dev meeting 2024 01 24

Present: @rgrinberg @Leonidas-from-XIV @moyodiallo @ElectreAAS Maxim Grankin, Boning Dong

• Changes to install in 3.11, Maxim

  • In 3.11 we deprecated installing to parent directories (../)
  • How long will the functionality be still available? Until 4.0 at least
  • As an alternative to this, can new sections be added?
    • sections are mostly inherited from OPAM, for compatibility
    • Files can be installed in the prefix root (dune install --prefix=DIR)
    • New sections with a flag as guard welcome to be submitted as PR

• Submodule support (@Leonidas-from-XIV)

  • Submitted
  • Missing support for relative file paths, although that is not a common use case (the Git CLI has guards against doing this) and if people need it it can still be added

• Git support (@rgrinberg)

  • We need to make sure packages that have Git sources use the Rev Store
  • @Leonidas-from-XIV to check that this is happening

• Missing items for the MVP (@Leonidas-from-XIV)

  • Pinning
  • Relocatable ocamlfind (@emillon would know about the progress on this)
  • Performance improvements
  • We are in good shape!
  • @rgrinberg has a list of outstanding issues that can be picked up

• Opam-repository changes (@Leonidas-from-XIV)

  • @Leonidas-from-XIV attended the meeting about scaling the opam-repo ocaml/opam-repository#23789
  • The proposal with currently the biggest amounts of proponents puts archived packages in a secondary repo
  • No issues for the dune packaging support

• Duplicate packages in multiple repos (@ElectreAAS)

  • @ElectreAAS wants to fix a part of the packaging code that @Leonidas-from-XIV was concerned about
  • We pass a list of valid package candidates to the solver in the order that the user specified the repositories in
  • The solver would give us a package + version and to determine which repo it was from we would need to search for it
  • This could potentially create issues where the package we find and the package the solver wanted are different if name + version are the same
  • @rgrinberg fixed this by never giving passing the solver the same name + version twice
  • OPAM behavior is identical in this case (verified in a later discussion with @kit-ty-kate)
  • It would be nice to have a test to check this behavior