Massive CodeGen test failures on AMD64

Hi,

I'm currently getting lots of test failures on AMD64. Most of them are
in CodeGen (e.g. indirect-goto.c, bitfield.c, enum.c, ...), some are in
CodeGenObjC, and two - and perhaps these cause all the others - are in
Sema: init.c and static-init.c. In the two Sema cases, I suspect that
the cast from array to int is the culprit, since int is too small to
hold a pointer and I saw some updates to the checking go by on the ML.

But are the CodeGen failures likely consequences of this or has
something different changed?

I can give you all the result files if you want, but they're not exactly
informative. Messages like "expected 1 lines but got 0".

Sebastian

Hi Sebastian,

I think there are multiple issues here. The majority of failures are
likely my fault and due to the switching to the new x86_64 ABI, I
haven't seen the failures because my target triple doesn't default to
a 64-bit one. I'll update the relevant test cases to force a
reasonable triple.

By the way, Daniel Berlin added clang testers to his buildbot. They
aren't triggered on every clang commit, but they are a nice resource
(which I admit I haven't been checking frequently) for seeing where
things go wrong. For example, the majority of failures were introduced
between here:
http://google1.osuosl.org:8011/builders/clang-x86_64-linux/builds/271
and here:
http://google1.osuosl.org:8011/builders/clang-x86_64-linux/builds/272
which includes the commit where I turned on the new x86_64 ABI
handling. See http://google1.osuosl.org:8011/waterfall for the
overview.

Thanks for bringing this to my attention!
- Daniel

Daniel Dunbar wrote:

Hi Sebastian,

I think there are multiple issues here. The majority of failures are
likely my fault and due to the switching to the new x86_64 ABI, I
haven't seen the failures because my target triple doesn't default to
a 64-bit one. I'll update the relevant test cases to force a
reasonable triple.
  

OK, so the Sema failures have nothing to do with it. I'll see what I can
do about those then.

By the way, Daniel Berlin added clang testers to his buildbot. They
aren't triggered on every clang commit, but they are a nice resource
(which I admit I haven't been checking frequently) for seeing where
things go wrong. For example, the majority of failures were introduced
between here:
http://google1.osuosl.org:8011/builders/clang-x86_64-linux/builds/271
and here:
http://google1.osuosl.org:8011/builders/clang-x86_64-linux/builds/272
which includes the commit where I turned on the new x86_64 ABI
handling. See http://google1.osuosl.org:8011/waterfall for the
overview.
  
That's great. Thanks for the info. If you need another linux-x86_64
machine, let me know.

Sebastian