[svn.haxx.se] · SVN Dev · SVN Users · SVN Org · TSVN Dev · TSVN Users · Subclipse Dev · Subclipse Users · this month's index

Re: Windows buildbot issues

From: Paul Burba <ptburba_at_gmail.com>
Date: Mon, 5 Jan 2009 14:23:42 -0500

On Sat, Dec 27, 2008 at 12:47 PM, D.J. Heap <djheap_at_gmail.com> wrote:
> On Sat, Dec 27, 2008 at 4:37 AM, Bert Huijben <bert_at_vmoo.com> wrote:
>> Hi D.J.,
>> The windows subversion buildbot seems to have some errors while executing
>> some/most svn.exe commands from python.

Hi Bert,

Are you seeing this error with actual svn.exe commands or only when
attempting to run a Python script as a command? Are you seeing the
problem yourself when running the tests? If not, could you try
running them in parallel and see if it occurs then?

>> It says '%1' is not a valid
>> executable, where '%1' should normally be filled with the executable name
>> (probably this is a bug in python). The test then says it succeeded,
>> without
>> any real work.
>> I would guess that we have a python script problem or that one of the DLLs
>> used by svn.exe is locked. Could you please look at the buildbot and try
>> to
>> find what causes this problem?

I'm able to replicate this problem with trunk_at_35030 but *only* when
running the tests in parallel and *only* when the offending 'command'
passed to svntest/main.py:open_pipe() is actually a Python script
rather than an executable. Now AFAICT the only time
main.py:open_pipe() is passed a Python script as the 'executable' is
when running the tests in parallel, which would explain why I only see
this when running the tests in parallel.

Since Arfrever's r34863 the test suite use subprocess.Popen() rather
than Popen3() in main.py:open_pipe(), if Python >=2.4, so maybe there
is a bug with subprocess.Popen() on Windows. Or maybe all the
non-Windows buildbots are using Python < 2.4 and it is simply invalid
to pass a Python script as the the executable to subprocess.Popen() --
Arfrever, what is your understanding of this? As you are the author
of r34863 I thought you might have some insight...

The attached patch fixes the problem for me on Windows, but before
committing it I'd like to know whether this really is a Python bug, a
bug in the test suite, or simply some oddity in my environment.


P.S. FWIW I was using ActiveState Python 2.5.2, upgraded to 2.6.1, and
still get the same problem.

> Is it just the buildbot or anyone who runs the Windows tests? I will look
> at it when I have a chance -- the holiday week is not good for me, though,
> so it may be a few days.
> DJ


Received on 2009-01-05 20:24:09 CET

This is an archived mail posted to the Subversion Dev mailing list.