dev meeting 2024 10 16

Agenda

• Issues cleaning: how could we organise? (@maiste)

Minutes

Attendees: @gridbugs @ManasJayanth @maiste @moyodiallo @ElectreAAS @rgrinberg @Leonidas-from-XIV

• organizing the backlog of issues in ocaml/dune

  • we have many issues open, several possibly overlapping
  • should we take some time to pass over all the issues and see if any can be closed?
    • Rudi has read through all the issues this at some point
    • we're all free to do work maintaining the issues
    • labeling/triaging issues can help
    • is it reasonable to ask users to try out potential fixes?
      • feel free to close items if you think you've fixed them

• dune binary dev tools

  • why do it with dune?
    • because we decided to make dune the frontend of ocaml
  • how will it work with different compiler versions?
    • could use the available field so that packages are only available when the correct compiler is present
  • reason did something similar, releasing a binary lsp server
  • why not do it with other packages such as the compiler
    • no reason we couldn't as long as the package is relocatable
    • can't do it with the compiler as it's not relocatable (yet!)
  • even if we distribute ocamllsp as a binary it still won't work unless the code has been compiled
    • so the compiler will also need to be installed

• which experimental features to stabalize in dune and enable by default

  • toolchains
    • it's needed so that the solver chooses a version of the compiler that works

• this issue: https://github.com/ocaml/dune/issues/11002

  • could only allow specific variable names by default until a flag is set to allow more
  • new variables are added in opam files and from the command line so there's no way to completely disallow certain variable names in dune
  • could also fix by adding a different syntax to dune for referring to non-standard variables

• solver in dune

  • tarides has a vague plan to improve error messages
  • within the zero-install solver there is an intermediate representation that represents smt (or sat?) problems as "zero install packages" and this is not helpful to us and complicates the solver.
    • it has many features that don't apply to opam packages (e.g. "commands", "role replacements", "machines")
    • might be easier to plug opam directly into the smt solver
    • could also rewrite the solver completely though this is probably overkill
    • simplifying the solver will allow us to generate errors with enough context to make them useful
    • could also speed up the solver
  • eagerly loads opam files into the formula which makes it slow
    • applies constraints after building the formula so it loads the wrong versions

• should it be an error if a package depends on a disjunction of packages, some of which don't exist?

  • allowing this could simplify some workflows involving pinning
  • but it's a bit strange to allow packages to refer to non-existat packages
  • could at least make this an error in dune and see if anyone complains