Repository navigation
BUG: SeedSequence accepts types outside of spec #27380
Description
Describe the issue:
SeedSequence supports None, int, Sequence[int] for entropy. It turns out that anythign with len(obj)==0 in the leading dimension will be accepted and is identical to 0. This includes ndarrays with any number of dimensions.
This might be more of a FR than a bug, but the lack of standardization could increase end users error risk.
Reproduce the code example:
import numpy as np
np.random.SeedSequence(0).generate_state(1)
np.random.SeedSequence([]).generate_state(1)
np.random.SeedSequence([[],[]]).generate_state(1)
np.random.SeedSequence(np.empty((0,10,10))).generate_state(1)
np.random.SeedSequence(np.empty((0,10,10), dtype=float)).generate_state(1)Error message:
No error. The output of the above is
All produce
array([2968811710], dtype=uint32)
array([2968811710], dtype=uint32)
array([2968811710], dtype=uint32)
array([2968811710], dtype=uint32)
array([2968811710], dtype=uint32)Python and NumPy Versions:
2.1.1
3.12.4 | packaged by Anaconda, Inc. | (main, Jun 18 2024, 15:03:56) [MSC v.1929 64 bit (AMD64)]
Runtime Environment:
No response
Context for the issue:
It simply seems a bit less careful than it should be, and could allow accidental seed values in.
Activity
rkern commentedon Sep 12, 2024
The 0-padding behavior for
[]should certainly be documented (we 0-pad out topool_sizeuint32integers so that spawning doesn't create conflicts).We certainly could tighten up our validation in
_coerce_to_uint32_array(). What we implemented there was really an older proposal that didn't make the final cut that allowed a lot more than what we currently document because we thought we might want to allow things like hex-int'0x1badcafe'strings for ease of integration. That could certainly use another pass to align it with the final documentation.As a rule though, the documentation defines the standard and things outside of what's documented MAY raise a helpful exception but MAY also just execute with undefined behavior. Possibly worth tightening up if it's easy, but not urgently.