# Numba Meeting: 2025-08-19

Attendees: Val, Guilherme, Graham, Swapnil, Todd, Stuart, Kaustubh, Siu
FPOC (last week): Val
FPOC (incoming): Kaustubh

NOTE: All communication is subject to the [Numba Code of Conduct](https://github.com/numba/numba-governance/blob/accepted/code-of-conduct.md).

Please refer to [this calendar](https://calendar.google.com/calendar/embed?src=5rqnddm4gjrdfjm31fsv182epk%40group.calendar.google.com) for the next meeting date.

## ~~Office hour signup~~

## Discussion

- should llvmlite API do something backwards compatible where possible. e.g. llvmlite.binding.initialize() is now "dead" in LLVM 20, should this still exist and warn folks about it being dead, or should it just be removed>?
- we are breaking old code anyway, due to the changes of LLVM 20
- ~~no way out from opaque pointers~~ and need to use new pass manager
- prefer to "break hard" rather than issue warnings, it's not likely to be backwards compat.... users need to make changes...
- `llvmlite.binding.initialize()` need to `raise`
- https://github.com/numba/llvmlite/issues/780
- Debian and Fedora already build this for 390x
- if your MCJIT works it should be fine?
- are they looking for official upstream support?
- clarify the ask
- Note on issue with build logs from Debian: https://github.com/numba/llvmlite/issues/780#issuecomment-3201064593
- LLVM 20
- cmake upgrade probably ok
- changes for the llvmdev 20 recipe changes are merged to the PR
- some notes have been written as to what works and what doesn't, because not everything works in LLVM 20, a few things are fundamentally broken
- once all that is merged, the llvmlite changes for LLVM 20 can be merged
- Python 3.14
- down to one segfault and 6 test failures

## New "Ready for Review" PRs

## 1. New Issues
- [numba#10175](https://github.com/numba/numba/issues/10175) - More border handling options (mode) for stencil computations
- [numba#10177](https://github.com/numba/numba/issues/10177) - Test failures on GHA win runners
- be fixed by a PR in future
- [numba#10180](https://github.com/numba/numba/issues/10180) - Yield-returning a structref does not run the destructor.
- there are known issues with lifetimes in generator contexts
- [numba#10182](https://github.com/numba/numba/issues/10182) - Poor performance for stencils in parallel
- not certain what the issue is
- could it be an issue with `min`?
- Add has different performance - comment added to issue
- [numba#10184](https://github.com/numba/numba/issues/10184) - [Bug Report] Numba cache and prange conflict in nested function calls.
- seen this before, something about the initialization sequence when caching
- maybe a quick fix: add `parallel=True` to the shell function

### Closed Issues

## 2. New PRs
- [numba#10172](https://github.com/numba/numba/pull/10172) - Improving the histogram function
- [numba#10176](https://github.com/numba/numba/pull/10176) - Fix/gha mem
- [numba#10181](https://github.com/numba/numba/pull/10181) - [CI Only] Testing memory aware test scheduling
- [numba#10183](https://github.com/numba/numba/pull/10183) - Revert NumPy 1.23 support removal
- [llvmlite#1259](https://github.com/numba/llvmlite/pull/1259) - update osx-64 sdk on recipes and CI scripts to use MacOSX10.15 SDK
- [llvmlite#1260](https://github.com/numba/llvmlite/pull/1260) - llvmdev 20 build patches

### Closed PRs
- [numba#10173](https://github.com/numba/numba/pull/10173) - [ci only] debug sys.monitoring failure
- [numba#10174](https://github.com/numba/numba/pull/10174) - [ci only] debug monitoring; what did sys.mon got?
- merged - [numba#10178](https://github.com/numba/numba/pull/10178) - Fix test_monitoring_multiple_threads
- merged - [numba#10179](https://github.com/numba/numba/pull/10179) - update azure pipeline configs to use numpy2.3
- merged - [llvmlite#1258](https://github.com/numba/llvmlite/pull/1258) - Update pre-commit hook python-jsonschema/check-jsonschema to v0.33.3

(last numba: 10171; llvmlite 1257)