Add per-requirement --config-settings

[q0w]

Closes #11325

[uranusjr]

Hmm, tests are failing but they seem to be unrelated. Something may be wrong in the CI?

[q0w]

I've tried to investigate, whats wrong with ssl but unsuccessfully

[sbidoul]

I tentatively added this to the 23.1 milestone as it'd be nice to have it at the same time as the removal of --install-option.

It looks like this PR is doing two things:

1. adding --config-settings as a per-requirement option
2. also propagating per requirements config settings to their transitive dependencies?

It seems (2.) was not done for --global-option and --build-option so I'd like to flag this to make sure everyone is on board with that.

I don't have a ton of practical experience with these options, but my instinct says it might be more flexible not to do that and let the user precisely specify which req must get which config settings.

[pfmoore]

I don’t have any practical experience to draw on either, but I strongly believe that we shouldn’t be introducing (2) silently. I’d rather see it as a separate PR so the implications can be discussed in isolation, but even if that isn’t possible I’d want to see some clear justification for the behaviour. It should also be explained clearly in the docs (as it’s far from obvious - see the comment by @uranusjr above) and explicitly noted as new behaviour in the changelog. I don’t believe we have any other case of settings being inherited by dependencies like this.

I’m personally -1 on behaviour (2), but mostly because it feels really confusing to me, and I expect it to be a source of bugs as users misunderstand it as well.

[q0w]

So should I remove the build flag propagation?

[sbidoul]

So should I remove the build flag propagation?

@q0w I think so yes.

Can you also add a test where config settings are passed both on the CLI and as per-requirement options?
We need to decide what to do in that case, but my current assumption is that we should merge both.
Please also add a test where the same config setting is passed on the CLI and and as per-requirement option, or multiple identical config settings are passed as per-requirement options (see #11681).

You may also want to base your work on #11876.

[q0w]

Can you also add a test where config settings are passed both on the CLI and as per-requirement options?

added pypa/pip@2f6f36c (#11634)

[sbidoul]

Keep in mind that this PR should not be merged before #11876 is.

[q0w]

after #11876 i cant pass merged cli/reqs config_settings to install_req_from_parsed_requirement

[sbidoul]

after #11876 i cant pass merged cli/reqs config_settings to install_req_from_parsed_requirement

Well that config_setting argument was not used so I removed it in #11876. But you can still re-introduce it if that's the best way to do it :)

[sbidoul]

Thanks @q0w ! A last round comment and it it should be good to go.

[pfmoore]

This looks pretty close to being ready for merging. If we can get this done in the next week or so, it'll make it into 23.1 which would be excellent.

[sbidoul]

@q0w following the discussion in #11915 (comment) we'll need to rework this PR to remove the merging with CLI requirements. Sorry again for the confusion. If you think you will not have time to do it by mid next week, just let me know and I'll finalize it.

[q0w]

@sbidoul Removed merging config settings

[sbidoul]

Thanks!