Minutes_2023_05_09

Numba Meeting: 2023-05-09

Attendees: Siu Kwan Lam, Val, Graham Markall, Andre Masella, brandon willard, Guilherme, Ianna Osborne, Luk, Michael C, Mingjie Wang, mmurray, stuart, Todd A. Anderson,
FPOC (last week): Graham FPOC (incoming): Guilherme (swapped)

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

Please refer to this calendar for the next meeting date.

0. Discussion

• 0.57.1 patch fix. Considerations:
  • issue https://github.com/numba/numba/issues/8939 - PR https://github.com/numba/numba/pull/8948
  • https://github.com/numba/numba/issues/8940
  • https://github.com/numba/numba/issues/8936 and PR https://github.com/numba/numba/pull/8947
  • https://github.com/numba/numba/pull/8895 that fixes https://github.com/numba/numba/issues/8888
    • OK to go in pending high confidence in the fix, from testing with RAPIDS and other libraries that make heavy use of the CUDA target
    • To report back next week
  • https://github.com/numba/numba/issues/8949 - another NumPy function type regression from the overload_glue migration
• 0.58 schedule / timeframe / plans
  • Potential draft roadmap thru EOY for discussion / comments:
    • gantt
  • Schedule for Numba dependencies releases
• Some small but not too hard things to do:
  1. wire in loc= to all the internal exception raises.
     • Siu: would be good to show the last user-source loc
  2. add in snprintf use to "IndexError" error messages to report the actual index that's OOB at runtime.
     • Dynamic exception argument is probably easier than snprintf
  • Github issues to be created if someone would like to volunteer to pick them up.
• https://numba.discourse.group/t/how-to-spell-type-signatures-for-default-argument/1922
  • How to spell signatures with default values
  • Numba currently uses Omitted(*) to type optional arguments. Proposed solution is to use optional(*), which seems more natural. Opinions?
    • Luk: Proposed solution seems more natural.
    • Stuart: Agreed with Luk.
  • Siu will open an issue for this feature request to be implemented.

New "Ready for Review" PRs

• compiler-rt https://github.com/numba/llvmlite/pull/909

1. New Issues

• numba#8936 - [0.57.1]Use of isinstance() error in numpy.where
  • Need to check if any other internal uses of experimental things
  • https://github.com/numba/numba/pull/8947 preferable fix
• numba#8939 - [0.57.1] Regression with literal_unroll in 0.57.0
  • Fix: https://github.com/numba/numba/pull/8948
  • Need to check fix with two loops next to each other to make sure we don't get name collisions in the same scope
• numba#8940 - [0.57.1] Regression: NameError: name 'unsafe_empty_inferred' is not defined
  • Potential fix: CPU Context registry needs to import the unsafe array module so that the intrinsic will run by virtue of the import
  • May need adding to CUDA
• numba#8941 - Checklist of deferred items from Advanced Indexing PRs
• numba#8942 - Support set() in numba typed dictionaries?
  • may work with typed set
• numba#8943 - zip(..., strict=True) not supported yet
• numba#8945 - Support numpy.frombuffer keyword arguments: count and offset
• numba#8946 - GPU support for jitclass
• numba#8949 - [0.57.1]np.angle returns incorrect type since 0.57.0
• llvmlite#941 - Test failure in 0.40.0 on 64-bit PowerPC: test_get_process_triple
  • Triple.cpp: https://sourcegraph.com/github.com/llvm/llvm-project/-/blob/llvm/lib/TargetParser/Triple.cpp

Closed Issues

• numba#8938 - numba accelerated numpy.argsort gives different result to numpy

2. New PRs

• numba#8937 - Remove old Website development documentation
• numba#8944 - Add exceptions to np.isclose
• numba#8947 - Introduce internal _isinstance_no_warn
• numba#8948 - Fix issue #8939
• llvmlite#942 - Add ORCJITv2 support

Closed PRs