While debugging PR 28274 I came across the following problem:
If I apply the second hunk of the patch below, the C++ frontend
crashes on the following code:
======================
void foo();
void foo() {}
======================
<built-in>:0: note: OLD DECL2 'struct std::type_info'
zsh: 25734 segmentation fault cc1plus -quiet ~/file.cc
If I also apply the first hunk, erverything works as expected:
<built-in>:0: note: OLD DECL1 'struct std::type_info'
<built-in>:0: note: OLD DECL2 'struct std::type_info'
file.cc: In function 'void foo()':
file.cc:2: note: OLD DECL1 'void foo()'
file.cc:2: note: OLD DECL2 'void foo()'
file.cc: At global scope:
file.cc:1: note: OLD DECL1 'struct __cxxabiv1::__fundamental_type_info'
file.cc:1: note: OLD DECL2 'struct __cxxabiv1::__fundamental_type_info'
There seems to be a problem with garbage collection somewhere.
Instead of adding the first hunk, commenting out the call
ggc_free (newdecl);
at the end of duplicate_decls (decl.c) makes things work, too.
Happens on mainline, 4.1 branch, and 4.0 branch
(x86_64-unknown-linux-gnu with checking enabled).
Might be related to PR 27100.
===================================================================
--- gcc/gcc/cp/name-lookup.c (revision 115789)
+++ gcc/gcc/cp/name-lookup.c (working copy)
@@ -720,7 +720,9 @@ pushdecl_maybe_friend (tree x, bool is_friend)
}
else
{
- tree olddecl = duplicate_decls (x, t, is_friend);
+ tree olddecl;
+ inform ("OLD DECL1 %qD", t);
+ olddecl = duplicate_decls (x, t, is_friend);
/* If the redeclaration failed, we can stop at this
point. */
@@ -729,6 +731,8 @@ pushdecl_maybe_friend (tree x, bool is_friend)
if (olddecl)
{
+ inform ("OLD DECL2 %qD", t);
+
if (TREE_CODE (t) == TYPE_DECL)
SET_IDENTIFIER_TYPE_VALUE (name, TREE_TYPE (t));
===================================================================
--
Summary: ICE after duplicate_decls
Product: gcc
Version: 4.2.0
Status: UNCONFIRMED
Keywords: ice-on-valid-code, GC
Severity: normal
Priority: P3
Component: c++
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: reichelt at gcc dot gnu dot org
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28525
