EPIC: Typing enhancement #469

Activity

1. leofang

   added

   cuda.coreEverything related to the cuda.core moduleEverything related to the cuda.core module

   enhancementAny code-related improvementsAny code-related improvements

   P1Medium priority - Should doMedium priority - Should do

   on Feb 24, 2025
2. leofang

   added sub-issues on Feb 24, 2025

   • CI: Enforce type checks #312
   • Stop repeating input/output type info in docstrings #454
   • Avoid relying on typing.TYPE_CHECKING to the maximum possible extent #468
3. leofang

   added

   EPICSoul of a releaseSoul of a release

   on Feb 24, 2025
4. carterbox commented on Mar 7, 2025

   

   carterbox

   on Mar 7, 2025

   Last edited by carterbox

   Contributor

   More actions

   Something that I noticed when using migrating nvmath-python to cuda.core is that not all types are importable from the same level. For example

   from cuda.core.experimental import Device, Stream
   from cuda.core.experimental._event import Event  # shouldn't need to import _event to get Event
   
   def some_function(device: Device, stream: Stream, event: Event) -> None:
       pass

   If you're a downstream project and you also want to use type hints, the Event type is not part of the public API, but it is returned from parts of the public API, so it should be accessible without having to import a private module.
5. rparolin

   added the

   Epic

   issue type on Sep 24, 2025
6. rparolin

   added this to the cuda.core parking lot milestone on Oct 13, 2025
7. rparolin

   removed this from the cuda.core backlog milestone on Apr 2, 2026
8. rparolin

   added this to the cuda.core next milestone on May 21, 2026
9. rparolin

   assigned

   isVoid

   on May 29, 2026
10. mdboom commented on Jun 10, 2026

    

    mdboom

    on Jun 10, 2026

    Contributor

    More actions

    I think #2061 largely closes this bug, with the exception of the subissue #454 (stop repeating types in docstrings).
11. seberg commented on Jun 15, 2026

    

    seberg

    on Jun 15, 2026

    Last edited by seberg

    Contributor

    More actions

    One thing I noticed in gh-2215 (it is made worse by that PR), is that the auto generation of .pyi files from cython files works fine but preserve decorators.
    That means that it keeps @cython.final (and now @cython.critical_section). I think @cython.final is typing friendly (@cython.critical_section probably not), however, we shouldn't even have cython as a runtime requirement.

    I suspect we need to make this change upstream in https://github.com/jon-edward/stubgen-pyx, though.

    (Although, if we are OK with the cython dependency for the stubs, then it can also be just nudging/making sure Cython types critical_section nicely.)
12. rparolin

    modified the milestones: cuda.core 1.2.0, cuda.core next on Jul 27, 2026