Siu Kwan Lam

Developer note: This is an issue with warnings propagated from `PartialTypeInference` due to the presence of `isinstance`. ``WarningsFixer`` should capture the warnings from it and not raise them.

Looks like a data model problem. I will take a look

I think it requires too much work and too little to gain in bypassing the Python interpreter for file I/O. But exposing unicode strings in Numba to `c_char_p` is actually...

We do have a warning: We have tried redirection but github pages disables it due to spam protection. We talked about this is developer meeting. Maybe we can delete the...

https://github.com/numba/numba/blob/338b0a2a98a3ede6fba7888a03726804d2059a61/numba/core/dispatcher.py#L288-L293 You can call `.disable_compile(False)` to re-enable compilation. The awkward part about it is the double-negative.

This can get tricky and need careful planning. In LLVM C++, `MDNode` can be unique or non-unique. IIRC, Unique MDNode are immutable. That's what llvmlite implemented only. All metadata node...

Additional references: - https://llvm.org/docs/TransformMetadata.html#metadata-on-loops > Because metadata nodes are immutable (with the exception of MDNode::replaceOperandWith which is dangerous to use on uniqued metadata) - https://github.com/llvm-mirror/llvm/blob/2c4ca6832fa6b306ee6a7010bfb80a3f2596f824/lib/IR/Metadata.cpp#L858-L868

@swap357 reported that `NUMBA_CPU_NAME="znver3"` -> `llvm ir: 40,753 lines, 2387kb` `NUMBA_CPU_NAME=""` -> `llvm ir: 6,737 lines, 475kb` `NUMBA_CPU_NAME="x86-64"` -> `llvm ir: 9,200 lines, 611kb`

As of https://github.com/numba/llvmlite/commit/dee4d034b4c8c171e08a7224a96836c726d92cf7, I am seeing much smaller code with LLVM20. `znver3` output (from the reproducer) is only 1.07x longer than the `icelake` output. Here's the data formatted as a...

A lot of the logic in Numba's `DIBuilder` is tightly coupled with the type system. For it to live in llvmlite, it will need to be generalized for any type...