Skip to content
Valentin Haenel edited this page Jul 8, 2020 · 1 revision

Numba Meeting: 2020-07-07

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

0. Feature Discussion

  • Open meeting next week. Agenda?!

    • Discourse feedback
    • Roadmap needs Community growing?
    • Daily task of maintainers
      • short presentation
    • Expectation on timeframe
    • look at zenhub
    • How to get community involved in
      • 1st/2nd pass review
      • labelling: difficulty, category by component
      • Link to label for "easy" PR/issues in CONTRIBUTING?
    • Any ideas from the community?
    • meeting themes
      • TODO: announce meeting and ask for theme; then vote on theme
  • Burndown starts tomorrow.

    • Need to do burndown meeting

1. New Issues

  • #5955 - LoweringError: LLVM IR Parsing Error during scanpy sc.pp.calculate_qc_metrics
    • need a cast somewhere
  • #5954 - Move doc links over to RTD
    • just a reminder
  • ** #5953 - Extreme compilation times with long tuples
    • IR is massive
  • ** #5951 - Decide whether or not to support the target kwarg in the numba.jit decorator
    • Someone needs to put in the work, to see if this is even a valid code path.
    • Does it really need to be public API? Can we rename to _target instead?
    • 1-cycle deprecation for target. _target as internal API
    • Fix up valid target code paths, remove invalid ones, e.g. 'gpu'
  • #5950 - ndarray.flags.writeable = False for structured array compiles/silently fails to assign with njit
    • Normally the compiler should catch this, but we don't do this properly yet for structured arrays
    • Accidentally doing the right thing?
    • This should warn during compilation
  • #5948 - strange behavior of @guvectorize
    • Dependent on how Numpy allocates the res array
    • Should have gone to discourse
  • #5942 - doing " += 1" in arrays does not add one in prange loop
    • Maybe doesn't recorgnize the assignment
    • A slice assigned to a slice: parfor within a parfor
  • #5934 - Cannot Fill 0d Structured/Record Arrays
    • Structured arrays will need some updating
    • Perhaps a task for the community?
    • Now that more fascilities are available, an update might be feasible, but needs a ticket

Closed Issues

  • #5961 - invalid number of args for optional arguments of jitclass methods.
  • #5957 - I can't install numba on rpi 4b
  • #5947 - numba -s fails with: AttributeError: 'str' object has no attribute 'items'
  • #5943 - Compilation is falling back to object mode WITH looplifting enabled
  • #5939 - Unary variation of np.round doesn't work as expected.

2. New PRs

  • #5962 - Fix SVML tests with LLVM 10 and AVX512
  • **** #5960 - Setting mypy on CI
    • good experiment to see how well it works in numba codebase
    • mypy Protocol
  • #5959 - Making typed.Dict a typing MutableMapping
  • #5958 - Making typed.List a typing Generic
  • #5952 - Fix "Internal error ... local variable 'errstr' referenced before assignment during BoundFunction(...)"
  • #5946 - Implement literal dictionaries and lists.
  • #5945 - Fixes the jit decorator with target='cuda'
  • #5944 - re-enable OpenMP in wheels
    • needs manual testing
  • #5941 - Fix/cuda atomic nanminmax
  • #5938 - Initial support for dynamic gufuncs
  • #5936 - Improve array analysis for user-defined data types.

Closed PRs

  • #5956 - Update numba_sysinfo.py
  • #5949 - Fix for: numba -s fails with: AttributeError: 'str' object has no attribute 'items'
    • Solved: #5947
  • #5940 - Add discourse badge
  • #5937 - [WIP] Make signature optional on guvectorize
  • #5935 - Fix default arguments with multiprocessing

3. Next Release: Version 0.51.0, RC=22 July, Final 29 July?

  • Requests for 0.51

  • high risk stuff for 0.51.

  • 0.51 potential tasks (To be updated)

4. Upcoming tasks

  • Opening up the numba meeting
Clone this wiki locally