Nogil

[ahojnnes]

No description provided.

[JDWarner]

@ahojnnes Looks great so far, let us know when ready to merge.

[jni]

This is tremendously exciting...!

[mrocklin]

Hooray for releasing the GIL. Curious have errors come up? We often rely on the GIL without realizing it.

When this is complete I would personally love to see a public writeup on the experience of releasing the GIL within scikit-image. I think that the community should do this more broadly; scikit-image could lead a bit here.

[ahojnnes]

Assuming that all tests pass, I am through with "nogilling" all Cython code.

@mrocklin I think, it was pretty straightforward. It is very helpful to use cython -a ... to figure which code actually accesses Python objects (usually only the yellow lines). Going through the code once again also helped to replace some Python code with pure C code.

[stefanv]

Stellar work! Doesn't look like NAN is recognized under Windows :/ Also, perhaps good to rebase on master.

[ahojnnes]

Thanks Stefan. This should hopefully do the trick.

[stefanv]

This is a major step forward--thanks @ahojnnes !

[mrocklin]

Hey look, momentum pandas-dev/pandas#10199

[jreback]

I DID ruthlessly steal your test_parallel (open-source, gotta luv it!)

[ahojnnes]

No problem, go ahead and copy as much as you want ;-) I am glad it gets reused and look forward to feedback or ideas how to improve the testing.