Fix get_affinity for linux

[ddaeschler]

This pull request fixes get_affinity under linux to properly deal with the scheduler CPU bitmask.

[giampaolo]

Why, what was the problem? Are you able to have a test case which fails before the patch and passes afterwards?

[giampaolo]

Note: the travis build appears to break after your patch so it seems something is wrong:
https://travis-ci.org/giampaolo/psutil/jobs/32151822

[giampaolo]

Ah sorry, I realized just now this is a PR for #522.

[ddaeschler]

This patch is related to #522.

On Ubuntu 12.04.5 LTS (precise) sizeof(cpu_set_t) == 128 bytes. _psutil_linux.c was instead passing a long (8 bytes). In some configurations appears to cause EINVAL to be returned. I imagine this is the case with any system that reports more than 64 CPU cores.

I think I've discovered what was wrong with my initial patch that is causing the test failure.. I just need to figure out how to fix it and update this pull request.

[ddaeschler]

Integration failed, but now it's stating "unknown failure on travis". Not sure what to make of that.

[ddaeschler]

I caught a compilation error for one of the builds that was expecting C90 compliant code. I moved declarations around so that all the variables are declared at the top of the function before any other calls. Hoping this maybe fixes the CI issue.

[ddaeschler]

All set.

[giampaolo]

Do you think this might help you?
http://hg.python.org/cpython/file/18a311479e8b/Modules/posixmodule.c#l7891

[ddaeschler]

That source is using dynamic allocation so it would support more than 1024 CPU cores, whereas what I committed would support only 1024. I can do it that way if you'd prefer. This PR is working now though on a few setups I've tested.

[giampaolo]

I guess it would be better, so yes, if you still like it, please do it.

[ddaeschler]

That code returns a set instead of a list. Would you like the psutil implementation to do the same?

[giampaolo]

I'd still like to see a list as I don't want to introduce backward
incompatibilities in a minor version.
We can change that to a set() in the next major version (actually there's a
ticket for this).

On Mon, Aug 11, 2014 at 2:40 PM, ddaeschler [email protected]
wrote:

That code returns a set instead of a list. Would you like the psutil
implementation to do the same?

—
Reply to this email directly or view it on GitHub
#527 (comment).

Giampaolo - http://grodola.blogspot.com

[ddaeschler]

Ok. I'll implement this PR with a list. I can also change it to a set when the time comes

[giampaolo]

Thanks a lot for your effort.

On Mon, Aug 11, 2014 at 2:52 PM, ddaeschler [email protected]
wrote:

Ok. I'll implement this PR with a list. I can also change it to a set when
the time comes

—
Reply to this email directly or view it on GitHub
#527 (comment).

Giampaolo - http://grodola.blogspot.com

[ddaeschler]

This PR is ready with the new code as far as I can tell

[giampaolo]

Merged (finally). Sorry for waiting this long.

[mlowicki]

@giampaolo Great! When is next release planned?