On 12/13/2017 07:12 AM, Jakub Jelinek wrote:
> Hi!
>
> PR83391/PR83396 failed because debug bind stmts were put before labels.
> Alex said that is undesirable, and that right now we want to allow
> just debug begin stmt markers before or intermixed with labels.
>
> This patch ensures that through verification, which is what defines
> what is and isn't valid GIMPLE. If we ever reconsider it, either allow
> further stmts or disallow even debug begin stmt markers, we can easily also
> tweak the verifier. The patch has been successfully bootstrapped/regtested as
> part of the:
>
> https://gcc.gnu.org/ml/gcc-patches/2017-12/msg00811.html
>
> https://gcc.gnu.org/ml/gcc-patches/2017-12/msg00808.html
>
> https://gcc.gnu.org/bugzilla/attachment.cgi?id=42861
>
> https://gcc.gnu.org/bugzilla/attachment.cgi?id=42866
>
> https://gcc.gnu.org/ml/gcc-patches/2017-12/msg00794.html
>
>
> patchset, without the msg00811.html patch it of course doesn't survive
> bootstrap, as we insert debug bind stmts before labels in those cases.
>
> Ok for trunk once the msg00811.html patch or something similar is committed?
>
> 2017-12-13 Jakub Jelinek <ja...@redhat.com>
>
> * tree-cfg.c (verify_gimple_in_cfg): Verify no non-label stmts
> with the exception of debug begin stmt markers appear before
> labels.
OK.
jeff
