[FR] Lock entry for foo[bar,qux]: tell which deps come from bar vs. qux

[jvolkman]

I'm using PDM's resolver in rules_pycross by way of parsing its lock files. Recently I've been working on adding support for extra dependencies. My plan is to model something like foo[bar,qux] as three packages: foo, foo[bar], and foo[qux], with the latter two depending on foo plus whatever other optional dependencies.

This is almost how PDM structures its lock files, with one exception: foo[bar,qux] is modeled as a single package depending on foo and the union of optional dependencies from bar and qux, but there's no way to tell which optional dependencies are contributed by which group.

One way I've thought of to fix this is to add another layer to the dependency graph: foo[bar,qux] would depend only on foo[bar] and foo[qux] and nothing else, and the single-extra packages would each list their own dependencies plus the dependency on foo.

Another way would be to keep the current structure, but to include an extra marker with each dependency in foo[bar,qux] that says which one provides it.

I like the first approach best: it seems straightforward and has some nice deduplicative properties: if foo[bar], foo[qux], and foo[bar,qux] are all present in the locked set, currently the dependencies for bar and qux would be listed twice, but with the first approach each would be listed only one.

If this sounds reasonable, I'd be willing to work on a PR.

[frostming]

The second way makes more sense to me, but you need to make sure the extra markers will be evaluated correctly when traversing the lock entries.