Minutes_2019_09_10

Attendees: Pearu, Aaron, Stuart, Todd, Ehsan, Stan, Siu, Val, Hameer, James

0. Feature Discussion

• Python 3.8:
  • Can use conda packages from conda install -c conda-forge/label/pre-3.8 for testing? (No NumPy built there yet)
  • Might need to tackle this with manual builds for now, and wait to pull into CI.
  • TASK:
    • limit to one OS to prune 90% of the problems; hopefully mostly bytecode problems.
• https://github.com/numba/numba/issues/4546
  • TASK:
    • make overloading on other targets work
• Adjusting num-threads at runtime
  • SUGGESTION:
    • static allocate for MAX size
    • check for dynamic size at runtime

1. New issues

• #4540 - Problem lowering code using timedelta arrays
  • Need to implement type
  • Give pointer to how
• #4538 - Cannot traverse DeferredStructModel
  • TODO: siu look at this
• #4537 - Function arguments and key arguments lost with @vectorize
  • low-priority high-effort
• #4535 - while True: causes 'missing Environment'
  • TODO: ask user if there's a bigger case with side-effect.
• #4534 - TypeError from trying to cache higher order functions
  • TODO: siu
• **** #4533 - hpat error with master
• #4531 - so slow,why!
  • Waiting on user to clarify
• #4530 - np.where with boolean fails without parallel=True
  • Needs to make the fix official
• #4529 - LoweringError in scanpy's calculate_qc_metrics function
  • int width problem
• #4528 - Faster array indexing (closer to what Numpy can do)
• #4527 - numba.errors.TypingError: Failed in nopython mode pipeline
  • problem with subtract on array[dtype=char]
• #4525 - Untyped global name 'str': cannot determine Numba type of <class 'type'>
  • TODO: reply str building is not good usecase
• #4524 - Numba fails to understand float('inf')
• **** #4522 - jitclass vs njit decorator performance
  • Stuart will talk about this
  • Trouble with refcount on self
    • LLVM pass to better refcount pruning?
    • High-level pass to avoid binding self?
    • Fix exception raising not decref'ing?
  • TODOs:
    • investigate the njit 2x slower than cython
• #4521 - Document how to use tuples with numba.typed.List and numba.typed.Dict
• #4520 - Tuples containing bools do not work with numba.typed.List or numba.typed.Dict
  • Missing cast from i1 to i8.
• #4519 - Array changes unexpectedly when used with numpy.linalg functions
  • not jitclass problem.
  • array linalg layout problem
  • complicated
• #4515 - Segfault when importing numba in conda environment
  • likely ABI problem conda vs manylinux1
• #4514 - Unable to use typed.Dict with jitclass?
• #4509 - List.__getitem__ needs to handle unsigned int
  • PR in progress
• #4508 - LoweringError iterating over 2D array @numba.jit(nopython=True)
• #4507 - Call cuda code From Device Function?

Already Closed

• #4526 - Error when converting numpy array to bool
• #4511 - Typing jit signature of readonly arrays inputs

2. New Open PRs

• #4541 - fix two typos in architecture.rst
• **** #4539 - [WIP] Support jit_options/inline for overload_method
  • Accept code duplication here
  • Make it a HIGH PRIORITY to merge @overload* in the next cycle.
• #4536 - Handle kw args in inliner when callee is a function
• #4532 - Fix static_getitem with Literal type as index
• #4523 - [WIP] make @lower_constant return new references
• #4518 - add support for np.append
• **** #4517 - Option to track heap allocations
• #4516 - [WIP] Make setitem accept 0d np-arrays
• **** #4513 - New compiler.
  • Ready for review
• #4512 - Raise specific error at typing time for iteration on >1D array.
• #4510 - Make List.__getitem__ accept unsigned parameters

Already merged

4. Next Release: Version 0.46.0, RC=Sept 30

• Finish pending rewrite passes
• Python 3.8 support
• Document best practices for constructing new compiler pipelines
• Define autodiscovery system for Numba extensions (like numba_scipy or HPAT) that don't need direct user import
• Allow opt-in dispatching of functions by literal value
• Making caching aware of transitive dependencies
• Define declarative typing system for @overload (to be used in future releases)
• Numba Runtime C API for extensions to register reference-counted memory with the runtime.
• Start using new CI system in parallel with existing one
• Priority bug fixes:
  • Low performance of JIT method calls (requested by Pandas devs)
  • Others TBD