Minutes_2020_05_05

Numba Meeting: 2020-05-05

Attendees: Todd, Siu, Hameer, Aaron, Stuart, Guilherme, Pearu

0. Feature Discussion

• numba 0.49.1 and llvmlite 0.32.1

  • numba RC1 released last Friday with wheels on pypi
  • Docfix needed Numba: #5669
  • Release planned for Thurs
  • conda-forge builds is lack behind (still at 0.48)

• Exceptions (update?)

  • Related PR for heap tracking https://github.com/numba/numba/pull/4517

• numba.types a public API?

  • many examples, release notebooks are broken
  • Add for 0.49.1 final? Yes

• Rewrite the inliner

  • current implementation is fragile; very sensitive to specific options.
  • Stuart is rewriting the inliner
    • WIP PR https://github.com/numba/numba/pull/5673

• May 14 is start of Burndown

• discussion on function subtyping

  • part discussion on merits of producer/consumer declaring
  • also on whether something more dynamic could be made work
  • also on motivation
  • TODO: outcome, discuss again next week, more people to investigate:
    • gist: https://gist.github.com/luk-f-a/77fb61b25cadbc497337ab9b53b1d0c9
    • PR: https://github.com/numba/numba/pull/5579

1. New Issues

• #5665 - No error for jit(nopython=True, forceobj=True)
• #5664 - Cythonizing a module containing numba functions
  • need example
• #5662 - Unexpected behavior taking the minimum of an array of np.inf when parallel=True.
  • inf larger than anything
• #5661 - numba 0.49 crashes when parameter is either number or None
  • SSA
• **** #5655 - Respect PEP 3102 keyword-only arguments in nopython mode
  • Related discussion: Use keyword-only for Numba source code
    • the decorators really need to use keyword-only args
    • consider it in public-APIs and internal funcs with a lot of arguments
• #5654 - can not use numpy.max as funtion parameter
  • likely duplicate
• #5650 - Create integer from unicode string with int()
  • val has a workaround on gitter
• #5643 - Issue in a njit function when determining types
  • types.* are public API but lack public API docs and errmsg
• ** #5641 - About enabling jit-decorated functions as first-class functions when using in sequences
  • Pearu's
• #5640 - guvectorize-CUDA broadcast fail
• #5637 - Cannot unify array vstack
  • type unstable
• #5635 - Unable to create set from unicode containing numba.typed.List
  • multiple problems
  • related PR: https://github.com/numba/numba/pull/5639

llvmlite:

• 587 - Importing llvmlite on Github Actions MacOS crashes Python

Closed Issues

• #5663 - TypeError vectorized assignment in ´prange` loop with if statements
• #5656 - Numba compiled code prints RuntimeWarning where there should be none
• #5649 - Anaconda 3 and PowerBI Connection Error
• #5647 - Unable to install numba - any help much appreciated!
• #5646 - Numba njit restarts my IPython kernel
• #5644 - UnsupportedError: Failed in nopython mode pipeline (step: analyzing bytecode) Use of unsupported opcode (IMPORT_NAME) found
• #5642 - Could numba support big integers for a * b?
• #5636 - Traceback (most recent call last): File "C:\Users\dell\AppData\Local\Programs\Python\Python36\lib\site-packages\numba\core\typeconv\typeconv.py", line 4, in <module> from numba.core.typeconv import _typeconv ImportError: DLL load failed: The specified module could not be found.
• #5634 - I am getting when importing numba in python3.7.2.Please provide solution for windows.
• #5633 - Parallel example function error
• #5632 - ImportError: Numba could not be imported.
• #5631 - ImportError: Numba could not be imported.

2. New PRs

• **** #5660 - Make coverage.py work with compiled code.
  • emits/faking a python line event for the interpreter
  • maybe very slow
  • many potential alternative or extensions using this technique
• #5659 - Add support for parfors creating 'F'ortran layout Numpy arrays.
• #5653 - Update doc with respect to try-finally case
• #5645 - Fix SSA redundant CFG computation
• #5639 - Make UnicodeType inherit from Hashable

Closed PRs

• #5658 - Explicit bail out in unified_function_type
• #5657 - Update CHANGE_LOG for 0.49.1
• #5652 - Pin to sphinx=2.4.4 to avoid problem with C declaration
• #5651 - Fix some SSA issues
• #5648 - Unset the memory manager after EMM Plugin tests
• #5638 - Make literal_unroll function work as a freevar.

3. Next Release: Version 0.50.0, RC=May 28th

• Requests for 0.50

• high risk stuff for 0.50.

  • Declarative typing (Siu and Stu)
    • declaring accepted types in things like @overload
    • so we can have better errmsg
    • allow overloading "contextual information"; i.e. targets (CPU, GPU), fastmath flag, exception handling
    • make sure we don't break existing @overload/@lower_builtin use cases
  • Remove macro expansion (Graham)
    • replace "macros" with standard typing/lowering; i.e. @register / @lower (@overload not yet working with CUDA)

• llvm 9

• 0.51 potential tasks

  • TypedSet
    • base on TypedDict
    • Related:
      • https://github.com/numba/numba/pull/5639