llvm dev - Mar 2015 - [LLVMdev] Self-hosting failure in ARM again

Folks,

It seems we got the same issue with Clang alignment as before:

http://lab.llvm.org:8011/builders/clang-cmake-armv7-a15-selfhost/builds/3025
http://lab.llvm.org:8011/builders/clang-cmake-armv7-a15-selfhost-neon/builds/485
http://lab.llvm.org:8011/builders/clang-cmake-thumbv7-a15-full-sh/builds/118

Commits between 231213 and 231255.

There are a few issues that could have brought it:
 * Seth's changes to Clang (unlikely)
 * Michael's changes to DAGCombiner (likely)
 * David's generic changes in LLVM (unlikely)

I'll be bisecting this like I did earlier, but I'd like to keep you
guys posted of the potential reverts, since this is not jut breaking
tests, but breaking Clang self-hosting (which cost us weeks during the
release).

Ahmed,

You have a better idea what the real problem is. I think maybe we
should bump its priority up again, or this will keep popping every
month.

Did you file a bug?

cheers,
--renato

Hi Renato,

I think it's a different issue this time.

The bots (not only ARM, btw) are getting an FE failure in compiling
Support/CommandLine.cpp.
Since some of David's in this range were to CommandLine.h, they appear to me
to be the more likely culprits. In particular, r231221 touched ~basic_parser(),
which is what we seem to be breaking on.

Apologize in advance in case it IS my DAGCombine change. :-)
  Michael

-----Original Message-----
From: Renato Golin [mailto:renato.golin at linaro.org] 
Sent: Wednesday, March 04, 2015 18:07
To: LLVM Dev; Clang Dev
Cc: Ahmed Bougacha; David Blaikie; Kuperstein, Michael M; Seth Cantrell
Subject: Self-hosting failure in ARM again

Folks,

It seems we got the same issue with Clang alignment as before:

http://lab.llvm.org:8011/builders/clang-cmake-armv7-a15-selfhost/builds/3025
http://lab.llvm.org:8011/builders/clang-cmake-armv7-a15-selfhost-neon/builds/485
http://lab.llvm.org:8011/builders/clang-cmake-thumbv7-a15-full-sh/builds/118

Commits between 231213 and 231255.

There are a few issues that could have brought it:
 * Seth's changes to Clang (unlikely)
 * Michael's changes to DAGCombiner (likely)
 * David's generic changes in LLVM (unlikely)

I'll be bisecting this like I did earlier, but I'd like to keep you guys
posted of the potential reverts, since this is not jut breaking tests, but
breaking Clang self-hosting (which cost us weeks during the release).

Ahmed,

You have a better idea what the real problem is. I think maybe we should bump
its priority up again, or this will keep popping every month.

Did you file a bug?

cheers,
--renato
---------------------------------------------------------------------
Intel Israel (74) Limited

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.

On 4 March 2015 at 16:17, Kuperstein, Michael M
<michael.m.kuperstein at intel.com> wrote:
> The bots (not only ARM, btw) are getting an FE failure in compiling
Support/CommandLine.cpp.
> Since some of David's in this range were to CommandLine.h, they appear
to me to be the more likely culprits. In particular, r231221 touched
~basic_parser(), which is what we seem to be breaking on.
I see, we may have two issues at hand.

> Apologize in advance in case it IS my DAGCombine change. :-)
No worries, that's a Clang issue, or at least it was last time, where
the structures are not aligned correctly.

I'll report back once I have a culprit. :)

cheers,
--renato

On Wed, Mar 4, 2015 at 8:06 AM, Renato Golin <renato.golin at linaro.org>
wrote:
> Folks,
>
> It seems we got the same issue with Clang alignment as before:
>
>
http://lab.llvm.org:8011/builders/clang-cmake-armv7-a15-selfhost/builds/3025
>
http://lab.llvm.org:8011/builders/clang-cmake-armv7-a15-selfhost-neon/builds/485
>
http://lab.llvm.org:8011/builders/clang-cmake-thumbv7-a15-full-sh/builds/118
Hmm, are these actually alignment-related?  They all seem to be the
same assert failing:

    Assertion `(!DD->isTrivial() || DD->hasAttr<DLLExportAttr>())
&&
"Should not emit dtor epilogue for non-exported trivial dtor!"'
failed.

I'm tempted to say the CommandLine changes triggered some clang bug?
> Commits between 231213 and 231255.
>
> There are a few issues that could have brought it:
>  * Seth's changes to Clang (unlikely)
>  * Michael's changes to DAGCombiner (likely)
>  * David's generic changes in LLVM (unlikely)
>
> I'll be bisecting this like I did earlier, but I'd like to keep you
> guys posted of the potential reverts, since this is not jut breaking
> tests, but breaking Clang self-hosting (which cost us weeks during the
> release).
>
> Ahmed,
>
> You have a better idea what the real problem is. I think maybe we
> should bump its priority up again, or this will keep popping every
> month.
>
> Did you file a bug?
Yep, http://llvm.org/bugs/show_bug.cgi?id=22728

Good luck with the bisecting!

-Ahmed
> cheers,
> --renato