Forum: Ruby-dev [Closed] TracePoint can skip c_return with rb_rescue()

308cbef6e86dfc49cce3b2d4cf42aedc?d=identicon&s=25 unknown (Guest)
on 2014-06-19 14:44
(Received via mailing list)
Issue #9961 has been updated by Koichi Sasada.

Status changed from Open to Closed
% Done changed from 0 to 100

Applied in changeset r46465.

----------
* vm.c (rb_vm_rewind_cfp): add new function to rewind specified cfp
  with invoking RUBY_EVENT_C_RETURN.
  [Bug #9961]
* vm_core.h: ditto.
* eval.c (rb_protect): use it.
* eval.c (rb_rescue2): ditto.
* vm_eval.c (rb_iterate): ditto.
* test/ruby/test_settracefunc.rb: add a test.
* vm_core.h (rb_name_err_mesg_new):

----------------------------------------
Bug #9961: TracePoint can skip c_return with rb_rescue()
https://bugs.ruby-lang.org/issues/9961#change-47292

* Author: Koichi Sasada
* Status: Closed
* Priority: Normal
* Assignee: Koichi Sasada
* Category: core
* Target version: current: 2.2.0
* ruby -v: 2.2
* Backport: 2.0.0: REQUIRED, 2.1: REQUIRED
----------------------------------------
下記のようなテストに失敗します。

```ruby
  def test_rb_rescue
    events = []
    curr_thread = Thread.current
    TracePoint.new(:a_call, :a_return){|tp|
      next if curr_thread != Thread.current
      events << [tp.event, tp.method_id]
    }.enable do
      begin
        -Numeric.new
      rescue => e
        # ignore
      end
    end

    assert_equal [
    [:b_call, :test_rb_rescue],
      [:c_call, :new],
        [:c_call, :initialize],
        [:c_return, :initialize],
      [:c_return, :new],
      [:c_call, :-@],
        [:c_call, :coerce],
          [:c_call, :to_s],
          [:c_return, :to_s],
          [:c_call, :new],
            [:c_call, :initialize],
            [:c_return, :initialize],
          [:c_return, :new],
          [:c_call, :exception],
          [:c_return, :exception],
          [:c_call, :backtrace],
          [:c_return, :backtrace],
        [:c_return, :coerce],            # don't miss it!
        [:c_call, :to_s],
        [:c_return, :to_s],
        [:c_call, :to_s],
        [:c_return, :to_s],
        [:c_call, :new],
          [:c_call, :initialize],
          [:c_return, :initialize],
        [:c_return, :new],
        [:c_call, :exception],
        [:c_return, :exception],
        [:c_call, :backtrace],
        [:c_return, :backtrace],
      [:c_return, :-@],
      [:c_call, :===],
      [:c_return, :===],
    [:b_return, :test_rb_rescue]], events
  end
```

何が起きているかというと、

(1) -Numeric.new で rb_rescue() の後で rb_funcall() した先で例外が起こる
(2) rb_rescue() では、c_return を無視して cfp を切り詰める

ということにより、c_return がスキップしてしまう、という現象となりました。

次のパッチで解決します。

```diff
Index: eval.c
===================================================================
--- eval.c  (revision 46464)
+++ eval.c  (working copy)
@@ -803,7 +803,7 @@ rb_rescue2(VALUE (* b_proc) (ANYARGS), V
   }
     }
     else {
-  th->cfp = cfp; /* restore */
+  rb_vm_rewind_cfp(th, cfp);

   if (state == TAG_RAISE) {
       int handle = FALSE;
@@ -862,7 +862,7 @@ rb_protect(VALUE (* proc) (VALUE), VALUE
   SAVE_ROOT_JMPBUF(th, result = (*proc) (data));
     }
     else {
-  th->cfp = cfp;
+  rb_vm_rewind_cfp(th, cfp);
     }
     MEMCPY(&(th)->root_jmpbuf, &org_jmpbuf, rb_jmpbuf_t, 1);
     th->protect_tag = protect_tag.prev;
Index: vm.c
===================================================================
--- vm.c  (revision 46464)
+++ vm.c  (working copy)
@@ -288,6 +288,23 @@ rb_vm_pop_cfunc_frame(void)
     vm_pop_frame(th);
 }

+void
+rb_vm_rewind_cfp(rb_thread_t *th, rb_control_frame_t *cfp)
+{
+    /* check skipped frame */
+    while (th->cfp != cfp) {
+#if VMDEBUG
+  printf("skipped frame: %s\n", vm_frametype_name(th->cfp));
+#endif
+  if (VM_FRAME_TYPE(th->cfp) != VM_FRAME_MAGIC_CFUNC) {
+      vm_pop_frame(th);
+  }
+  else { /* unlikely path */
+      rb_vm_pop_cfunc_frame();
+  }
+    }
+}
+
 /* obsolete */
 void
 rb_frame_pop(void)
Index: vm_core.h
===================================================================
--- vm_core.h  (revision 46464)
+++ vm_core.h  (working copy)
@@ -901,6 +901,7 @@ VALUE rb_name_err_mesg_new(VALUE obj, VA
 void rb_vm_stack_to_heap(rb_thread_t *th);
 void ruby_thread_init_stack(rb_thread_t *th);
 int rb_vm_control_frame_id_and_class(const rb_control_frame_t *cfp, ID
*idp, VALUE *klassp);
+void rb_vm_rewind_cfp(rb_thread_t *th, rb_control_frame_t *cfp);

 void rb_gc_mark_machine_stack(rb_thread_t *th);

Index: vm_eval.c
===================================================================
--- vm_eval.c  (revision 46464)
+++ vm_eval.c  (working copy)
@@ -1093,18 +1093,7 @@ rb_iterate(VALUE (* it_proc) (VALUE), VA
     th->errinfo = Qnil;
     retval = GET_THROWOBJ_VAL(err);

-    /* check skipped frame */
-    while (th->cfp != cfp) {
-#if VMDEBUG
-        printf("skipped frame: %s\n", vm_frametype_name(th->cfp));
-#endif
-        if (VM_FRAME_TYPE(th->cfp) != VM_FRAME_MAGIC_CFUNC) {
-      vm_pop_frame(th);
-        }
-        else { /* unlikely path */
-      rb_vm_pop_cfunc_frame();
-        }
-    }
+    rb_vm_rewind_cfp(th, cfp);
       }
       else{
     /* SDR(); printf("%p, %p\n", cdfp, escape_dfp); */
```

例によって、2.1, 2.0 でもおきます。

(Numeric.new なんてできるとは知らなかった)
This topic is locked and can not be replied to.