Stop injecting wheel as a build dep fallback
#12449
Conversation
PEP 517 doesn't mandate depending on `wheel` when a `__legacy__` setuptools fallback is used. Historically, it used to be assumed as necessary, but later it turned out to be wrong. The reason is that `setuptools`' `get_requires_for_build_wheel()` hook already injects this dependency when building wheels is requested [[1]]. It also used to have this hint in the docs, but it was corrected earlier [[2]]. It could be argued that this is an optimization as `pip` will request building wheels anyway. However, it also shows up in the docs, giving the readers a wrong impression of what to put into `[build-system].requires` when they start a new project using setuptools. This patch removes `wheel` from said `requires` list fallback in the docs and the actual runtime. [1]: https://github.com/pypa/setuptools/blob/v40.8.0/setuptools/build_meta.py#L130 [2]: pypa/setuptools#3056
webknjaz
added
type: bug
webknjaz
force-pushed
the
maintenance/pep517-fallback-wheel
branch
from
b6dae52
to
3769ad7
Compare
|
Motivation/backstory: I was reading https://gregoryszorc.com/blog/2023/10/30/my-user-experience-porting-off-setup.py/ (it was shared @ pypa/packaging.python.org#1468) and it got me puzzled — I thought I'd gotten rid of all the prominent places newbies may copy-and-paste the suggestion to depend on |
|
Similar PR in build: pypa/build#716. |
|
This is also still recommended in PyPA's sample project: https://github.com/pypa/sampleproject/blob/main/pyproject.toml#L153 |
That should be raised separately in that repo. |
|
Thanks for this, @webknjaz! |
|
Thanks, @pfmoore! Is there any chance that could be related to this patch? It doesn't seem to be the cause but is still worth checking... |
|
It should be easy to verify by restarting the CI (or at least one job) of the previous commit — https://github.com/pypa/pip/commit/e88d39ae49ab11a8b80609a018e5a36ea4ccad89/checks. |
|
I don't believe it was this run, but I've hit restart just to check. |
|
FYI CI runs are failing because of changes in setuptools which had a significant impact on Pip tests: #12458 |
|
@notatallshaw thanks for confirming! I had a hunch that maybe the test env isn't pinned well enough, but was not fully sure. |
|
Thanks @notatallshaw - I had a suspicion it was that. @webknjaz I'm not sure we should be pinning setuptools in the test env. The occasional failure, while annoying, is overall probably less so than having to update our pins at the sort of rate setuptools changes (and in general, I think we do want to know we work with the latest setuptools). |
|
Understandably so. OTOH, there's a middle ground — use pins for PRs and merges but add unpinned tests for nightly/cron runs. |
PEP 517 doesn't mandate depending on
wheelwhen a__legacy__setuptools fallback is used. Historically, it used to be assumed as necessary, but later it turned out to be wrong. The reason is thatsetuptools'get_requires_for_build_wheel()hook already injects this dependency when building wheels is requested [1]. It also used to have this hint in the docs, but it was corrected earlier [2].It could be argued that this is an optimization as
pipwill request building wheels anyway. However, it also shows up in the docs, giving the readers a wrong impression of what to put into[build-system].requireswhen they start a new project using setuptools.This patch removes
wheelfrom saidrequireslist fallback in the docs and the actual runtime.