Minutes_2023_12_12

Numba Meeting: 2023-12-12

Attendees: FPOC (last week): Graham FPOC (incoming): Graham until 15 Dec

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

Please refer to this calendar for the next meeting date.

0. Discussion

• Plan for 0.59:
  • Move the reflected list part of the test in numba.tests.test_print.TestPrint.test_print_values into it's own test, apply a @skipTest, file a ticket to debug and fix it.
    • Stuart's view is that this is probably a localised latent problem that compile_isolated was managing to hide, an issue either with reflected list or print (and print is already known to have issues with the typed containers, reflected list also has known bugs!), this gets main green again.
    • Consensus is that this is the way to go.
  • Take a view on https://github.com/numba/numba/pull/9353#issuecomment-1850547561
    • Stuart's view: this test should be skipped and a ticket should be filed, closed by #9259.
    • Fixing the performance of @overload for one test is not really something to attempt this late in the release cycle, and the shipped compiler won't be any slower than it is now, the problem has been present in @overload since its inception.
    • Consensus is also that this is the way to go.
  • Get #9352 into reasonable shape.
    • Separate branch to be pushed up to examine some changes in detail.
  • Cut RC.
• Plan for llvmlite 0.42 (Graham)?
  • Cut the RC - no further action needed
  • Release note writeup may be needed.
  • Run the PR script to generate the list.
• LLVM plans (Graham):
  • 15 / 16 / 17
  • Opaque pointers
  • Legacy pass manager
  • OrcJIT

New "Ready for Review" PRs

ISSUES/PR TBD

1. New Issues

• numba#9339 - Allow dynamic gufuncs with target="parallel"?
• numba#9340 - about the shape argument for cuda.shared.array() and literal types inside the definitin of a kernel
• numba#9344 - Fix exception message when the number of arguments does not match in a gufunc
• numba#9345 - Support calling gufunc without output buffers
• numba#9348 - Random cache corruption
• numba#9350 - FEATURE REQUEST: enable more fine grained control over Numba cache

Closed Issues

• numba#9343 - Error : Inconsistency with supported architecture with llvmlite

2. New PRs

• numba#9338 - Add np.in1d, np.isin, np.setxor1d, np.setdiff1d, extend np.intersect1d.
  • Need to balance prioritisation between NumPy 2.0 and adding NumPy functions, and communicate accordingly
• numba#9346 - Setting NUMBA_CAPTURED_ERRORS=old_style will now raise warnings.
• numba#9347 - Make Numba's deprecation warnings subclasses of the builtin ones.
  • Potentially related: https://numba.discourse.group/t/is-there-a-way-to-resolve-the-multiprocessing-error-from-aws-lambda-with-numba/2322
• numba#9351 - Made Python 3.12 support rst note more verbose
• numba#9352 - Removing object mode fallback from @jit.
• numba#9353 - Remove numba.generated_jit
• numba#9354 - Alter the test_print.py test to match previous.

Closed PRs

• merged - numba#9334 - Fix parfor reduction issue with Python 3.12.
• merged - numba#9335 - Add validation capability for user generated towncrier .rst files.
• merged - numba#9336 - Add TargetLibraryInfo pass to CPU LLVM pipeline.
• merged - numba#9337 - Revert #8583 which skip tests due to M1 RuntimeDyLd Assertion error
• merged - numba#9341 - Add configuration variable to force llvmlite memory manager on / off
• merged - numba#9342 - Fix flake8 checks for v6.1.0
• numba#9349 - [CI only] Pr/9330
• merged - llvmlite#1019 - Add convergent as a supported FunctionAttribute and CallInstrAttribute.
• llvmlite#1020 - Aarch64memorymanager ppc debug

(last numba: 9354; llvmlite 1020)

3. Short-term Roadmap

gantt: https://github.com/numba/numba/issues/8971