[clang] ea22fdd - [Clang][DebugInfo] Cease turning instruction-referencing off by default

[David Blaikie via cfe-commits]

On Wed, Dec 22, 2021 at 11:47 AM Jeremy Morse <jeremy.morse.llvm at gmail.com>
wrote:

> On Wed, Dec 22, 2021 at 4:34 PM David Blaikie via cfe-commits
> <cfe-commits at lists.llvm.org> wrote:
> > Is there a way to turn this off now, if it's broken for some user/use
> case? I guess using an -mllvm flag, instead of an -Xclang flag?
>
> Yup, that should be achieved by "-mllvm
> -experimental-debug-variable-locations=false", there's a test for that
> in llvm/test/DebugInfo/X86/instr-ref-flag.ll.
>

Fair enough - probably would've been good to have that flag in this commit
message (hopefully some folks can find it here in the email thread) - and I
think I probably would've left the clang flag in (& changed its default) to
"ship" the feature (though, as you said on the other thread - LTO wouldn't
be affected by that flag flip, so would also have to make the change down
in LLVM too - sort of nice that they happened separately, really, to create
an incremental rollout)

Ah well - sounds alright, thanks for the details!


>
> > Generally I think for a significant change like this the process would
> be to add it under a flag, off by default, encourage people to test it by
> opting in via the flag, eventually change the default for the flag (letting
> people know they can opt-out using the flag if they see problems & to
> report them to you so they can be fixed & hopefully they can stop opting
> out) & then remove the flag when it's stable. Reverting a patch to disable
> a newly-on-by-default feature is a bit heavyweight and it can be nice to
> have a flag to control it while it's getting more exposure/experience.
>
> Indeed, I'd hoped this was largely achieved with
> https://lists.llvm.org/pipermail/llvm-dev/2021-July/151965.html , I
> guess what's missing is better communicating in how to opt out now,
> alas. It's unfortunate that people bisecting back are going to find
> this commit; I couldn't think of another way of rectifying things
> without pushing up several reverts and the re-applies at the same time
> though.
>
> --
> Thanks,
> Jeremy
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20211222/cb463451/attachment-0001.html>