Minutes_2020_12_01
Valentin Haenel edited this page Dec 2, 2020
·
1 revision
Attendees: Siu, Stu, Todd, Guilherme, Val, Hameer
- ship 0.52
- open meeting next week, topics!?
- Siu will announce today asking for topics.
- Jim(awkward), Luk suggest a way to re-use Numba typing/codegen for a different array implementation
- Maybe related to https://github.com/numba/numba/pull/6148
- Todd asks if anyone is doing Sparse array in Numba
- Stuart has a local experiment on sparse arrays.
- Hameer has some experience. Higher dim sparse array is hard to impl in numba; i.e. iterators.
- set up numba-extras repo
- use GH actions
- automate pypi/conda publishing
pytest
- maybe
black
- Hameer suggests we pin
black
version to avoid churn
- Hameer suggests we pin
- type annotation would need discussion (community to implement setup etc)
- use GH actions
- python 3.9
- more bytecode changes than we thought (late doc update)
- Stuart writing a peephole opt to compress bytecode instructions 😨
- more likely to be 0.53 with so much changes
- Code owner review ping #6502
- JAX-numba bridge https://numba.discourse.group/t/calling-numba-from-jax/372
-
#6526 - Wrong results using Numba and astype() compared to pure Numpy
- impl needs fixing
-
#6523 - typed.List.extend returns self in some cases
- impl needs fixing, PR at #6524
-
#6522 - Caching a function that takes in a jitclass instance
- known jitclass problem
- #6520 - typeguard fail for numba.tests.test_jitclasses.TestJitClass
- #6518 - LiteralStrKeyDict error for non-identifier key names
- #6525 - inconsistency in assignment to global variables
- #6527 - setup.py: fix py version guard
- #6524 - Fix typed.List return value
- #6521 - Skip annotated jitclass test if typeguard is running.
- #6515 - CUDA: Add support for the jit_module function
- #6514 - Catch user code with compound type use and offer advice.
- #6519 - Avoid unnecessary ir.Del generation and removal
- #6517 - CUDA: Add NVVM test of all 8-bit characters
- #6516 - Fix RTD build by using conda.
- Relase branch: mid-Jan