Da Li (李达)

Ah, cool! Perhaps can you also consider how to do the similar thing for structref? It also has at least one location using `njit` directly without any options. Here: https://github.com/numba/numba/blob/e15cd6870042125ff025e459f5f7558449b22cf8/numba/experimental/structref.py#L371-L375

Test code: ```python from numba import njit from numba.extending import overload def foo(): raise NotImplementedError @overload(foo, override=False) def ol_foo0(): def impl0(): return 0 return impl0 @njit def goo(): return foo()...

> we think that target-extension API can fulfill the same needs I think I didn't get it. Do you imply if we change the decorator usage from `@overload(..., override=True)` to...

Thanks! Let me figure out how to apply this feature into my use case.

close this, since this is my experimental PR, and won't get reviews as well. I didn't try the way [Guilherme Leobas](https://github.com/guilhermeleobas) pointed out. When I try this, will report my...

re-edit: After I rebuilt numba dispatcher, the code snippet in the first comment throws an error: ```python File "/Users/dali/Code/open-numba/numba/core/types/common.py", line 52, in __init__ if isinstance(dtype, Buffer): File "/opt/homebrew/anaconda3/envs/numbaenv/lib/python3.10/abc.py", line 119,...

Yeah, in `numba/core/callconv.py`, more than one location can generate `excinfo`-related `store` instr, but only one location adds the metadata for it. Can push a PR soon.

could be related to https://github.com/numba/numba/issues/9186, where I observed this `excinfo` change hurts the performance.

@guilhermeleobas, yes, I tried to add more metadata in #9551 . And my internal test case's performance is better with that. ex: w/o this PR, calls of `NRT_decref` are 3000...

also tag with #9409 meta issue.