David Glasser wrote:
> On 10/17/07, David Glasser <glasser@davidglasser.net> wrote:
>> On 10/17/07, David Glasser <glasser@davidglasser.net> wrote:
>>> On 10/17/07, Philip Martin <philip@codematters.co.uk> wrote:
>>>> "David Glasser" <glasser@davidglasser.net> writes:
>>>>
>>>>> Oh, and also --parallel is broken: it makes a bunch of threads, each
>>>>> of which call popen3 and then wait, but there's no guarantee that the
>>>>> test they wait on will be the test they popen3'd.
>>>> It also modifies an array in multiple threads without any locking; I
>>>> posted a mutex patch but then gave up when I encountered the wait
>>>> problem.
>>> From searching, it does appear that .append on a list should be
>>> atomic. However, I think we'll be happier if we just change it to use
>>> Queue. Will work on that now...
>> See r27264. In addition to only using explicitly synchronized data
>> structures, I think the new design is conceptually cleaner, and it
>> eliminates a lot of sleeps; seems to speed me up a small amount too.
>> Unfortunately the wait issues still exist.
>
> Hopefully fixed the wait issues in r27266.
Looking at INSTALL, we require Python 2.2 for Windows. Looking here:
http://effbot.org/downloads/#subprocess
they supply subprocess for Python 2.2 and 2.3. So we know that some versions of
subprocess work on older versions of Python.
So maybe we can just include subprocess in the test suite and use that. We just
have to fall back to the popen2.Popen3() code for Python 2.{0,1} on non-Windows
systems.
Blair
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Thu Oct 18 02:48:29 2007