Ah this is a bit awkward, because say ph.assert_dtype() will always use in_dtype both for its error messages e.g. ... [func(uint8, int16)], and sometimes use it for inferring expected. In say ph.assert_keepdimable_shape(), in_shape is indeed always used as the "expected" to assert against.

Still yeah it'd be a good idea to to review the args order a bit, and definitely write some docs. I hadn't spent too much time on documenting utils because I end up removing half the ones I write anyway, but now would be a good time to regroup.