chore(ci): speed-up tests
Maybe we should just get os.cpu_count() or len(os.sched_getaffinity(0)) as a default & override where it needs to be overriden ? (probably only TravisCI)
Last time when I try this it crashed because some collision in pip install using pip cache. Maybe the current timeout maybe also connected with this.
Last time when I try this it crashed because some collision in pip install using pip cache. Maybe the current timeout maybe also connected with this.
This was investigated in https://github.com/pypa/cibuildwheel/issues/1254, I think we worked around this in https://github.com/pypa/cibuildwheel/pull/1301 , so shouldn't be an issue any more.
Not huge speedups, but it looks reasonable. Any particular reason to hard-code the concurrency on travis/azure rather than letting os.cpu_count() do it?
Let's merge this for the speedups, then revisit if we can use a smarter method to get the num threads.