Skip to content
Valentin Haenel edited this page Sep 22, 2021 · 1 revision

Numba Meeting: 2021-09-21

Attendees: Siu Kwan Lam, stuart, Valentin Haenel, brandon willard, Guilherme, Luk, Nick Riasanovsky, Shi Fan, Todd Andeson

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

0. Feature Discussion

  • 0.54.1
  • Inline closure SSA problem/regression
    • plan for 0.55
    • to fix with LLVM-style prefixing function name to inlined variable name
  • default branch rename
    • wait after 0.54.1 tagged
    • stuart + val tiger team
  • Val talks about numba changes for Python3.10

1. New Issues

  • #7399 - Allow getting pointer (address) to Numba's int/float variables
  • #7400 - Allow pyobject annotation on outgoing variables in objmode
  • #7402 - guvectorize does not implement the numpy ufunc interface in versions 0.53+
  • #7406 - Support for Numpy convolve mode 'valid'
  • #7409 - numba.vectorize interface inconsistency between target="cpu" and target="parallel"
  • #7410 - Support for Record type in jitclass
  • #7411 - jit(locals) are propagated to inline functions with same variable names

Closed Issues

  • #7396 - if statement inside a for loop fails
  • #7405 - Caching Silently Fails with Omitted Arguments
  • #7407 - Untyped global name 'dblquad': Cannot determine Numba type of <class 'function'> : Scipy dblquad not work in Numba
  • #7408 - Type inference failing when using a heapq of namedtuples

2. New PRs

  • #7397 - Add option to catch only Numba numba.core.errors derived exceptions.
  • #7398 - Add support for arrayanalysis of tuple args.
  • #7401 - FEAT: add np.moveaxis support
  • #7403 - Fix for issue 7402

Closed PRs

  • #7404 - fix typo in literal_unroll docs

3. Next Release: Version 0.55.0/0.38.0, RC=Early Nov

Clone this wiki locally