Minutes_2020_06_30
Valentin Haenel edited this page Jun 30, 2020
·
1 revision
Attendees: Hameer, Stuart, Todd, Pearu, Ehsan, Siu, Val, Guilherme,
- 0.50.1 released
- roadmap?
- core maintainers may not have time given
- expand to have more contributors
- list of simple things:
- labels: "good-first-issues", "good-second-issues"
- more internal docs
- expectation management to get up to speed in numba.
- Val: timeframe >6mos to 24mos
- bootstrap new numba devs with help from other numba devs
- Guilherme thinks numpy/scipy functions are easy for newcomers
- Roadmap theme for time frame
- wishlist or priority list for community to tackle
- many items are researchy with hard to estimate ETA
- Add a list of easier issues (largely refactor):
- point to code
- owning maintenance of parts of numba
- Decide on how to proceeed with https://github.com/numba/llvmlite/pull/602
- Rebuild wheels instructions https://github.com/numba/llvmlite/tree/master/buildscripts/manylinux1
- No patch, but rebuild wheels
-
#5933 - Wrong result with np.linalg.inv for 1x1 size, modifies input in-place
- has a PR fix
-
#5932 - Unification error for datetimes (minimum/maximum)
- np1.18
- unification error
-
#5931 - Support for exec() at precompile time
- need answering
-
#5930 - Strange/wrong Error message when giving list of function pointers to jitted function
- need answering
-
#5929 - numpy ediff1d tests have TypeError in
to_end
(numpy 1.19)- remove a patch we made to fix older numpy
-
#5927 - SVML weirdness 🙃
- Maybe an LLVM bug
- non-deterministic
-
#5924 - allow staticmethod's to be compiled with jit(forceobj=True)
- Good idea, but perhaps need a different design
- Use a decorator on the function instead
-
#5923 - Failed in nopython mode pipeline (step: convert to parfors)
- Might work as R*=255 + m * 255
- Hopefully precedence is correct
-
#5917 - Release 0.50.1 checklist
- Closing as released
- #5916 - Fix numba.typed.List pre-allocation
- #5915 - Accelerate numba.typed.List constructor when used from the interpreter.
-
#5911 - Feature Request:
__hash__
for Enum and IntEnum -
#5910 - AOT compiled function inside JIT function fails.
- Can probably make this work, with a little bit of help
- #5909 - typedlist slower than Python list
- #5908 - TBB threading layer deadlock on terminate
-
#5907 - Numpy wrappers not used for scalar return values
- Yes, it's an issue
- Numba only has one scalar type
-
#5905 - Yielding while iterating over a typed list results in a segfault
- Parent in iterator is null
- Might be really hard
- #5904 - Performance issue with prange when an user-defined function is called
- #5903 - Keyword-only arguments and default arguments always do the wrong thing when combined
- #5926 - operands on datetime64 data types
- #5925 - Use np.inf and -np.inf for max and min float values respectively.
- #5922 - Reduce objmode and unpickling overhead
- #5921 - Don't try to analyze the broadcast if the shape of some part is not known.
- #5906 - Fix optional type in static_getitem
- #5920 - Update master changelog for latest change in 0.50.1
- #5919 - Update changelog 0.50.1
- #5918 - Fix cuda test due to #5876
- #5914 - Update change_log for 0.50.1
- #5913 - Update change_log for 0.50.1
- #5912 - Fix bad attr access on certain typing templates breaking exceptions.
-
Requests for 0.51
-
0.51 potential tasks (To be updated)
- Opening up the numba meeting