Skip to content

Minutes_2020_07_14

Jump to bottom Edit New page
Valentin Haenel edited this page Jul 20, 2020 · 1 revision

Numba Meeting: 2020-07-14

Attendees: Aaron Meurer, Ethan P, Juan Galvez, Lucio, Stuart, Siu, Val, Guilherme, Hameer, Todd A

NOTE FROM THE EDITOR: this was the second public community Numba meeting. As we continue to figure how best to run these, we currently simply record the most important statements in an attempt to summarize the conversation. Please feel free to follow up with us via our digital channels if something is out of context or unclear for you.

0. Feature Discussion

  • Discourse feedback

  • Feedback on deprecating Google groups mailing list.

    • There is still a reference in setup.py in author_email="numba-users@continuum.io
    • Stuart and Val both agree and no one is opposed

  • Daily task of maintainers presentation

    • TODO: (needs link to slide)
    • Ethan P: Plan for where user contributions should go, what can they help with?
    • Siu: Not much time lately, long term discussions taking place.
    • Stuart: It's been a 6-9 months like no other, huge code changes and a lot of growth in use and user contributions.
    • Val: Good things to do that are easy include going through the issue tracker and marking non-reproducing issues, flake8 fixing etc.
    • Lucio (on chat): I agree that increasing resources to share the load is important
    • Hameer: Could expert NumPy and Python users help by reviewing those PRs which are accessible.
    • Siu: PRs can be reviewed with respect to just sorting out e.g. Python things.
    • Lucio: what do you expect or would like to see in terms of community PR reviews? People just jumping in whenever they feel like? Or would like some order/system to it?
    • Stuart/Siu: any order is fine, not mutex until needed.
    • Ethan P: If I find a bug and I want to fix it, how can I get help from a core dev?
    • Siu: Prefers discourse
    • Val: Gitter for real time
    • Ethan P: what about weekends?
    • Stu: Opening an issue and following up on Gitter is perhaps a good strategy to enlist help and get guidance. But if it's a bug, it goes on the issue tracker!
    • Ethan P: Can CONTRIBUTING.md be updated on how to actually do stuff and stuff that needs doing?
    • ACTION: Core devs, go do that ^

  • Themed Discussion

    • What’s “the one missing feature” that you really want?!
      • Ethan P: Python parity! Documentation on what works/doesn't.
      • Siu: Want to speed up a subset of python, focus was on array. Stepped into other containers. Fear of python parity road, should this fear be reevaluated. Pragmatic focus on specialised problem and making that very fast.
      • Lucio (on chat): there's a whole analysis by a core CPython dev (forgot his name) that reaches the conclusion that full python semantics cannot be massively accelerated (referring to https://faster-cpython.readthedocs.io/fat_python.html)
      • Ethan P: Will open issue to clarify approaching compilation from a more Pythonic approach
      • Siu: Core devs can focus on making extensions to Numba easy, most complexity is actually in the core compiler. Means contributions can live outside Numba as extension packages, e.g. Awkward, SDC.
      • Hameer: +1 on making extensions easy. Are the QoL developments for extension writers in general, profiling and debugging.
      • Lucio (on chat): I think making extension types easy to create, and potentially one and the same with jitclasses, would enable a lot of growth in usage
      • Stu: extensions, profiling and AOT are mine.
      • Siu: also sees expansion of extensions as a good place to go.
      • ACTION: Continue on discourse thread in more details and see where it brings us (maybe continue next time if needed)**
    • Where should Numba be in 5 years time?
      • EDIT: did not get around to talking about this, perhaps next time.

1. New Issues

  • #5979 - Incorrect results from (deterministic) prange loop
  • #5977 - Equality tests on floats: add support for math.isclose() and numpy.isclose()
  • #5976 - Memory leak on trivial compilation
  • #5974 - ModuleNotFoundError: No module named 'numba.decorators'
  • #5973 - Segfault in TBB, maybe parfors related.
  • #5970 - Add note to docs about boundschecking being off by default and how to switch it on
  • #5968 - Numba 0.50.1 downloaded from conda fails to compile on GPU (but works when using pip)
  • #5967 - Typed list: Unable to add empty fortran array
  • #5965 - custom logging level for numba internals.
  • #5964 - Type inference problem with literal unroll

Closed Issues

  • #5966 - 0.50.1 missing source distribution on PyPI
  • #5961 - invalid number of args for optional arguments of jitclass methods.
  • #5957 - I can't install numba on rpi 4b

2. New PRs

  • #5981 - Document bounds-checking behavior in python deviations page
  • #5980 - Deprecate target kwarg for numba.jit.
  • #5978 - Add structref to as a mutable struct that is pass-by-ref
  • #5975 - restructure communication section
  • #5971 - Typecode cache for stable typecodes
  • #5969 - test Numba on the buildfarm with LLVM 10 enabled llvmlite
  • #5962 - Fix SVML tests with LLVM 10 and AVX512
  • #5960 - Setting mypy on CI
  • #5959 - Making typed.Dict a typing MutableMapping
  • #5958 - Making typed.List a typing Generic

Closed PRs

  • #5972 - fix flake8 for numba/runtests.py
  • #5963 - Fix broken link to numpy ufunc signature docs
  • #5956 - Update numba_sysinfo.py

3. Next Release: Version 0.51.0, RC=22 July, Final 29 July?

  • Requests for 0.51

  • high risk stuff for 0.51.

  • 0.51 potential tasks (To be updated)

4. Upcoming tasks

  • Opening up the numba meeting
Add a custom footer
Add a custom sidebar
Clone this wiki locally