Minutes_2019_10_22

Numba Meeting: 2019-10-22

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

0. Feature Discussion

• Numba 0.46.1 Status
  • Python 3.8 bytecodes really mess with control flow assumptions
• Error message verbosity?
  • Stuart has a related PR: https://github.com/numba/numba/pull/3942
  • https://github.com/numba/numba/issues/4344
  • turn off "suggest-user-to-submit-issue"
  • turn off "template with literal without literal"
    • no fallback to other overloading template unless it's a TypingError; panic on other kind of errors

1. New issues

• #4730 - Refactor call table code to replace with find_call_name or similar.
  • Yes, do this.
• #4728 - How to turn a normal python dict into a numba.typed.Dict?
  • need a copy constructor
  • Good second issue?
• #4727 - numpy.array_equal is not supported
  • Fast forward to future
• #4726 - target='cuda'
  • known issue
  • need better error because this is so common
• **** #4723 - Tracker for Numba usability improvements Q4 2019.
  • keypoints:
    • pandas using numba in 1.0 milestone
    • split impl for having cython and numba impl
• #4722 - Should there be unsafe implicit casts for function arguments?
  • need more information, Siu will ask follow-up questions
• #4720 - Some question about nogil mode
  • Stuart answered
  • nogil is only used to control the CPython wrapper. Internal calls between nopython functions do not change the state of the GIL.
• #4719 - Supporting DLPack
  • Seems reasonable?
• #4716 - How to json a nested dictionary?
  • should json module pick up that typed.Dict inherits from MutableMapping?
• #4714 - slice constructor inconsistent with python
  • Contributor is already working on this
• #4713 - Numba is not detecting the icc_rt libraries when installed with pip
  • can't replicate the issue
• #4712 - Bug with Python 3.8 and np.argmin
  • Much larger problems with Python 3.8, in progress
• #4711 - problem with numba @jit(cache=True) in python 3.7. The same code in python 3.6 dont give any error
  • already fixed in past release
• #4709 - LoweringError while allocating shared memory
  • this can't work (due to how cuda works), but failure should have been caught earlier in pipeline
• #4708 - Erroneous behaviour of np.transpose() after two transpositions
  • weird bug. needs investigation
• #4706 - np.heaviside in no_python mode
  • would be good contribution to get
• #4705 - Feature request: Support str(<int>) and more generally str.
  • complicated to do with all the variations that Python allows in integer strings

Already Closed

• #4704 - Internal Type Detection Failure?

2. New Open PRs

• #4729 - Allow array construction with mixed type shape tuples
  • Need to decide how to deal with 32-bit vs. 64-bit
  • To be fix by checking bad ptr returned from allocator
• #4725 - Requires llvmlite >=0.30.0
• **** #4724 - Only use lives (and not aliases) to create post parfor live set.
  • #scary: is there another way to do this?
  • is copy propagation not working for init blocks?
  • Ehsan will review
• #4721 - [WIP] Allow cffi value as function arguments
• **** #4718 - Change get_call_table to support intermediate Vars.
  • needs a testcase
• **** #4717 - [WIP] Making casts explicit in Numba IR
  • SSA
• #4715 - [WIP] Add bool overload for UnicodeType
• #4707 - Added support for slice.indices

Already merged/closed

• #4710 - BUG: Comparison for literal

4. Next Release: Version 0.46.1, RC=

• CPython 3.8

• Requests for 0.47 (last release for the year) - jitclass performance issues - llvm 9 trial - CTK libcudadevrt.a - CI needs to take 50% of current time - Val & Stu already looking at this - also checking Azure CI config to avoid wasting compute time - Caching: - transitive dependency - other issues: i.e. function as argument, with objmode - distributing cache - Immutable list and deprecating reflected list - Switch to pytest (see above) - Using Numba to generate LLVM/NVVM IR for different targets https://github.com/numba/numba/issues/4546 - @overload for gpu