[Distutils] waf can be used to compile Python extension modules

[Daniel Holth]

On Fri, Mar 21, 2014 at 10:25 AM, Paul Moore <p.f.moore at gmail.com> wrote:
> On 21 March 2014 13:46, Daniel Holth <dholth at gmail.com> wrote:
>> It mostly has to implement egg_info and install to be pip compatible.
>
> At a *very* brief scan, pip uses just the following setup.py
> invocations (this definitely needs to be properly audited before
> relying on it):
>
> For installation from sdist
> * setup.py egg_info [--egg-base XXX]
> * setup.py install --record XXX [--single-version-externally-managed]
> [--root XXX] [--compile|--no-compile] [--install-headers XXX]
>
> For editable installs
> * setup.py develop --no-deps
>
> Not sure, used in RequirementSet
> * setup.py clean
>
> For pip wheel
> * setup.py bdist_wheel -d XXX
>
> Note that installation from wheel requires *no* setup.py interface.
> Nearly all of this is for installation from sdist (which long-term is
> expected to be deprecated - install via wheels is the intended "way of
> the future", and PR #1572 is the first step on the way to that). While
> theoretically you can build wheels any way you like, I expect "pip
> wheel XXX" to be the canonical approach (it's how PR#1572 works) and
> at the moment that means implementing setup.py bdist_wheel.
>
> So, at a bare minimum, you could get away with just supporting
> setup.py bdist_wheel (and not supporting editable installs or installs
> direct from source). It might need PR #1572 to make this palatable, of
> course.
>
> Also, there's the fact (which I glossed over) that pip currently
> doesn't run setup.py directly, but does some nasty hackery to inject
> setuptools. Before working on standard build interfaces, we probably
> need to refactor that out somehow. Maybe a simple flag (that could be
> set in setup.cfg) that says "use setup.py without monkeypatching"
> would be enough...

No worries... if you are re-implementing setup.py, then setuptools is
monkey-patching code that you won't be executing.