Minutes_2022_02_08

Numba Meeting: 2022-02-08

Attendees: Siu Kwan Lam, Val, Benjamin Graham, Graham Markall, Nick Riasanovsky, Todd A. Anderson (Intel Labs), Ehsan Totoni, Guilherme Leobas, Luk, brandon willard, Kaustubh Chaudhari, Vladimir Lukianov

NOTE: All communication is subject to the Numba Code of Conduct.

Please refer to this calendar for the next meeting date.

0. Feature Discussion

• Short leftover discussions from last week:
  • #7797 - prune README
• Numba vision discussion
  • Bodo/Ehsan:
    • Uses Numba as part of a high-performance data science platform.
    • Numba is a very critical component: 1) JIT is the main user interface; 2) Numba is used as a compiler toolkit and for implementing kernels (in addition to C++)
  • monkey-patch if contribution not feasible
  • would like to contribute more
  • Vision:
    • lower the barrier to performance for python devs
    • a compiler toolkit for python
  • Requirements:
    • Comprehensive coverage of Python basics
    • Simple and intuitive errors
      • current errors are scary to users
    • Compiler development features
      • Bodo has Dataframe support that are very complicated
      • Make numba a better
    • Production qualify code base
      • so new contributors can be effective earlier
  • who do we target?
    • compute-focused?
    • everyone? (web developers; i.e. async)
    • IntelPython/ToddA: target basic users; advance users have more options
  • Luk: Julia is closest to Numba. Questions why Numba cannot do what Julia can.
  • IntelPython/ToddA: echo Compiler-toolkit. Support for new backends. Perhaps the transition to MLIR will help with supporting in HW.
  • Aesara/Brandon: New HW backend is important. Want to use Numba as the single point for support different backends.
  • Luk: Numba is a small team, small project trying handle what a typical language/compiler project tries to do.
  • Val:
    • Anarcho Utilitarianism https://gist.github.com/esc/a41f52139e3338e00dc17c183de4ba89
  • Questions:
    • What is the core? How to decide what should be in the core?
    • How can we make it easier for people to contribute?

1. New Issues

• #7806 - CUDA: Division by zero stops the kernel
• #7807 - pickle issue
• #7808 - has no attribute class_type when running NUMBA_DISABLE_JIT
• #7811 - Assignment of jitclass-owned array fails when list comprehension is used in the calling function
• #7812 - Providing many keyword arguments triggers assertion error with Python 3.10 due to unsupported bytecode
• #7816 - Python sets as arguments to jitted functions

Closed Issues

• #7810 - Documentation error

2. New PRs

• #7805 - Enhance source line finding logic for debuginfo
• #7813 - Extend parfors test timeout for aarch64.
• #7814 - CUDA Dispatcher refactor
• #7815 - CUDA Dispatcher refactor 2: inherit from dispatcher.Dispatcher
• #7817 - Update intersphinx URLs for NumPy and llvmlite.

Closed PRs

• #7809 - Updates the gdb configuration to accept a binary name or a path.

3. Next Release: Version 0.56.0/0.39.0, RC,