Skip to content

Coq Call 2023 07 11

Emilio Jesús Gallego Arias edited this page Jul 11, 2023 · 11 revisions

Topics

  • Reducing the STM (Emilio)
    • goal: reduce the surface Stm API, see what we can deprecate / remove, close bugs
    • vio2vo: is that used? Due to #4123 not likely
    • vio files: replaced in practice by vos?
    • query workers?
    • xml protocol: goals hints and evars are unused in coqide
    • stm: some parts of the error recovery code of 8.6 didn't work
    • classification: can we use the vernac_interp type to derive it?
    • coqc vs simple compiler?
  • Issues pertaining to DMs (Emilio)
  • Adding primitives to better control (i.e., force) reduction (Rodolphe, BedRock Systems)
    • This is something we discussed and came up with with PMP a while ago.
    • We have a draft PR, and would like feedback on whether something like this could eventually get in.

Roles

  • Chairman: Théo
  • Secretary: Hugo, then Emilio

Notes

  • Reducing the STM (Emilio)

    • see list of features above questioned by Emilio
    • Emilio suggests some lightening and refactoring of the STM
    • Théo proposes that Emilio opens PR on which discussions can happen
  • Issues pertaining to DMs (Emilio)

    • Théo suggests that it is not necessary that a solution to the various issues are given within the Coq repository
    • Emilio proposes to have specific discussions in September
  • New reduction primitives:

    [Presentation by Rodolphe]:

    • Adding new primitives to Coq's kernel to have better control of reduction in proof by reflection.
    • Use in IRIS, problems with forcing reduction
    • An IRIS proposition is reified into a datatype
    • During Kernel type-checking there is no way to force reduction under binders
    • At QED time that's needed, likely to gain a factor of 2x
    • New primitive: let_lazy, like let, but body is typecheck with reduced version
    • But you need a way to stop that reduction at certain points
    • Usually the leaves of the reified Ast, which are "private"
    • Main question: how likely this gets in?
    • Pierre-Marie was optimistic.
    • Blocking primitives implemented but not pushed
    • Proposal still not enough for the IRIS use case

    [Discussion]:

    • Q: How does this compare to math-comp's unlock / lock?
    • A: AFAWCT math-comp cannot reduce under binders
    • Some discussion about how that would interact with existing system in math-comp, also more validation and documentation is needed.
    • Some more discussion about how the system does work
    • Q: Who to get in touch with to continue work on this?
    • A : In addition to Coq devs, G. Gonthier and Assia / Cyril may be good reviewers.
    • Q: Would it work to have a constructor that forces the term?
    • A: No because still the let is "CBN"
    • Q: current PR targets mainly non-users (internal)
    • A: Yes, for now (more discussion as to see more visible applications)

    [Conclusion]:

    • We agree this in an important feature to have
    • Next step: push implementation of blocking primitives, more testing / validation
    • Next step: discussion / review with other experts (mainly math-comp) => understand the relationship of IRIS and math-comp needs
    • Once it is ready we expect it to land in Coq
Clone this wiki locally