly throws an exception.
2. jsSet_length when executing the code marked with "// assume that the representation is sparse" effectively removes all properties with values less then the current length when String is used to represent its value. Note that simply changing lines "if (d == d && d < length) delete(id);" to "if (d == d && d >= longVal) delete(id);" is not good because it would remove properties like "4.5" or "007", the full array index check has to be used instead.
Here is a test case that catches the problem:
var BIG_INDEX = 4294967290;
var a = Array(BIG_INDEX);
a[BIG_INDEX - 1] = 'a';
a[BIG_INDEX - 10000] = 'b';
a[BIG_INDEX - 0.5] = 'c';
a.length = BIG_INDEX - 5000;
var s = '';
for (var i in a) s += a[i];
print('s="'+s+'"');
this should print s='cb' (or 'bc': EcmaScript does not fix the order), but currently it gives s=''.
3. There are race conditions in jsSet_length and getIds.
The first contains:
if (hasElem(this, i))
ScriptRuntime.delete(this, new Long(i));
which would lead to call to delete in the Array prototype if 2 threads would invoke this code. Simply calling ScriptableObject.delete without any checks for existence is enough here.
getIds assumes that the count of present elements in the dense array does not change, which is not true when another thread deletes elements from dense.
The attached patch fixes these issues.
Regards, Igor
pschwartau%netscape.com
brendan%mozilla.org
seawood%netscape.com
rginda%netscape.com
mike+mozilla%meer.net
alecf%netscape.com
jst%netscape.com
dbradley%netscape.com
jaggernaut%netscape.com
cls%seawood.org
dougt%netscape.com
nboyd%atg.com
jband%netscape.com
bryner%netscape.com
dbaron%fas.harvard.edu
gerv%gerv.net
shaver%mozilla.org
timeless%mac.com
scc%mozilla.org
rogerl%netscape.com