MAINT: Better error message for linalg.norm

[ZhihuiChen0903]

For example:
numpy.linalg.norm(2, np.inf)

raises:
ValueError: Improper number of dimensions to norm.

The error messageis a bit vague.
This pull request changes the error message so that it explicitly states the expected dimensions.

Expected error message:
Input must be 1-D or 2-D when axis=None (unless ord is None)

[mattip]

It seems your error message would be displayed when calling np.norm(a, axis=(1, 2 3)), which would be confusing.

[ZhihuiChen0903]

@mattip
Is it possible to provide more specific error in the error message?

[mattip]

You could do a specific check for len(axis) your error condition in an else clause on line 2534. Please add a test or point to an existing test that trips the new code.

[InessaPawson]

@ZhihuiChen0903 Thank you for sharing your solution! Do you think you could submit a requested test (#17882 (comment)) some time soon?

[charris]

I'm going to close this. The error message could be improved, but is seems it needs more work to get it right. @ZhihuiChen0903 feel free to make another PR along these lines.