[Bug c++/12491] [3.3/3.4 Regression] [eh] Destructor fails to compile when optimizations (inlining) are enabled

["ebotcazou at gcc dot gnu dot org" <gcc-bugzilla@gcc.gnu.org>]

------- Additional Comments From ebotcazou at gcc dot gnu dot org  2003-12-06 09:18 -------
Thanks for the clarification. I dug deeper and I'm now convinced that it's an EH
problem: we have 6 EH regions, laid out as follows:

   fixup  cleanup
     2  --> 1
     |  \ 
     |   \
     4  --> 3
     |  \
     |   \
     6  --> 5

where --> stands for 'peer' and non-horizontal edges stand for 'inner'/'outer'.

The problem is that the real_region of R4 is R2, which is also a fixup region
and thus is eliminated.

The following patch cures the ICE but I'd like confirmation by an EH specialist
that it is sane:

Index: except.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/except.c,v
retrieving revision 1.254
diff -u -p -r1.254 except.c
--- except.c	14 Nov 2003 10:44:08 -0000	1.254
+++ except.c	6 Dec 2003 09:05:00 -0000
@@ -937,8 +937,13 @@ remove_fixup_regions (void)
 	&& (fixup = cfun->eh->region_array[INTVAL (XEXP (note, 0))])
 	&& fixup->type == ERT_FIXUP)
       {
-	if (fixup->u.fixup.real_region)
-	  XEXP (note, 0) = GEN_INT (fixup->u.fixup.real_region->region_number);
+	/* Walk up the chain of fixup regions.  */
+        do {
+	  fixup = fixup->u.fixup.real_region;
+	} while (fixup && fixup->type == ERT_FIXUP);
+
+	if (fixup)
+	  XEXP (note, 0) = GEN_INT (fixup->region_number);
 	else
 	  remove_note (insn, note);
       }

[Mike, I CCed you because you're the author of except.c]


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mrs at apple dot com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=12491