dev meeting 2025 06 11

Agenda

• release 3.19.1 (@Alizter)

  • needs https://github.com/ocaml/dune/pull/11879
  • any other fixes?

• time to deprecate the monorepo benchmark (Steve, who won't be present but please discuss this anyway ;) )

  • nobody is maintaining it, Steve is the only one who knows how to maintain it (though there are docs on the Tarides internal wiki)
  • originally introduced so we could benchmark dune on a large monorepo so Jane Street could get a sense for if it was fast enough for them to adopt
  • this adoption has now taken place (afaik)
  • ostensibly the benchmark is also there to detect regressions in upstream dune, but Jane Street seems to have forked dune in the meantime for the purposes of oxcaml

• OxCaml extension and pform (@maiste)

  • For OxCaml, I tried to add an extension to guard the variable %{oxcaml_supported} but it creates a dependency loop with Pform
  • Two options:
    • Manually register the oxcaml with a ref (Found it ugly)
    • Instead of a variable create a new field with_oxcaml_supported.

Meeting notes

Attendees: @maiste @rgrinberg @shym @Alizter @rikusilvola @aguluman @Leonidas-from-XIV

Release 3.19.1 (@Alizter)

• Ali had worked on improving the way programs are launched, but it had lead to a regression
• Reverted back to the old way
• It was released earlier this morning to opam-repository
• CI was stuck because it could only use released versions, @maiste has created two PRs to enable CI to use the unreleased dev-version
  • https://github.com/ocaml/dune/pull/11884
  • https://github.com/ocaml/dune/pull/11883
• We should otherwise stick to our usual release cadence
• There are some issues setting up a clean switch
  • But it seems developing on OCaml 5.3.0 works perfectly fine aside from some warnings about alert in the bootstrap process

Obsolete monorepo benchmark (@gridbugs)

• Only maintained by @gridbugs
• Made for adoption within Jane Street
• Jane Street does not need it any more
• It takes a long time to build
• It can be removed
• This is the one benchmark which takes a really long time, the other benchmarks are still useful

OxCaml support flag (@maiste)

• There should be a possibility to determine whether building oxcaml is suported
• It can be a pform but should be guarded behind the oxcaml extension
• But determining whether to support it or not causes a circular dependency
• set_decoding_environment can be used to register extensions
• We have had this situation before, Melange might be doing something similar, Coq definitely does, the appraoch can be copied
• It's stored in an universal map, shouldn't cause any circular dependencies