Minutes_2019_02_26

Attendees: Stuart, Stan, Siu, Ehsan, Todd

0. Feature Discussion

• Ready for RC?
  • Dict PRs?
    • WIP + In Review
    • Segfaulting w/ refct
  • BuildFarm issues?
    • Python 3.5 windows VC failing _dictobject.c
  • Resolution:
    • Bump RC to next week
    • Tried to merge PRs if they don't cause new failures in the farm and make target-specific patches to resolve the problems
    • Good to have PRs:
      • IPyParallel patch

1. New issues

• #3804 - Add support for np.matmul
  • good first issue
• #3803 - LoweringError: Failed in object mode pipeline (step: object mode backend)
  • object mode fail
  • would have worked in nopython mode if matmul existed
• #3802 - (Auto?!)Generate wrappers for call convention change
  • need wrappers to help
  • docs explaining calling convention
• #3801 - Please report the error message and traceback, along with a minimal reproducer
  • unusable bug report
  • need to close
• #3798 - Simplify and expose maximum register use in code in cuda.kernel
  • can do some of the suggestions in this issue
  • can't assign registers to source lines, because that's not how the compiler pipeline works

• #3797 - Move _nonoptional a common intrinsic

• #3795 - Cuda.jit does not seem to respect max_registers keyword
  • possibly related to #3798
• #3793 - send method of generators
  • Needs actual coroutine support
  • Need to investigate how LLVM supports coroutines
• #3790 - Expose all the numba env var controls through command line arguments
  • should look at how dask does this
  • need to put info into help string in numba CLI
• #3789 - isnan(int) isinf(int) should work
  • works in both Python and NumPy
  • hard to write code that accepts both ints and floats while checking nans
  • good first issue
• #3788 - Call to cuMemcpyDtoH results in CUDA_ERROR_LAUNCH_FAILED
  • may be answered on mailing list
• #3787 - Numba vectorize take 2d array
  • already answered
• #3784 - Feature request: implement range as a type
  • need transformations to transform iterators into more efficient form, otherwise every for loop will get slower
  • also have problems passing around iterators as arguments
• #3781 - Failure when trying to use a JitClass
  • jitclass instances as globals don't work in other njit functions.
  • can improve error message
• #3780 - Failure when trying to use return value of yield
  • coroutines not supported
  • also overlaps with #3779
• #3779 - Multiple return types of yield are not detected if one of them is None
  • #3782 prevents this case from being allowed
  • lowering can't handle optional yield type currently
  • should be possible to support like return

Already Closed

• #3805 - Numba compilation error
• #3794 - Problem with cuda.jit ptx?
• #3786 - Writing to local graphics memory causes llvm error: Invalid user of intrinsic instruction!
• #3783 - Numba problem in for loop
• #3778 - Compiled code returns wrong values or crashes

2. Open PRs

• #3800 - Bugfix for np.interp
  • almost ready to merge once author fixes some issues
• #3799 - Support refcount'ed types in numba dict
  • WIP, segfaults to fix
• #3796 - DO NOT MERGE: Add in the c3i_test channel and DLL mod
  • can close this
• #3792 - Fix non-module object used as the module of a function.
  • needs minor test fix
  • probably ready to review after that
• #3791 - Implement interpreter-side interface for numba dict
  • ready to review, but pending merge of #3777
  • need __repr__ and __str__
• #3785 - Clarify use of range as function only.
  • easy to review
• #3782 - Raise typing error on yield optional.
  • ready to review

Old Active

See https://github.com/numba/numba/projects/10

===========================

4. Next Release: Version 0.43, RC=March 5?, Final=March 12?

• Initial dictionary support
• ???