pip wheel should have a "don't download wheels" option

[pfmoore]

When building wheels via "pip wheel" I often want to just build those wheels that I can't get from PyPI (for example, for upload to a local index). Current behaviour builds the specified requirements and all their dependencies, and also downloads any already-existing wheels into the target wheelhouse.

It would be good to have an option to not download already-existing wheels, so that the resulting wheelhouse only contains wheels that aren't available already. It's then possible to just upload the whole contents of the wheelhouse to the local index, without needing to do an "is this already on PyPI" check.

[rbtcollins]

BTW I think the title is wrong: we don't want to not-download-wheels, because we may need wheels to satisfy build time dependencies. We want to not write copies of them into the output directory.

[rbtcollins]

Oh and we'll definitely need the wheels to get dependency metadata today too.

[pfmoore]

Understood. I don't know the best name for the option yet (naming is hard :-)), but from a user perspective I don't want the wheels "downloaded" (the fact that pip needs to download them internally to check dependencies is an implementation detail).

[rbtcollins]

How about --only-built-wheels
On 8 Apr 2015 6:40 pm, "Paul Moore" notifications@github.com wrote:

Understood. I don't know the best name for the option yet (naming is hard
:-)), but from a user perspective I don't want the wheels "downloaded" (the
fact that pip needs to download them internally to check dependencies is an
implementation detail).

—
Reply to this email directly or view it on GitHub
#2657 (comment).

[Russell-Jones]

--only-build-wheels what? And isn't making "wheel" part of an option for pip wheel redundant? I'd suggest --skip-existing or --get-missing-only

[sublee]

--skip-existing looks good. Is this issue still active?

[xavfernandez]

Somewhat related to #1310 indeed. This issue is still active as far as I know.

[sublee]

I'm trying to make a Ci job. It builds requirements of my project as wheel archives nto in-house wheelhouse Web server. As you know, pip downloads all already built archives. So this CI job is always slow.

If we have no --skip-existing yet, is there some another way to make this CI job be faster?

[pradyunsg]

How relevant is this now @pfmoore?

[pfmoore]

Personally, it's no longer something I need (mainly because most of the major projects now have wheels available). I'm happy to withdraw it. Do any of the other commenters on this issue still see a need for it?