Minutes_2019_03_12

Attendees: Stan, Siu, Ehsan, Stuart

0. Feature Discussion

0.43.0 release
- Add note to release notes indicating known issues with NumPy 1.16
- action:
  - stuart open PR to note the above + xfail/skip related tests
  - fix setup script (see https://github.com/numba/numba/issues/3846)
llvmlite https://github.com/numba/llvmlite/pull/466
- opens up opportunity for LLVM pass in python
- LLVM IR manipulation API and lifetime tracking
complex type returns from C-ABI for scipy functions

#3855 - Extending and strengthening numba.extending.overload
#3854 - Support np.delete
#3851 - Rewrite of refcount pruning pass w/o regex
#3848 - LoweringError after NotImplementedError
#3846 - Install fails on clean Python 3.5 environment
- Add Python 3.5 back into wheel builds
- Fix setup.py to note that numpy is a build requirement
#3844 - "data type not understood" with readonly array argument
- already fixed
#3841 - Passing *args to function calls fails when parallel=True
- gufunc issue with tuples
#3840 - numba with cudatoolkit on Windows not finding libraries
- we think this is already fixed, no response from user
#3839 - Switch to LLVM 8 backend
- Delayed because LLVM 8 has another RC
#3838 - Add or document signature syntax for zero dimensional arrays
- Need to think harder about how zero-dimensional arrays function before making signature easier to type
- Should document workaround in FAQ
#3836 - Unicode String iteration
- see PR
#3835 - Support list() ctor for unicode types
- see PR (same as above)
#3834 - support __mul__ for strings
- Good "first" issue

#3849 - Python UnboundLocalError not recognized by numba
#3843 - Numba 0.42.1 on Linux : 'No threading layer could be loaded' when config.THREADING_LAYER = 'omp'
#3833 - Numba 0.43 RC checklist

#3853 - Fix some minor issues in the C source.
- failing some things in CI still
#3852 - Continuation of PR #2894
- finishing out jitclass default constructor arguments
#3850 - WIP: refactor np.interp
- trying to match what NumPy does better
#3847 - Allow fine-grained control of fastmath flags to partially address #2923
- Need to do more research on how this works in clang
- need to check that SVML is still OK
#3845 - Fix mutable flag transmission in .astype
- ready to merge
#3842 - Unicode type iteration
- will mark ready to merge
- Need better API for iterators