Da Li (李达)

with this PR, if add `@overload(np.apply_along_axis, override=True)` for the second definition, the output is correct.

Yeah, it's more like a POC. But I will push it forward to make it complete. And I can enjoy a "pre" review if anyone would like to comment.

Numba compilation still has at least 2 issues: 1. sometimes, if the return type of a jitted/overloaded function is Tuple, numba forcily doesn't cache this function into its cache dict....

An example to show why we need this PR: ```python from os import environ environ["NUMBA_CHROME_TRACE"] = "3.json" environ["NUMBA_DUMP_LLVM"] = "1" from numba import njit from numba.core import types from numba.typed...

I didn't change this file, but CI reports the flake8 errors in it. Emm... ``` numba/tests/test_unicode.py:2670:27: E226 missing whitespace around arithmetic operator ``` btw, the last real failure is about...

@sklam I'm not sure if I choose the right way to fix them. Maybe the real fix should happen at the start of "funny" things, rather than at the end...

interesting failure: ```python Traceback (most recent call last): File "/home/vsts/work/1/s/numba/tests/test_compiler_flags.py", line 40, in test_fastmath_in_overload self.assertEqual(b, "Has fastmath") AssertionError: 'No fastmath' != 'Has fastmath' - No fastmath ? ^^ + Has...

added a naive release doc and fixed the failed test case. TODO: add some new tests to check if this PR can deduplicate the expected unnecessary compilations.

this failure (numba.numba (Linux py310_np123)) is a CI infra issue, not code issue.

I don't have bandwidth to handle future change requests, so close it. In this PR, I basically give a clear POC to show where the bug is coming from but...