Bug 728604 - many self checks fail on PPC
Summary: many self checks fail on PPC
Alias: None
Product: Fedora
Classification: Fedora
Component: llvm
Version: 16
Hardware: powerpc
OS: Unspecified
Target Milestone: ---
Assignee: Michel Alexandre Salim
QA Contact: Fedora Extras Quality Assurance
Depends On:
Blocks: F16Betappc
TreeView+ depends on / blocked
Reported: 2011-08-05 18:44 UTC by Karsten Hopp
Modified: 2012-03-14 13:10 UTC (History)
4 users (show)

Fixed In Version:
Doc Type: Bug Fix
Doc Text:
Clone Of:
Last Closed: 2011-10-27 16:16:08 UTC
Type: ---

Attachments (Terms of Use)

Description Karsten Hopp 2011-08-05 18:44:05 UTC
Description of problem:
1 warning generated.
fatal error: malformed or corrupted PCH file: 'declaration ID out-of-range for AST file'
FileCheck error: '-' is empty.
1 error generated.
  Expected Passes    : 2860
  Expected Failures  : 22
  Unexpected Failures: 36

Version-Release number of selected component (if applicable):

How reproducible:

Steps to Reproduce:
1. ppc-koji build dist-f16 --scratch llvm-2.9-2.fc16.src.rpm
Actual results:

Additional info:
This is currently blocking F16 composes on PPC

Comment 1 Michel Alexandre Salim 2011-08-06 09:18:42 UTC
grep ^FAIL: build.log shows that most of the failures are C++-related; however, some of them are failed C and ObjC test cases too.

Having no PPC machine to test on, I'm not sure which to choose of the two solutions:
- disable clang on PPC
- just disable its test suite

Do the PPC maintainers have a preference? Thanks.

Comment 2 Karsten Hopp 2011-08-06 09:59:16 UTC
We're hoping to get a PPC machine for Fedora packagers exactly for this type of problems later this year. Until then I'd prefer to have the testsuite disabled. Disabling clang might cause lots of dependency errors in other packages.

Comment 3 Michel Alexandre Salim 2011-08-06 10:26:32 UTC
Hm, you're right, flterm and pure have build dependencies on clang (I package the latter, and it probably should be fixed -- I tried switching to clang for building before and it didn't quite work out).

Do you need the test suite disabled on ppc64 as well? From the changelog it looks like you enabled them. I don't see a 'ppc64-koji' launcher so I can't test that build target.

Comment 4 Karsten Hopp 2011-08-06 10:41:51 UTC
ppc-koji starts builds on ppc and ppc64 similar to 'koji' which starts builds on x86 and x86_64.
Please disable the testsuite on ppc and ppc64.

Comment 5 Michel Alexandre Salim 2011-08-06 11:02:19 UTC
Right, I noticed after sending my message that it's doing both builds. Still awaiting scratch build completion; if it's successful I'll commit and close this bug -- I'm assuming the PPC compose will rebuild any updated packages automatically?

It's probably too late to report this to upstream - they don't support older versions at all - but I'll make sure to try the first 3.0 pre-release on PPC as well and report the results before the final release goes out.

Comment 6 Michel Alexandre Salim 2011-08-06 15:11:46 UTC
LLVM tests fail on ppc64 as well, so I'm disabling that too. Scratch build now finishes fine:


and the changes have been pushed. Setting this to ON_QA; please close when the ppc* compose succeeds.

Comment 7 Dan Horák 2011-09-07 09:41:46 UTC
Wouldn't be it better to always run the test, but only don't fail the build on platforms where failuers are expected like on ppc or s390? E.g. use something like

make check \
%ifarch ppc s390
|| :

s390(x) seems to have similar problems like ppc/ppc64. llvm-2.9-0.2.rc1.fc16 contains failed tests, but the build was successful, for llvm-2.9-2.fc16 the build fails because the failing tests.

Comment 8 Karsten Hopp 2011-10-27 16:16:08 UTC
llvm-2.9-4.fc16 built fine on ppc*, closing

Note You need to log in before you can comment on or make changes to this bug.