Skip to content
This repository

Test failure in Travis #120

Open
kmaehashi opened this Issue · 15 comments

2 participants

Kenichi Maehashi Shuzo Kashihara
Kenichi Maehashi
Owner

I've updated the travis config in 1fa7a95 and this somewhat improved the situation:

Shuzo Kashihara
Owner

:+1:

Shuzo Kashihara
Owner

It caused mixable_stat's constructor does not initialize member variable e_ and n_ in mixable_stat.cpp. And that mixable_stat_test is not appropriate for unit testing.

Shuzo Kashihara
Owner

Previous commit(member initialization) 1854d1c that could not fix when built by gcc in Travis.

Kenichi Maehashi
Owner

I wish if we could ssh & gdb into there :(

Just FYI, Travis is running in 32-bit box. Migration to 64-bit environment is planned, in late September or early October 2012 as they described in here.

Shuzo Kashihara
Owner

I didn't know that is in 32-bit because Jubatus uses 64 bit __sync_fetch_and_add GCC Atomic-Bulitins.

I added labels to pending.
I think better to disable Travis, or to use Travis without gcc(clang based) for a short while. How do you think?

Kenichi Maehashi
Owner

Oh, I didn't know that __sync_fetch_and_add doesn't work in 32-bit. I think using Travis without gcc would be a best option. I'll commit the changed config.

Shuzo Kashihara
Owner

Current travis fails by jubatus-mpio configuration with clang:
https://travis-ci.org/jubatus/jubatus/builds/3750405
that says

checking whether the C compiler works... no
78configure: error: in `/home/travis/builds/jubatus/jubatus/jubatus-mpio':
79configure: error: C compiler cannot create executables

I got reappearance of failure:
https://gist.github.com/4537539

mpio and msgpack-rpc use -O4 which defined in configure.in.
The problem is that clang driver and ld command cannot link LLVM bitcode.
We can avoid choosing like -O3 or -O2 option instead of -O4.

man of Clang says:

On supported platforms, -O4 enables link-time optimization; object files are stored in the LLVM bitcode file format and whole program optimization is done at link time.

That is the feature, and clang needs llvm gold-plugin when compile -O4.
http://lists.cs.uiuc.edu/pipermail/llvmbugs/2011-May/018110.html
http://lists.cs.uiuc.edu/pipermail/llvmbugs/2011-May/018111.html

http://llvm.org/docs/GoldPlugin.html

Shuzo Kashihara
Owner

I fixed mpio and msgpack-rpc for clang compilation to success, but travis fails in 2b0f8de build 188 , so currently make travis disabled.
mpio/msgpack-rpc fail with Clang 3.0 in my Debian 7.0 wheezy x86_64 environment, but I got successful with Clang 3.3(trunk 172618)/LLVMgold.

I suspect about behavior of Clang 3.1 or earlier.

Shuzo Kashihara suma referenced this issue in jubatus/jubatus-mpio
Open

msgpack-rpc tests fail mpio compiled with clang #3

Shuzo Kashihara
Owner
suma commented

Test failure on 32bit maybe solved by http://d.hatena.ne.jp/y_tag/20130310/p1

Shuzo Kashihara suma referenced this issue from a commit in gintenlabo/jubatus
SubaruG gintenlabo Fix issue #349 f3d1196
Shuzo Kashihara suma referenced this issue from a commit
Shuzo Kashihara suma Update travis config: refs #120 6741488
Kenichi Maehashi
Owner

Is it now OK to close this issue?

Maybe we can enable gcc and master branch test. https://github.com/jubatus/jubatus/blob/master/.travis.yml#L59

Shuzo Kashihara
Owner

I close.
Yes > gcc for master.

Shuzo Kashihara suma closed this
Kenichi Maehashi
Owner

Yes > gcc for master.

Ok, I'll re-enable it in master branch. (reopened & changed assignee)

Kenichi Maehashi
Owner

I re-enabled tests against gcc, however it fails.
https://travis-ci.org/jubatus/jubatus/jobs/15748101#L2487

Kenichi Maehashi
Owner

Does anyone interested in investigating this issue? ;)
If not, I'll exclude gcc test again (gcc tests are already covered by our Jenkins, so no worry about it)

Kenichi Maehashi
Owner

I've disabled gcc test in Travis; changed milestone to "Pending".

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Something went wrong with that request. Please try again.