[Feature #10602] Add new API rb_profile_thread_frames()

Add a new API rb_profile_thread_frames(), which is essentialy a
per-thread version of rb_profile_frames().

While the original rb_profile_frames() always returns results about the
current active thread obtained by GET_EC(), this new API takes a Thread
to be profiled as an argument.

This should come in handy when profiling I/O-bound programs such as
webapps, since this new API allows us to learn about Threads performing
I/O (which do not have the GVL).

Profiling worker threads (such as Sidekiq workers) may be another
application.

Implements [Feature #10602]

Co-authored-by: Mike Perham <mike@perham.net>

ext/-test-/debug/profile_frames.c

@@ -37,8 +37,29 @@ profile_frames(VALUE self, VALUE start_v, VALUE num_v)
     return result;
 }
 
+static VALUE
+profile_thread_frames(VALUE self, VALUE thread, VALUE start_v, VALUE num_v)
+{
+    int i, collected_size;
+    int start = NUM2INT(start_v);
+    int buff_size = NUM2INT(num_v);
+    VALUE buff[MAX_BUF_SIZE];
+    int lines[MAX_BUF_SIZE];
+    VALUE result = rb_ary_new();
+
+    if (buff_size > MAX_BUF_SIZE) rb_raise(rb_eRuntimeError, "too long buff_size");
+
+    collected_size = rb_profile_thread_frames(thread, start, buff_size, buff, lines);
+    for (i=0; i<collected_size; i++) {
+        rb_ary_push(result, rb_profile_frame_full_label(buff[i]));
+    }
+
+    return result;
+}
+
 void
 Init_profile_frames(VALUE klass)
 {
     rb_define_module_function(klass, "profile_frames", profile_frames, 2);
+    rb_define_module_function(klass, "profile_thread_frames", profile_thread_frames, 3);
 }

include/ruby/debug.h

@@ -50,6 +50,25 @@ RBIMPL_ATTR_NONNULL((3))
  */
 int rb_profile_frames(int start, int limit, VALUE *buff, int *lines);
 
+/**
+ * Queries mysterious "frame"s of the given range.
+ *
+ * A per-thread version of rb_profile_frames().
+ * Arguments and return values are the same with rb_profile_frames() with the
+ * exception of the first argument _thread_, which accepts the Thread to be
+ * profiled/queried.
+ *
+ * @param[in]   thread The Ruby Thread to be profiled.
+ * @param[in]   start  Start position (0 means the topmost).
+ * @param[in]   limit  Number objects of `buff`.
+ * @param[out]  buff   Return buffer.
+ * @param[out]  lines  Return buffer.
+ * @return      Number of objects filled into `buff`.
+ * @post        `buff` is filled with backtrace pointers.
+ * @post        `lines` is filled with `__LINE__` of each backtraces.
+ */
+int rb_profile_thread_frames(VALUE thread, int start, int limit, VALUE *buff, int *lines);
+
 /**
  * Queries the path of the passed backtrace.
  *

test/-ext-/debug/test_profile_frames.rb

@@ -39,6 +39,20 @@ class SampleClassForTestProfileFrames
   end
 end
 
+class SampleClassForTestProfileThreadFrames
+  def initialize(mutex)
+    @mutex = mutex
+  end
+
+  def foo(block)
+    bar(block)
+  end
+
+  def bar(block)
+    block.call
+  end
+end
+
 class TestProfileFrames < Test::Unit::TestCase
   def test_profile_frames
     obj, frames = Fiber.new{
@@ -139,6 +153,39 @@ class TestProfileFrames < Test::Unit::TestCase
     }
   end
 
+  def test_profile_thread_frames
+    mutex = Mutex.new
+    th = Thread.new do
+      mutex.lock
+      Thread.stop
+      SampleClassForTestProfileThreadFrames.new(mutex).foo(lambda { mutex.unlock; loop { sleep(1) } } )
+    end
+
+    # ensure execution has reached SampleClassForTestProfileThreadFrames#bar before running profile_thread_frames
+    loop { break if th.status == "sleep"; sleep 0.1 }
+    th.run
+    mutex.lock # wait until SampleClassForTestProfileThreadFrames#bar has been called
+
+    frames = Bug::Debug.profile_thread_frames(th, 0, 10)
+
+    full_labels = [
+      "Kernel#sleep",
+      "TestProfileFrames#test_profile_thread_frames",
+      "Kernel#loop",
+      "TestProfileFrames#test_profile_thread_frames",
+      "SampleClassForTestProfileThreadFrames#bar",
+      "SampleClassForTestProfileThreadFrames#foo",
+      "TestProfileFrames#test_profile_thread_frames",
+    ]
+
+    frames.each.with_index do |frame, i|
+      assert_equal(full_labels[i], frame)
+    end
+
+    th.kill
+  end
+
+
   def test_matches_backtrace_locations_main_thread
     assert_equal(Thread.current, Thread.main)
 

vm_backtrace.c

@@ -1584,11 +1584,10 @@ 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)
+static int
+thread_profile_frames(rb_execution_context_t *ec, 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_control_frame_t *top = cfp;
     const rb_callable_method_entry_t *cme;
@@ -1650,6 +1649,20 @@ rb_profile_frames(int start, int limit, VALUE *buff, int *lines)
     return i;
 }
 
+int
+rb_profile_frames(int start, int limit, VALUE *buff, int *lines)
+{
+    rb_execution_context_t *ec = GET_EC();
+    return thread_profile_frames(ec, start, limit, buff, lines);
+}
+
+int
+rb_profile_thread_frames(VALUE thread, int start, int limit, VALUE *buff, int *lines)
+{
+    rb_thread_t *th = rb_thread_ptr(thread);
+    return thread_profile_frames(th->ec, start, limit, buff, lines);
+}
+
 static const rb_iseq_t *
 frame2iseq(VALUE frame)
 {