[Distutils] A possible refactor/streamlining of PEP 517

[Nick Coghlan]

On 4 July 2017 at 12:46, Nathaniel Smith <njs at pobox.com> wrote:
> if hasattr(os, "symlink"):
>     def prepare_build_files(tmp_dir):
>         os.symlink(os.getpwd(), tmp_dir)
>
> and now pip will do in-place builds. (Except on Windows, of course;
> Windows users get screwed as usual, but from scipy's perspective
> that's pip's fault.)

scipy shouldn't be relying on pip as their frontend for local
development builds (except via pip -e), and if frontends see backends
actively subverting their build policies, they'll be well within their
rights to blacklist those backends (or, more likely, push them into a
chroot or container, rather than just using a subprocess with normal
user level access to the host filesystem).

However, Donald's persuaded me that the cases where:

1. build_sdist will be called (i.e. not starting with a known sdist); and
2. build_sdist will throw an exception

are going to be sufficiently rare that we can drop the input
preparation hook, and instead just let the build fail in such cases
(the conclusion of the previous round of discussions suggested we
weren't going to be OK with that, so I'm happy to change my view based
on the updated info).

Going down that path also still leaves frontends with the option of
looking for a pre-existing PKG-INFO file when handed an arbitrary
directory and taking that into account when deciding which build
strategy they want to use (in-tree or out-of-tree), and which copying
strategy they use in the out-of-tree build case.

Cheers,
Nick.

P.S. `build_sdist` should still indicate failure by throwing an
exception, not by returning a non-string result. However, frontends
should also fail the build when a backend doesn't adhere to the
specification (so returning NotImplemented would technically be an
alternative way to trigger a build failure, just not a recommended
one)

-- 
Nick Coghlan   |   ncoghlan at gmail.com   |   Brisbane, Australia