[Python-3000] threading, part 2 (original) (raw)

Jason Orendorff jason.orendorff at gmail.com
Fri Aug 11 17:47:39 CEST 2006

On 8/11/06, tomer filiba <tomerfiliba at gmail.com> wrote:

why is thread.raiseexc(id, excobj) a bad API?

It breaks seemingly innocent code in subtle ways. Worse, the breakage will always be a race condition, so it'll be especially hard to reproduce and debug.

class Foo: ... def close(self): self.f.close() self.closed = True

Any code that uses the "closed" attribute obviously depends on it being properly set, right? This close() method gets this right. It sets "closed" if and only if the self.f.close() call succeeds. There are circumstances where this will fail: MemoryError, KeyboardInterrupt, a broken trace function, a broken setattr(), del builtins.True... but all are extreme cases. I think thread.raise_exc() should be considered extreme too. Otherwise, its existence must be considered to degrade the reliability of the above code.

I'm not saying "don't add this". Maybe it's useful, particuarly as a fallback mechanism for killing a runaway thread. But it should be documented as an extreme measure.

-j

More information about the Python-3000 mailing list