Skip to content

dev meeting 2024 01 10

Jump to bottom
Marek Kubica edited this page Jan 24, 2024 · 1 revision

Present: @emillon @gridbugs @Leonidas-from-XIV @moyodiallo @nojb

  • Expansion in (modules) field (@nojb)

    • https://github.com/ocaml/dune/pull/9578
    • dynamic forms like (:include) or %{read-lines} can be used in (modules)
    • the included file needs to come from a different directory
    • the error message does not provide a good hint
    • other places with this kind of dynamic dependencies have the same characteristics
    • this can replace some usages of ocaml syntax
    • used at Lexifi for a sort of static plugin structure (modules selected at compile time)
    • this feature will be advertised in ocaml.org changelog

  • opam-compatible package name validation (@Leonidas-from-XIV)

    • we added extra checks for (depends) in cases where opam generation isn't used
    • some packages have errors in there but we're not sure how many
    • include check in 3.13.0~alpha1 and act depending on results
    • we can add bounds on affected packages if there are not too many
    • otherwise we can only perform the check (and type conversion) at use: when generating opam files and in pkg rules

  • 3.13 branching and updated release process (@emillon)

    • we now use the same branch for alpha versions and x.y.0
    • development can continue on main while the release is being done
      • just be careful in case some changes need to be backported
      • in particular, unrelated subsystems (e.g. coq) can continue being developed

  • single-command bootstrap (@emillon)

    • pointers: https://github.com/ocaml/dune/pull/9613 https://github.com/ocaml/dune/issues/9507 https://github.com/ocaml/dune/pull/9563
    • bootstrap process can use 2 strategies:
      • parallel: run compile commands in parallel and link the rest
      • single-command: run ocamlopt with a ton of arguments
    • today single is used if win32 or if -j 1 is set (implicitly or explicitly)
    • problem: parallel and single-command create different binaries
    • each strategy is reproducible though
    • assumption was that single would be faster than j1 on linux and jx on windows
      • from a quick benchmark the assumption is true on linux but false on windows
    • we lean towards removing that code path to simplify the bootstrap process and fix reproducibility issues

  • opam-repository scaling (@gridbugs)

Clone this wiki locally