Install: Build the extensions in parallel if the options has not been set

[Lauszus]

This roughly halved my compilation times when using 8 threads instead of just 1.

[matham]

Definitely a good idea, but I'm worried about making this decision for everyone. E.g. for make you can always specify how many threads.

I think we first should look at what other projects do (e.g. numpy), if they make it optional and what the default is!?

[Lauszus]

Numpy has parallel builds enabled by default, but limits it to maximum of 8 cores: https://github.com/numpy/numpy/blob/8087e1de04d011f7ce66bf80dbef4a0599b14f7a/numpy/distutils/misc_util.py#L78-L112.

This is where the actual code is compiled: https://github.com/numpy/numpy/blob/36a2bf0ce8b072d557af5b58a9003a6bc72caaef/numpy/distutils/ccompiler.py#L352-L361.

What I have done is similar however I do not limit the number of cores, but I can do that if you want to?

FYI here's the relevant code in distutils that checks the parallel attribute: https://github.com/python/cpython/blob/master/Lib/distutils/command/build_ext.py#L443-L469, as you can see it will get the number of cores of the CPU and use all of them, but if we set parallel to an integer, then that number of cores will be used.

[matham]

Ok, makes sense. I'd say that by default we should also limit the number of parallel jobs to like 4 for similar reasons as numpy. Numpy is a much larger project with more code so it makes sense for them to benefit of using upto 8 cores.

For us, there probably isn't much benefit above like 3 or 4 cores so lets limit it to the min of the number of cores and like 4.

[Lauszus]

@matham fixed in: 63a825a.