http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57184
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |ASSIGNED
AssignedTo|unassigned at gcc dot |jakub at gcc dot gnu.org
|gnu.org |
--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-05-06
11:47:58 UTC ---
Created attachment 30039
--> http://gcc.gnu.org/bugzilla/attachment.cgi?id=30039
gcc49-pr57184.patch
So, I see 3 different ways to fix this:
1) something like:
--- gcc/dwarf2out.c.jj2013-04-26 08:49:53.000000000 +0200
+++ gcc/dwarf2out.c2013-05-06 13:33:20.077898452 +0200
@@ -14933,6 +14933,9 @@ reference_to_unused (tree * tp, int * wa
}
else if (TREE_CODE (*tp) == STRING_CST && !TREE_ASM_WRITTEN (*tp))
return *tp;
+ /* expand_expr can't handle COMPOUND_LITERAL_EXPRs. */
+ else if (TREE_CODE (*tp) == COMPOUND_LITERAL_EXPR)
+ return *tp;
return NULL_TREE;
}
where we'd just give up on COMPOUND_LITERAL_EXPRs in the initializers for debug
info purposes, unfortunately we generate worse debug info in that case
2) rtl_for_decl_init could notice this case, unshare_expr the expression and
tweak it, but that means extra 2 walk_tree calls on the initializer before
expand_expr, with two different callbacks
3) just allow expand_expr on ADDR_EXPR of COMPOUND_LITERAL_EXPR if
EXPAND_INITIALIZER
Attached patch implements 3), which is quite simple and allows to have good
debug info in this case too.
