RFC: a more flexible flake-parts API for dream2nix

[oati]

trafficstars

Currently, the flake-parts API has two "layers":

• dream2nix.inputs.<name>, with the options source, settings, inject, and overrides
• dream2nix.inputs.<name>.projects.<name>, with the options name, version, relPath, subsystem, translator

I think these two layers should really be collapsed into one layer, with a "translator inputs" option for each project. This would solve a number of problems:

• A translator's goal is to produce a dream-lock. For the most flexibility, it should be allowed to do so "by any means." For example, the hackage translator takes a name and a version and produces a dream-lock. source doesn't really make sense as an option for this translator.
• sourceOverrides and packageOverrides (in the "inputs" layer) only make sense in the context of a subsystem, but subsystem is defined later in the "packages" layer
• relPath can always be incorporated into source, and feels redundant. e.g. ${some-path-or-derivation}/subdirectory
• For some translators, version has no effect on the output. (cabal-plan writes the same version into the dream-lock regardless of the version input.) This could be made less confusing (and overrides more explicit) with custom "translator inputs" types for each translator.
• "name" is defined four times in the current API, and it is not clear how each option affects the behavior of dream2nix:
  • dream2nix.inputs.<name>
  • dream2nix.inputs.<name>.pname
  • dream2nix.inputs.<name>.projects.<name>
  • dream2nix.inputs.<name>.projects.<name>.name