This fixes the ICE seen in PR60060 which stems from us asking to
output debug-info for a function-scope global twice - once via
emitting the function and its BLOCK-associated vars and once
via the debug-hook through lto_write_globals ->
emit_debug_global_declarations.
The list of "global" variables does not agree with those from
the frontends (that function-scope global is not in GFortrans list).
Thus the following simply avoids the mess by only ever emitting
debug-info for non-function scope globals here. As
wrapup_global_declarations looks like a complete no-op for LTO
(DECL_DEFER_OUTPUT is never 1, it's not even streamed ...) I
chose to inline emit_debug_global_declarations. That whole area
asks for a FE-middle-end interface sanitization ...
LTO bootstrapped / tested on x86_64-unknown-linux-gnu.
Does that look ok?
Thanks,
Richard.
2014-02-11 Richard Biener <rguent...@suse.de>
PR lto/60060
* lto-lang.c (lto_write_globals): Do not call
wrapup_global_declarations or emit_debug_global_declarations
but emit debug info for non-function scope variables
directly.
Index: gcc/lto/lto-lang.c
===================================================================
*** gcc/lto/lto-lang.c (revision 207655)
--- gcc/lto/lto-lang.c (working copy)
*************** lto_write_globals (void)
*** 1082,1098 ****
if (flag_wpa)
return;
! /* Record the global variables. */
! vec<tree> lto_global_var_decls = vNULL;
varpool_node *vnode;
FOR_EACH_DEFINED_VARIABLE (vnode)
! lto_global_var_decls.safe_push (vnode->decl);
!
! tree *vec = lto_global_var_decls.address ();
! int len = lto_global_var_decls.length ();
! wrapup_global_declarations (vec, len);
! emit_debug_global_declarations (vec, len);
! lto_global_var_decls.release ();
}
static tree
--- 1082,1092 ----
if (flag_wpa)
return;
! /* Output debug info for global variables. */
varpool_node *vnode;
FOR_EACH_DEFINED_VARIABLE (vnode)
! if (!decl_function_context (vnode->decl))
! debug_hooks->global_decl (vnode->decl);
}
static tree
