[IPython-dev] Requiring distribute on Windows
Fri Jun 24 14:20:53 CDT 2011
On Fri, Jun 24, 2011 at 12:20, Brian Granger <email@example.com> wrote:
> On Fri, Jun 24, 2011 at 11:52 AM, MinRK <firstname.lastname@example.org> wrote:
>> On Fri, Jun 24, 2011 at 11:01, Brian Granger <email@example.com> wrote:
>>> In the course of testing IPython.parallel on Windows, Min and I have
>>> discovered that ipcluster won't work on Windows unless distribute is
>>> installed and used to install IPython. The reason is that setuptools
>>> has a bug that interacts with multiprocessing in a subtle, nasty way.
>>> Min has been using distribute all along for testing and dev., so we
>>> didn't see it until now. This leads to a question:
>>> Should we required distribute to be installed and used to install
>>> IPython on Windows?
>> The issue is that distribute writes scripts with the traditional `if
>> __name__ == '__main__'` block, whereas
>> vanilla setuptools does not, breaking any use of multiprocessing in
>> entry points on Windows (all child processes will enter the outer
>> loop, launching an infinite number of children).
>> I *have* figured out a way to catch this in the entry point itself
>> using inspect, so we shouldn't have to depend on distribute, if this
>> fix is acceptable:
> I will look at this later today. That would probably be better than
> requiring distribute though.
I discovered a cleaner check using multiprocessing instead of inspect:
>>> I am not sure how this will affect the .msi installer. We might be
>>> able to relax the requirement for that installer as the command line
>>> scripts are created differently.
>>> Brian E. Granger
>>> Cal Poly State University, San Luis Obispo
>>> firstname.lastname@example.org and email@example.com
>>> IPython-dev mailing list
> Brian E. Granger
> Cal Poly State University, San Luis Obispo
> firstname.lastname@example.org and email@example.com
More information about the IPython-dev