RFA: Fix PR41906 (segfault in EH lowering)

RFA: Fix PR41906 (segfault in EH lowering)

[Michael Matz]

Hi,

with -fpermissive we accept catch-alls that aren't at the end of a catch 
chain, hence we shouldn't segfault on them.  Formerly we regarded them as 
unreachable (which indeed they are), and it's easy enough to continue to 
do so while lowering the EH constructs, so I don't think we should change 
the behaviour of -fpermissive in this regard.

Regstrapping on x86_64-linux (all langs+Ada) in progress.  Okay if that 
passes?


Ciao,
Michael.
-- 
	PR c++/41906
	* tree-eh.c (lower_catch): Stop after expanding a catch-all.

testsuite/
	* g++.dg/tree-ssa/pr41906.C: New testcase.

Index: tree-eh.c
===================================================================
--- tree-eh.c	(revision 154674)
+++ tree-eh.c	(working copy)
@@ -1666,6 +1666,8 @@ lower_catch (struct leh_state *state, gi
 	  x = gimple_build_goto (out_label);
 	  gimple_seq_add_stmt (&new_seq, x);
 	}
+      if (!c->type_list)
+	break;
     }
 
   gimple_try_set_cleanup (tp, new_seq);
Index: testsuite/g++.dg/tree-ssa/pr41906.C
===================================================================
--- testsuite/g++.dg/tree-ssa/pr41906.C	(revision 0)
+++ testsuite/g++.dg/tree-ssa/pr41906.C	(revision 0)
@@ -0,0 +1,12 @@
+/* { dg-do compile } */
+/* { dg-options "-fpermissive -w" } */
+/* We aren't interested in the warning, but in the ICE.  */
+void foo();
+extern void abort (void);
+
+void bar()
+{
+  try { foo(); }
+  catch (...) {}
+  catch (int) {abort ();}
+}

Re: RFA: Fix PR41906 (segfault in EH lowering)

[Richard Guenther]

On Thu, Nov 26, 2009 at 5:50 PM, Michael Matz <matz@suse.de> wrote:
> Hi,
>
> with -fpermissive we accept catch-alls that aren't at the end of a catch
> chain, hence we shouldn't segfault on them.  Formerly we regarded them as
> unreachable (which indeed they are), and it's easy enough to continue to
> do so while lowering the EH constructs, so I don't think we should change
> the behaviour of -fpermissive in this regard.
>
> Regstrapping on x86_64-linux (all langs+Ada) in progress.  Okay if that
> passes?

Ok.

Thanks,
Richard.

>
> Ciao,
> Michael.
> --
>        PR c++/41906
>        * tree-eh.c (lower_catch): Stop after expanding a catch-all.
>
> testsuite/
>        * g++.dg/tree-ssa/pr41906.C: New testcase.
>
> Index: tree-eh.c
> ===================================================================
> --- tree-eh.c   (revision 154674)
> +++ tree-eh.c   (working copy)
> @@ -1666,6 +1666,8 @@ lower_catch (struct leh_state *state, gi
>          x = gimple_build_goto (out_label);
>          gimple_seq_add_stmt (&new_seq, x);
>        }
> +      if (!c->type_list)
> +       break;
>     }
>
>   gimple_try_set_cleanup (tp, new_seq);
> Index: testsuite/g++.dg/tree-ssa/pr41906.C
> ===================================================================
> --- testsuite/g++.dg/tree-ssa/pr41906.C (revision 0)
> +++ testsuite/g++.dg/tree-ssa/pr41906.C (revision 0)
> @@ -0,0 +1,12 @@
> +/* { dg-do compile } */
> +/* { dg-options "-fpermissive -w" } */
> +/* We aren't interested in the warning, but in the ICE.  */
> +void foo();
> +extern void abort (void);
> +
> +void bar()
> +{
> +  try { foo(); }
> +  catch (...) {}
> +  catch (int) {abort ();}
> +}
>