Nick Maclaren wrote:
>
> In article <[EMAIL PROTECTED]>,
> hg <[EMAIL PROTECTED]> writes:
> |>
> |> I posted an equivalent question earlier ... but am still not sure:
> |>
> |> I currently (under Linux) have a program that uses Queue.put
> |> (raw_input('')) in a signal handler and Queue.get() in the main/only
> |> thread.
> |>
> |> It works but I want to know if it is legal / what type of synchro API I
> |> have the right to use in a signal handler.
>
> Strictly, no and none.
>
> C and POSIX's specifications of signal handling is a complete mess.
> ALL handling of ALL real signals is undefined behaviour, though few
> programmers realise that and POSIX depends on it. As that would make
> all Unices (and Microsoft systems) unusable, you can reasonably assume
> that there is some signal handling that will work.
>
> But what? Which is where you came in.
>
> The situation is that most things that seem to work do actually work,
> in some environments and under some circumstances. You can never be
> sure when they will stop working, however, and all bets are off when
> you port signal handling code to a new system. And, really but REALLY,
> do not mix it with very high levels of optimisation or rely on it
> being in any way thread safe.
>
> I can't tell you whether the code of Queue is intended to work if an
> active Queue.get is interrupted by a signal which then does a Queue.put,
> but that is the sort of circumstance where things get very hairy. Even
> if Python has code to enable that case, it won't be 100% reliable on all
> systems.
>
> Sorry. But that is the situation :-(
>
>
> Regards,
> Nick Maclaren.
Fair, I'll find a way around then.
Regards,
hg
--
http://mail.python.org/mailman/listinfo/python-list
