asyncio.sleep(float('nan')) does not raise ValueError

[jakkdl]

Bug report

time.sleep(float('nan')) raises ValueError, but asyncio.sleep(float('nan')) is valid and sleeps for 0. This seems bad, and asyncio.sleep should probably also raise ValueError.

import asyncio
import time

try:
 time.sleep(float('nan'))
except ValueError:
 print("time.sleep raised ValueError")

async def foo() -> None:
 await asyncio.sleep(float('nan'))
 print("asyncio.sleep does not raise any error")

asyncio.run(foo())

$ python foo.py
time.sleep raised ValueError
asyncio.sleep does not raise any error

time.sleep and asyncio.sleep do differ in behaviour when it comes to negative numbers and changing that behaviour was considered a breaking change without strong reason: #83879
nan feels like it should very rare to be passed intentionally, but the potential gain might also be small enough that this isn't worth fixing.

Related: python-trio/trio#2493

Your environment

• CPython versions tested on: 3.8 to 3.11
• Operating system and architecture: Linux 6.3.4-arch1-1

Linked PRs

• gh-105331: Fix asyncio.sleep() bug #105501
• gh-105331: Fix asyncio.sleep() bug #105513
• gh-105331: Fix asyncio.sleep() bug #105641

[sobolevn]

It is also different for negative values.

For example:

>>> import asyncio
>>> async def a():
... await asyncio.sleep(-1)
... 
>>> asyncio.run(a())

and

>>> import time
>>> time.sleep(-1)
Traceback (most recent call last):
 File "<stdin>", line 1, in <module>
ValueError: sleep length must be non-negative

[jakkdl]

It is also different for negative values.

Yes, I wrote

time.sleep and asyncio.sleep do differ in behaviour when it comes to negative numbers and changing that behaviour was considered a breaking change without strong reason: #83879

[kumaraditya303]

I don't think we should change it at this point, I can accept a docs improvement for this but I'm -1 on changing the behavior.

[Zac-HD]

fwiw I'd expect it to follow whatever time.sleep() does with weird arguments, and think that there's a pretty strong argument in favor of consistency.

[gvanrossum]

Hm, I expect at this point there is probably plenty of working code out there that passes a negative delay to asyncio.sleep (perhaps the result of trying to sleep until some target time that's slightly in the past).

I don't see how NaN could be the result of such a computation, so I agree NaN is a "weird" argument (and leads to undefined behavior).

So I think I can only approve of the NaN check.