Minutes_2022_03_01

Numba Meeting: 2022-03-01

Attendees: Siu Kwan Lam, Graham Markall, Guilherme Leobas, Benjamin Graham, brandon willard, Nick Riasanovsky, Todd A. Anderson, stuart, Ehsan Totoni, Jim Pivarski

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

Please refer to this calendar for the next meeting date.

0. Feature Discussion

• Location of CUDA examples in docs
  • Issue: https://github.com/numba/numba/issues/4266
  • Notebook: https://gist.github.com/brandon-b-miller/68bdb7313d1f07d00b0f398e9d795d3b
  • To investigate whether to include jupyter notebooks into the RTD manual
    • tools:
      • Jupytext is a collection of formats that represent unevaluated Jupyter notebooks either as Markdown with special syntax to indicate breaks between cells and where teh code cells are or as Python with syntax in the Python comments that indicate breaks between cells. This could be a good way to generate notebooks for JupyterBook based on code that has already been tested by CI/pytest. That gets around an issue that it's difficult to catch errors when JupyterBook evaluates the notebooks (start with already-tested code).
      • JupyterBook turns a collection of Jupyter notebooks (or Jupytext) into reST and has Sphinx make a documentation site from it. I (Jim) don't know if it's possible to just make the reST and pass it into an existing Sphinx toolchain. The problem with JupyterBook is that they keep changing their YAML format for representing tables of contents. Pin to a version!
      • Nbsphinx: I (Jim) don't know about this; it might be used by JupyterBook.
• Numba Vision discussion after normal meeting

1. New Issues

• #7854 - prange seems to exit for iterator before all threads finish. Do I missing some code like join()?
• #7856 - Some random distributions do not accept size=None arguments
• ** #7857 - Using numba and julia (or any other LLVM loader?) at the same time leads to segfaults
  • Stuart suggests a runtime check on LLVM version if a libLLVM.so is in-process
• #7859 - Incorrect behavior with numpy array equality against a scalar unicode string
• #7862 - Wrong results with prange on Linux with Numba 0.55.1
• ** #7863 - Multiple ConcreteTemplates should not be considered separately
• #7867 - Prange incorrectly compute ifloordiv reduction
• #7868 - Numba incorrectly interpreting generator function
• #7869 - Strange Behaviour on Advanced Indexing
  • improve error message to say which element in the index tuple will cause advanced indexing

Closed Issues

• #7853 - numpy.argmin(arr, axis=int64), where (arr.ndim > 1), returns passes through the dtype of arr

2. New PRs

• #7858 - CUDA: Deprecate ptx Attribute and Update Tests
• **** #7860 - Path finding for JIT compiling CUDA launch sites
• #7864 - Fix cross_iter_dep check.
• #7865 - Remove add_user_function
• #7866 - Support for large numbers of args/kws with Python 3.10

Closed PRs

• #7855 - Ensure np.argmin/no.argmax return type is intp
• #7861 - Fix a spelling mistake in README

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