Un-deprecate source distribution re-installation behaviour

[pradyunsg]

This was deprecated in favour of explicitly passing --force-reinstall.

[sbidoul]

Should that really be an error ? Or just a Requirement already satisfied: ... message ?

[pradyunsg]

I think an error is better, because otherwise this would be a "silent" behaviour change -- the behaviour changed but the user has no way to know how/why; especially if they don't have checks to ensure that the behaviour doesn't affect them. Given that the deprecation message explicitly advertised a replacement, I think it's a good idea to fail loudly in these cases.

[pradyunsg]

@pypa/pip-committers Re-reading #8711 and #5780 (comment), I do feel a bit confused about this behaviour change.

From the feedback we've gotten since the deprecation was announced, it's clear that using --force-reinstall isn't a good-enough replacement -- that behaviour is not sufficient for the use cases that users have mentioned. It does look like we'll bring in another round of behaviour changes whenever #5780 gets resolved too.

I'm not a 100% sure what to do here TBH. I'm inclined to pull out this deprecation, and then kick the can down the road for this, into #5780 -- since it does look like we want pip install foo.tar.gz to continue reinstalling and pip install https://example.com/foo.tar.gz to reinstall in some situations. There's a bunch of behaviour changes planned in that issue, and I'd much prefer we do them as a coherent effort.

[uranusjr]

since it does look like we want pip install foo.tar.gz to continue reinstalling and pip install https://example.com/foo.tar.gz to reinstall in some situations. There's a bunch of behaviour changes planned in that issue, and I'd much prefer we do them as a coherent effort.

+1 This needs to be done in one shot (which is most of the reason why I haven't touched it one bit)

[pradyunsg]

Alrighty. Here's a patch removing the deprecation then!

I'm hoping that this doesn't fail CI in some weird way. :)

[uranusjr]

Tests failed due network issues. Unrelated.

[sbidoul]

@pradyunsg @uranusjr sorry to be late to the game but, is it not better to keep this deprecation ?

IMO, "Reinstalling a source distribution despite an installed package having the same name and version as the installed package" is stil a behaviour that is dubious and we will probably want to change in the future. So continuing to warn users that they are doing something unusual seems like a good thing to do ?

[uranusjr]

I think “reinstalling a (direct URL; the warning did not happen for version-specified requirements) source distribution despite an installed package having the same name and version” in itself isn’t actually dubious on paper—in fact, it’s exactly what I’d want if I pip install foo==1.0 and then pip install foo @ file://.../foo-1.0.tar.gz. The behaviour I wouldn't expect is running pip install foo @ file://.../foo-1.0.tar.gz again still triggers a reinstallation. So to me, the deprecation message is, at best, inaccurate and covers too many false positives, so it’s better to do without it now, and try again in a future version when we can actually figure out what exactly we want to deprecate.