Spurious test failures in open_write_stream

[Jelmer Vernooij]

On 08/29/2011 05:03 PM, John Arbash Meinel wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> On 8/29/2011 1:45 AM, Jelmer Vernooij wrote:
>> Hi,
>>
>> While trying to build bzr 2.4.0 on hardy, I am running into
>> spurious test failures in open_write_stream when being called from
>> _write_index.
>>
>> I've seen this particular exception being raised from various
>> different tests. For some reason I can only reproduce the problem
>> on the buildds an not on my local machine.
>>
>> Especially the fact that the failures are spurious and not locally
>> reproducible make it hard to fix this.
>>
>> Has anybody seen these errors before? This is the traceback from
>> one of the errors:
>>
>> ERROR:
>> bzrlib.tests.blackbox.test_version_info.TestVersionInfo.test_custom_implies_all
> ...
>
>>
> File
>> "/build/buildd/bzr-2.4.0/build/lib.linux-i686-2.5/bzrlib/transport/local.py",
>>
>>
> line 331, in open_write_stream
>> handle = osutils.open_file(abspath, 'wb') IOError: [Errno 21] Is a
>> directory
>>
>> Cheers,
>>
>> Jelmer
>>
>>
> I haven't seen this before myself. Getting EISDIR is pretty strange
> here. This is when it is writing the index files out, so the files
> should always have a '.iix', '.tix' etc, even if 'name' somehow became
> an empty string.
>
> Now, if this was ENOTDIR, that could be something else entirely, but
> EISDIR is just really strange.
Yeah, my thoughts exactly. At least that means I'm not missing something 
obvious here.

It seems that not running with --parallel stops the spurious test 
failures, so that means this issue is no longer blocking me.

It's still a mystery to me why this would break when running the tests 
in parallel though, since we're running all tests in separate 
directories (IIUC).

Cheers,

Jelmer