This is the mail archive of the
gcc-bugs@gcc.gnu.org
mailing list for the GCC project.
Re: [Bug middle-end/64498] [5 Regression] Cannot build Firefox with LTO: ICE in substitute_and_fold_dom_walker::before_dom_children
- From: Jan Hubicka <hubicka at ucw dot cz>
- To: "marxin at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>
- Cc: gcc-bugs at gcc dot gnu dot org, jakub at redhat dot com
- Date: Mon, 12 Jan 2015 10:01:35 +0100
- Subject: Re: [Bug middle-end/64498] [5 Regression] Cannot build Firefox with LTO: ICE in substitute_and_fold_dom_walker::before_dom_children
- Authentication-results: sourceware.org; auth=none
- References: <bug-64498-4 at http dot gcc dot gnu dot org/bugzilla/> <bug-64498-4-Q8wx2Th2mT at http dot gcc dot gnu dot org/bugzilla/>
> Problem is hidden in flag_var_tracking_assignments, which is set to false.
> Thus, MAY_HAVE_DEBUG_INSNS is set to false. Unfortunately, Firefox contains
> couple of third party libraries that have -g as a default option. So that DCE
> after a statements removal does not remove debug statements that use result of
> a statement and verifier fails.
flag_var_tracking_assignments is marked as Optimization in common.opt while rest
of debug machinery is not Optimization. This means that we try to preserve assignments
in functions that was compiled with -g at compile time, but then we never consume
it....
Shall we drop flag_var_tracking_assignments to 0 and remove the debug
statements after streaming in in these cases? Also inliner probably needs to
remove the statements when it inlines from function with
flag_var_tracking_assignments to one with !flag_var_tracking_assignments
Honza