Fix compile cached vcs packages #1649
Conversation
for more information, see https://pre-commit.ci
| # from ``ireq.link`` when the link is found in the wheel cache. | ||
| # ``ireq.link`` would point to the wheel cache, while this points | ||
| # to the found remote link (e.g. from pypi.org). | ||
| link=candidate.source_link, |
There was a problem hiding this comment.
Glad your copy_install_requirement extensibility made this easy!
There was a problem hiding this comment.
That's very handy indeed!
tests/test_cli_compile.py
Outdated
| # Install and cache VCS package | ||
| # TODO: install package in isolated environment | ||
| subprocess.run( | ||
| [ | ||
| sys.executable, | ||
| "-m" "pip", | ||
| "install", | ||
| vcs_package, | ||
| ], | ||
| check=True, | ||
| ) |
There was a problem hiding this comment.
On my machine, this doesn't cause a wheel for six to be cached in ~/.cache/pip/wheels. I did a quick sanity check with [sys.executable, "-m", "pip", "cache", "dir"] here to confirm that ~/.cache/pip/wheels is the relevant cache on my machine when running under tox -e py310-piplatest-coverage.
I had a similar issue when writing the MWE script in #1647. Pip would not cache the wheel unless I installed another version of the VCS package first.
Doing that doesn't seem to fix the issue here though. On my machine, replacing typeguard @ ... with six @ git+https://github.com/benjaminp/six@65486e4383f9f411da95937451205d3c7b61b9e1 in my MWE script doesn't actually cause the issue to occur—for some reason I can't get Pip to cache a wheel for this version of six.
There was a problem hiding this comment.
Suggested changes atugushev/pip-tools@fix-compile-cached-vcs-1647...gschaffner:pip-tools:fix-compile-cached-vcs-1647.
bb2e91edbd3d0f1bdd86c41e84c88621f8d35690 is failing—I can't find a way to get Pip to cache six-1.16.0-py2.py3*.whl. I am not sure why1, but the easiest fix is probably to change to another package that supports 3.7–3.10, has no dependencies for this range of Pythons, and that Pip will cache a wheel for, e.g. Plumbum 1.7.2. As far as I can tell, there isn't another candidate for this that's already used in pip-tools's suite elsewhere (the list of git+'s used elsewhere seems to be pip-tools, django, pip, and billiard).
1: My guess is that this is because six does not use setuptools_scm, but I haven't looked into this.
There was a problem hiding this comment.
The test works on CI. As you can see it fails without the fix:
AssertionError: assert 'six @ git+ht...05d3c7b61b9e1' == 'six @ file:/...-none-any.whl' - six @ file:///Users/albert/Library/Caches/pip/wheels/b4/42/6d/bab90ef8caaba6e66b7c1b7e35d83fbe1b67ae9a1101839ff2/six-1.16.0-py2.py3-none-any.whl + six @ git+https://github.com/benjaminp/six@65486e4383f9f411da95937451205d3c7b61b9e1
I think that should be enough.
There was a problem hiding this comment.
@gschaffner what would you think?
There was a problem hiding this comment.
Sorry for the delay!
Without the fix, the test mistakenly passes on my laptop when run as
$ git checkout 53f36d88bbaaad3ece3e964b015015690c6792e9 # or 835b7e58c4fb929e114d88fce7e1f911aef2c4cd
$ rm -r ~/.cache/pip/wheels
$ tox -e py310-piplatest-coverage -- --no-cov tests/test_cli_compile.py::test_compile_cached_vcs_package # or tox -e py310-piplatest-coverageCan you replicate? It seems like the test is not putting six into the cache.
(My initial guess was that the test was failing (as it should) on GHA because six was already in the wheels cache loaded by
This doesn't seem to be the case however—see that there is no wheel for six loaded in https://github.com/jazzband/pip-tools/runs/7439890847?check_suite_focus=true#step:7:223.)
There was a problem hiding this comment.
Ok, but isn't it all fine if you just replace six with any other package?
There was a problem hiding this comment.
That was my first thought. However, while the test works on CI, it can mistakenly pass in a different setup. This makes me think that the test is flaky and must be reworked in a way as suggested here.
There was a problem hiding this comment.
Ok, thanks, yeah I'm on board to get the fix out ahead of the test.
There was a problem hiding this comment.
Tracking issue #1686.
There was a problem hiding this comment.
@AndydeCleyre okay, all set.
atugushev
force-pushed
the
fix-compile-cached-vcs-1647
branch
from
53f36d8
to
57012f6
Compare
|
@gschaffner thanks! Feel free to. |
Fixes #1647
Contributor checklist
Maintainer checklist
backwards incompatible,feature,enhancement,deprecation,bug,dependency,docsorskip-changelogas they determine changelog listing.