Minutes_2020_03_10

Numba Meeting: 2020-03-10

Attendees: Graham, Pearu, Stuart, Val, Siu, Aaron, Todd, Gui

#5162 - is it converging?
- still fanning out
- Graham will try to summarize the ideas in there
- Move to 0.50
#5144 - status?
- delete the problematic test
state of jitclass
- Q: is makeover was planned for 2019 didn't happen, is it in the plan for 2020? Is there any value in spending time trying to do new stuff/improvements? Are they going to be entirely re-done?
  - Q3 at earliest.
  - Don't burn lots of time.
  - Might need a lot of rewrite.
  - Might want to re-think what it is for
first class function
- target to finish this week
- discard the first PR.
proposal 1: burndown behaviour/release cycle
- 4 dev weeks + 2 weeks burndown
- dev weeks:
  - works on high risk features
  - have kickoff at the start of cycle
    - declare the high-risk features which of us will be doing
    - review these features
- burndown weeks:
  - no major features
  - focus on bug fixing
- No objections.
proposal 2: triager? issues czar?
- start with try out period to help define this role.

#5367 - SimpleIteratorType issues
#5365 - Type inference for integers is odd
#5364 - RuntimeError: missing environment when iterating over closure in intrinsic
- odd; needs digging
#5363 - Make is really obvious that the "dev" docs are for unrelease/development versions
- doc
#5358 - Support 0d arrays in cuda
- ~~TODO: Siu~~
#5355 - typed.List's empty_list function cannot accept allocated parameter in nopython mode
- fix in #5356
#5354 - numpy.take cannot compile with typed.List as indices parameter
- relates to reflected list deprecation
#5353 - typed.List variables cannot be assigned from reflected_list
#5350 - yield statement yields leaky memory
- confirmed memory leak
#5349 - Parfors, LLVM IR Parsing error from conditional needing cast.
#5344 - literal_unroll != numba.literal_unroll?
- the common getattr thing
#5339 - Add a as_constant helper for forcing computations to compile-time

#5366 - Remove reference to Python 2.7 in install check output
#5361 - 5081 typed.dict
#5360 - Remove Travis CI badge.
#5359 - Remove special-casing of 0d arrays
#5357 - Small fix to have llvm and asm highlighted properly
#5356 - implement allocated parameter in njit
**** #5352 - Add shim to accommodate refactoring.
- integration tests are mostly passing
- will be announcing to get developers to get it tested
**** #5351 - SSA again
- Pass strips SSA just prior to lowering to match the strict semantic requirements for LLVM
- SSA can be constructed and stripped manually, good for testing/R&D.
- State now is slowly solving bugs in impl.
- Need to see what happens in parfor lowering.
- SSA labelled issues, add as tests.
- Parfors would prefer not to have SSA-reconstructor by its passes
- Warn/validate on the event of reconstruction that cause violations, aim for full SSA in 0.50.
  - plan it!
#5348 - Add support for inferring the types of object arrays
#5347 - CUDA: Provide more stream constructors
**** #5346 - Docs: use sphinx_rtd_theme
- Looks great!
#5343 - Fix cuda spoof
** #5342 - Support for tuples passed to parfors.
#5341 - Misc/testpr5172
- testing only
#5340 - [WIP] Package serialization functions into a pickleable class

Requests for 0.49
- ** #5189 (__cuda_array_interface__ not requiring context)?
- TBD
Future Relase Plan
1. LLVM9
2. Explore LLVM pass for refcount pruning