dev call 20230302

Dev Call: 2 March 2023

Present

Tim, Florian, Antonio, Simon, Petr V., Du Toit, Mohaned

Agenda

Still working on the release 0.9.

• Merged PRs:

  • Add HPyUnicode_FromFormat(v) and HPyErr_Format.
  • ENH: add HPyContextVar* interfaces and document how it is done
  • Improve HPy documentation
  • Introduce HPy_Compile_s and HPy_EvalCode.
  • Debug mode: Check correct context usage
  • Verify HPy_TypeCheck's 2nd arg in debug mode only.

• PRs in progress:

  • Support vectorcall protocol

• Numpy/HPy

  • Matti implemented following HPy APIs used in the Numpy/HPy port:

    HPy HPyContextVar_New(HPyContext *ctx, const char *name, HPy default_value);
    int HPyContextVar_Get(HPyContext *ctx, HPy context_var, HPy default_value, HPy *result);
    HPy HPyContextVar_Set(HPyContext *ctx, HPy context_var, HPy value);
    

  • Florian implemented (PRs are open or soon to be opened):

    int HPyType_IsSubtype(HPyContext *ctx, HPy sub, HPy type);
    const char *HPyType_GetName(HPyContext *ctx, HPy type);
    HPy HPyUnicode_FromEncodedObject(HPyContext *ctx, HPy obj, const char *encoding, const char *errors);
    HPy HPyUnicode_Substring(HPyContext *ctx, HPy obj, HPy_ssize_t start, HPy_ssize_t end);
    HPy HPyDict_Keys(HPyContext *ctx, HPy h);
    

  • still open:

    HPy HPySeqIter_New(HPyContext *ctx, HPy seq);
    int HPySlice_Unpack(HPyContext *ctx, HPy slice, HPy_ssize_t *start, HPy_ssize_t *stop, HPy_ssize_t *step);
    

Language Summit Proposal

• Petr suggests to have a short HPy intro as well since not everyone will know it.
• Tim points out that we should mentioned/compare/involve Mark's new C API proposal
• Simon thinks that concentrating too much on Mark's proposal could
• Tim: in contrast to Mark's C API, HPy also addresses the problem that the C API is very tangled with CPython implementation.
• Simon argues, the Microsoft team (which tries improve interpreter performance) would have an easier life if HPy would already be the default.
• Simon points out that if CPython ignores HPy, it will be dead in the water.
• Petr thinks that it will be very hard that CPython core devs commit to something concerning HPy (since also a lot of the devs do not really care about the C API); but he thinks that HPy is still in a good position.
• Simon asks: do CPython core devs share the same frustration concerning the C API?
  • Petr answers: frustrations are the same among core devs and a few (including Petr) are convinced that HPy (or an approach like this) is the way to go.
• Our (HPy team) intention of the language summit talk is to get information about steps forward.
• Petr doesn't think that a PEP would really help.
• We could aim to get a statement like: "Everyone interested in the C API should support HPy"
• Florian will prepare some numbers for Numpy/HPy game of life for PyCon US

Vectorcall PR

• See also discussion on PR 309
• Let's go for vectorcall calling convention (i.e. C function signature HPy (*fun)(HPyContext *ctx, const HPy *args, HPy_ssize_t nargs, HPy kwnames)) as primary calling mechanism and convention (also change HPyFunc_keywords to that).
• We should provide some helpers for easy migration from tp_call. This could be some macro that generates a trampoline or just some helper functions that do conversion to the different calling convention.
• We can start without the flag HPy_VECTORCALL_ARGUMENTS_OFFSET and maybe add it later if it turns out to be useful.

Misc

• Du Toit is preparing work for internship at Oracle. Currently investigating related work, i.e., existing tools and mechanisms like Cython or CPython's subinterpreters.