Minutes_2021_05_18

Numba Meeting: 2021-05-18

Attendees: Jim Pivarski, brandon willard, Graham Markall, Guilherme, Todd A. Anderson, Siu Kwan Lam, stuart, Valentin Haenel

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

Numpy 1.20 and dropping older numpy and maybe 32-bit platforms
- win32 testing by Awkward
- win32 is more common on cloud CI
- linux32 is harder to get on cloud CI
Stale PR bot / action?
- issues from previous investigation
  - GH query filtering becomes unreliable when number of issues/labels is large.
  - GH ticket https://support.github.com/ticket/personal/0/834117
- Will give it a try and see how well it works out
Numba youtube channel for developer tutorials:
- https://www.youtube.com/channel/UCLfCWffD2nEejc_WpISl95A
Numba-backend in Aesara https://github.com/pymc-devs/aesara/releases/tag/rel-2.0.8

#7034 - DEBUG_JIT is undocumented and only works for the CPU target
#7033 - Unsupported op-code encountered: list_to_tuple (only with Python 3.9)
#7032 - Slicing UniTuple in a loop fails
#7029 - .view() for NumPy types broken if the view type is a complex type
#7028 - memory leak caused by racing conditions in parallel loop
#7023 - Using operator module as mapping fails when compiling
- bug/feature request, literalstrkeydict is intercepting code too early.
#7022 - ir not imported
- consider getting rid of it.
#7014 - [BUG] cannot pickle generator when there is only one function that uses input

#7031 - Fix inliner to use a single scope for all blocks
#7030 - Fixes missing typing_extensions dependency on gitpython
#7027 - Fix nrt debug print
#7025 - Add missing vars into scope during SSA rewrite
#7024 - fixed llvmir usage in create_module method
#7021 - Debug cache location in failing test **** - #7019 - Support NumPy 1.20
- min numpy 1.17
- 32-bit platform missing conda packages for newer numpy