------- Comment #11 from jakub at gcc dot gnu dot org 2010-04-29 20:29 -------
So, the first difference that happens is that with -g the extern var is removed
from BLOCK_VARs, with -g it is kept there. This happens in
remove_unused_scope_block_p. Next during inlining in remap_decls
can_be_nonlocal decides the extern var can't be non-local (remapped_type for
int is NULL at that point - int hasn't been remapped yet, so it returns false).
This is something I don't understand at all - for variably_modified_type_p
(TREE_TYPE (decl), id->src_fn) perhaps it makes sense, but aren't all other
types just remapped to itself?
Thus, remap_decls decides to remap_decl it (in -g case only) and that actually
means the remapped var makes it into REFERENCED_VARS.
One question is whether auto_var_in_fn_p should return true or false for
DECL_EXTERNAL vars with DECL_CONTEXT being the FUNCTION_DECL. Apparently in
many places in tree-inline.c this predicate is used to tell whether the decl
should be remapped. So, either we need something like H.J.'s patch, but with
&& !DECL_EXTERNAL (var) (for VAR_DECL only), or e.g. tree-tailcall.c should
add a && !DECL_EXTERNAL (var) or && !is_global_var (var) check.
--
jakub at gcc dot gnu dot org changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |hubicka at gcc dot gnu dot
| |org
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43942
