On Fri, Jul 28, 2017, at 04:16 PM, Daniel Holth wrote:
It looks like we've run out of things to say about PEP 517, except, how soon can we get it into pip?
I admire your optimism! ;-)
While I partly hope that I get a unanimous disagreement, as it would be simpler, I have a nagging concern about something that someone mentioned ages ago: does it make sense for building sdists and building wheels to be part of the same backend?
Flit now makes both sdists and wheels, but for a long time it only made wheels, and the two parts are largely separate: it wouldn't take much work to use flit's sdist machinery but build the wheel with a different tool (e.g. if it had compiled parts).
Requiring one backend to build both formats may result in a significantly higher barrier to entry for backend developers: I don't know if I would have started writing flit if PEP 517 had already been finalised and I had to make both wheels and sdists to comply with it. They're also, at least to my mind, quite different kinds of thing: an sdist is almost like an archive of a VCS tag, whereas a wheel is the end result of any build steps the project needs.
So I'd like us to circle back round and reconsider allowing projects to specify 'use tool X to make wheels, and tool Y to make sdists'. If everyone else thinks that's unnecessary, I think we'd all be glad to finish this discussion up, but this concern has been growing in my mind for a while, and I want to get it out there before we finalise the PEP.