> I'm more or less persuaded by Nathaniel's argument that the source directory shouldn't be on sys.path
I do too. There should be an option in pyproject.toml to disable this behavior though so that numpy can build itself. 2017-08-25 16:21 GMT-05:00 Thomas Kluyver <tho...@kluyver.me.uk>: > On Fri, Aug 25, 2017, at 10:02 PM, xoviat wrote: > > While we're waiting for others to respond, what are we going to do with > respect to the sys.path issue? I don't think there has been discussion on > that yet. > > > I'm more or less persuaded by Nathaniel's argument that the source > directory shouldn't be on sys.path, because: > > - Most projects won't need it to be > - There are plausible ways it could go wrong > - I think we could make a backend which loads the real hooks from the > source dir, so there's a backdoor for projects that do want it. > > IIRC, however, Nick was still strongly in favour of including the source > directory on sys.path, and he can usually persuade me of his position with > a few emails. I do think that behaviour is more like the 'obvious thing' > that Python programmers would expect. > > Leo: > > > - removing srcdir from sys.path requires explicitly removing "" from > the head of sys.path, which is not normally done. > > Not necessarily. If the hook process is launched with a script at the > command line, it is the script directory, not the cwd, that is added to > sys.path. My WIP hook-calling module works this way: > https://github.com/takluyver/pep517/blob/ee43a9334c377d7c37badcc8527cb7 > a8500180f7/pep517/wrappers.py#L75 > > With this code, we would actually need to explicitly *add* the source > directory (CWD) to sys.path before importing the hook. > > > - it would break many setuptools based projects, probably forcing the > setuptools backend to reinsert it, defeating the purpose of removing it in > the first place. > > I don't think we should decide this based on what setuptools does; I > expect it to need quite a bit of special handling in any case. > > Thomas >
_______________________________________________ Distutils-SIG maillist - Distutils-SIG@python.org https://mail.python.org/mailman/listinfo/distutils-sig