Pearu Peterson

> (1) ... So is y mutable? I guess you'd have said no Yes, `y` is not mutable because you cannot mutate it via operations on `y`. > - but...

Recall, Python `__setitem__` does not support returning non-None values, that is, the assumption in https://github.com/data-apis/array-api/issues/845#issuecomment-2388381523 that one would be able to implement `__setitem__` such that ```python x[0] = 1 ```...

> If it helps, you can replace x[0] = 1 with *** in that code example Ok, fair enough. A better example would be that uses, say, some in-place operation...

@dependabot rebase

`find_common_type` is removed from newer numpy versions and this blocks using pyfma. Here's a quick hack for local scripts that unblocks using pyfma: ```python if not hasattr(numpy, "find_common_type"): def find_common_type(array_types,...

FWIW, as an user of finfo, I have found that duck-typing of finfo instance attributes is a very useful feature.

> @pearu can you clarify what you mean? `np.float64` duck-types as `float` in some ways but not in others. Is the current behavior sufficient? Yes. With earlier versions of numpy,...

> You seem to be agreeing with the point I made at [data-apis/array-api#405](https://github.com/data-apis/array-api/issues/405) that the return types for finfo should _not_ be `float`. Yes, except `numpy.finfo(float).eps` should be `float`.

> The array API doesn't define finfo for Python `float` or `complex`, so that's not a concern (unless you want to suggest that should be added). Right, thanks for pointing...

Yes, in principle. The rjit signature should read `rjit('double[](double[], double[])')`. For the heavydb backend, see test cases in https://github.com/heavyai/rbc/blob/main/rbc/tests/heavydb/test_array_math.py, for instance. For the numpy backend, I am sure it can...