Le 01/10/2017 à 21:20, Benjamin Peterson a écrit :
> 
>>> signal.pthread_kill and pdox's proposed time.pthread_getcpuclockid on
>>> https://github.com/python/cpython/pull/3756
>>
>> Given how specialized (and of little actual use) those functions are, I
>> don't think it's much of a problem if they misbehave if you deliberately
>> pass an invalid thread id.
> 
> Okay, so you think this PR is reasonable?

On the principle yes.  I am not interested enough in this feature to
look at the implementation.

Another possibility would be to have a new separate threading API
returning a "low-level opaque thread handle" that you could pass to
pthread_kill() and pthread_getcpuclockid().  threading.get_ident() could
still be used for logging purposes and others.

A related issue: what if you want to call pthread_kill() on a non-Python
thread?

> I suppose, though you can't usefully use it as an integer. I imagine a
> comparable and hashable opaque object would be compatible with most
> code.

What happens for code that uses e.g. "%x" to format thread ids?