Implement dedicated causes object

[notatallshaw]

This is proof of concept to address issues raised in https://github.com/sarugaku/resolvelib/pull/92#issuecomment-955139430

Derived information can then be calculated and cached as needed. In particular the downstream provider can run identifier in backtrack_causes.names where names is a property that calculates the underlying information lazily and only has to be re-calculated if the underlying causes are changed.

I'm trying to see if this approach would have consensuses, several things in the code would need fixing before merging.

[notatallshaw]

A nice thing about this approach is it allows the downstream library to be decoupled from the internal structure of the causes. Maybe helpful?

[uranusjr]

I think I like this approach the best, but we need to go a bit further down the decoupling path. Taking things like pypa/pip#10478 into account, perhaps we should take a similar approach to Requirement and Candidate and add an interface in Provider custom Cause object logic? (We can provide a default implementation if that’s a good idea.)

[notatallshaw]

That makes sense to me, I will take a look at how Requirement and Candidate are implemented and mirror that approach.

[hswong3i]

So #97 and 1.0.0 should now depends on this PR?

[uranusjr]

Yes.

Technically not this PR specifically, but the implementation of a new provider method that allows creating a custom conflict cause object that the resolver will modify and pass back to other provider hooks. That may happen in this PR, another PR, or a combination of several.

[hswong3i]

Seems a huge workload for Cause, which may need more than a month for implement it precisely… Maybe this breaking change could go into 2.0.0? Or just implement the API interface changes now for 1.0.0 without indeed and perfect implementation, yet?

[jbylund]

Technically not this PR specifically, but the implementation of a new provider method that allows creating a custom conflict cause object that the resolver will modify and pass back to other provider hooks. That may happen in this PR, another PR, or a combination of several.

Isn't that what #96 is? Using that approach a provider could pass make_causes_obj_from_raw_causes if it wanted to (though my preference would be for just set of strings).

[uranusjr]

Yes except I want it to be a provider method and create the entire cause object, not modify it.

[notatallshaw]

Not had any time to work on this this last month. However I expect to have some free time during the holidays. I'm probably going to create a new PR and close this one, implementing it like Requirement and Candidate the way @uranusjr suggests.

I don't feel any ownership over doing the work 😉, so if anyone else wants to take the approach and implement it themselves feel free to, otherwise I will come back with an update later this month.

[notatallshaw]

I think I like this approach the best, but we need to go a bit further down the decoupling path. Taking things like pypa/pip#10478 into account, perhaps we should take a similar approach to Requirement and Candidate and add an interface in Provider custom Cause object logic? (We can provide a default implementation if that’s a good idea.)

I've addressed these comments in a new PR: https://github.com/sarugaku/resolvelib/pull/99 and an accompanying pip PR: https://github.com/pypa/pip/pull/10732