I'm a bit concerned about change the --verbose and -v options. The build_src command is supposedly private, but is present in the API. Guess we might find out if it is a problem.

Note that testing uses --verbose=<number> to select verbosity, with short -v, -vv, -vvv options. We might want to go that route.

What happens to the reset verbose flag, do we just ignore it now?

What happens to the reset verbose flag, do we just ignore it now?

I didn't change that code at all.

Currently the default when building numpy is 'verbose'. The -v flag to runtests.py only impacts testing, not building. I am not sure we want to mix flags to the two different tasks

Edit: ahh, you mean to change to use an additional/different -v flag to only setup.py? The help for that command says:

$ python setup.py --help
...
Global options:
  --verbose (-v)      run verbosely (default)
  --quiet (-q)        run quietly (turns verbosity off)
...

What do you think of a strategy like:

• the default runtest.py is to call setup.py --quiet
• add a verbose-build flag to runtests.py that switches to setup.py --verbose
• all CI uses the verbose build

This might need more tweaking, since with current HEAD the build output is very terse (too terse???):

$ python setup.py --quiet build 
Processing numpy/random/bounded_integers.pxd.in
Processing numpy/random/philox.pyx
Processing numpy/random/bit_generator.pyx
Processing numpy/random/sfc64.pyx
Processing numpy/random/pcg64.pyx
Processing numpy/random/common.pyx
Processing numpy/random/mt19937.pyx
Processing numpy/random/bounded_integers.pyx.in
Processing numpy/random/mtrand.pyx
Processing numpy/random/generator.pyx
Running from numpy source directory.
/usr/lib/python3.6/distutils/dist.py:261: UserWarning: Unknown distribution option: 'define_macros'
  warnings.warn(msg)
Cythonizing sources
non-existing path in 'numpy/distutils': 'site.cfg'

It seems --quiet is too terse: it hides existing warning when building (for instance with clang-8). We do have the --warn-error option, but forcing that to true by default would fail on any platform that still has warnings. Maybe this PR should be viewed as an hack that should be removed when we clean up the remaining warnings?
xref gh-13816, this comment

@charris ping

Let's put this in as a step on the way. I don't worry about this much and don't think some experimentation will hurt.

Thanks Matti.