Bump FD_SETSIZE on Windows

[pitrou]

No description provided.

[conda-forge-linter]

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

[ocefpaf]

You need to bump the build number to get that change in.

[ocefpaf]

Can you write a quick sentence to why this change is needed?

[mrocklin]

Why decide on number in particular? Why not twice or half as many?

[pitrou]

Like any arbitrary limit, it just feels a priori sufficient :-) Note there's a slight runtime cost as the fd array is runtime allocated; I think the cost is practically negligible but still.
By default:

>>> a, b= socket.socketpair()
>>> %timeit select.select([a], [a], [a])
The slowest run took 12.48 times longer than the fastest. This could mean that a
n intermediate result is being cached.
100000 loops, best of 3: 2.89 µs per loop

With this PR:

>>> %timeit select.select([a], [a], [a])
The slowest run took 24.39 times longer than the fastest. This could mean that a
n intermediate result is being cached.
100000 loops, best of 3: 3.52 µs per loop

[pitrou]

@ocefpaf, thanks for the review comments. I think I've addressed them.

[pitrou]

I've rerendered in order to fix the CI bullds.

[jjhelmus]

I'm not apposed to this change but I do think some discussion is warranted.

The patches that are currently applied are either cosmetic or are needed to allow python to operate properly in a conda environment. This patch is different, it changes the behavior of python in a manner that is not necessary for python to work in a conda environment (at least as far as I can tell).

Additionally, this could create cases where a users script would work properly with the conda-forge python but not with the python from the defaults channel or the Python installed from the python.org installed.

I think making "improvement" to packages can be a slippery slope, who gets to decide what defaults should be changed or what trade-offs should be made.

Finally, has a similar change been suggested upstream, and if so, how was it received?

[pitrou]

@jjhelmus, the upstream issue reference is in the patch.

[mrocklin]

Also issue raised for default channel: ContinuumIO/anaconda-issues#1241

[jjhelmus]

Thanks for pointing out the upstream issue @pitrou, missed that link when I first looked at this. I'm in favor of this change as this should keep this compatible with future versions of Python in the defaults channel.

I think merging this change should wait until conda-forge moves to conda build 2.x to avoid creating a package missing the __pycache__ directory, see #70.

[jjhelmus]

@pitrou Now that the __pycache__ issue is solved (see #70) do you want to rebase or create a new PR with a bump to FD_SETSIZE? Python 3.6 already includes this change in the master branch but Python 3.5 still has the smaller size in the 3.5 branch.

[jjhelmus]

This patch has been applied to Python 3.5 #109 and Python 3.6 #101. Should this also be applied to Python 2.7, or can this issue be closed?

[jjhelmus]

Closing issue. Feel free to reopen if there is interest in applying this change to Python 2.7.