Minutes_2021_11_23

Numba Meeting: 2021-11-23

Attendees: Siu Kwan Lam, Graham Markall, Val, Guilherme Leobas, Nick Riansanovsky, Todd A Anderson, stuart, brandon willard, Luk, Kaustubh Chaudhari

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

llvmlite 0.38 RC plan
- Plan to tag thursday + build packages
- Conda pkgs for Friday, wheels slightly later
- Python 3.10 support merged
- All targets now supported
- Release tracking: https://github.com/numba/llvmlite/issues/792
Python 3.10 update (Numba)
- with-statement detection: seems to be OK
- aim to merge towards the end of the week
- Create RC: aim for Dec 1st
object mode caching
- https://github.com/numba/numba/issues/7572
- https://github.com/numba/numba/pull/3884 has important tests to check the func_ir.copy() related issues.
llvmlite RISC-V merge + PRs
- https://github.com/numba/llvmlite/pull/793
- https://github.com/numba/llvmlite/pull/775
- Packages moved to numba channel, 775 rebuild started
- Keep CI pointing at main because that's where we keep the known good llvmdev.
Added overload methods for Numpy's ufunc.reduce #7524

#7566 - Support nested tuples as tuples of literal elements
#7568 - Using literally suppresses info in compilation failure traceback
#7569 - for-with-break construct fails on Pythons 3.7 and 3.8
#7571 - Documentation Cuda example wrong indentation
#7572 - Object mode code fails to load from cache
- Yes please PR
#7573 - numba 0.54.1 casting IntEnum regression
#7574 - Add support for non-constant boundaries for numba.stencil
#7575 - cache missed for str arguments
#7576 - test_no_accidental_warnings (numba.tests.test_import.TestNumbaImport) fails on Python 3.9 Windows
- PyWin32 latest version doesn't use imp
#7577 - values in a list become odd in nopython mode and no warning or error reported
- Probably not to be fixed for reflected lists - mayybe support with reflected lists
- Typed list can be either of type None or something else, but not mixing None and others
#7578 - Compilation with parallel=True gives garbage results for a very simply parallelizable function
#7579 - CUDA: Lowering error using character sequences in record types
#7585 - Support for len(dict.key()), len(dict.values()), len(dict.items())

Request for 0.55
- broadcast_to https://github.com/numba/numba/pull/7119
  - driven by need for vectorize support in aesara
  - Now merged