Danny Faught <[EMAIL PROTECTED]> writes:
>Uri Guttman wrote:
>> are you doing your own signal handling in perl? if so, you should switch
>> to event.pm. perl's signal handling is well known to be unsafe.
>
>I'm putting together some code samples to explain how event loops work,
>but for any serious even-driven work, I'd use one of the available
>libraries. FWIW, the slides will be on my web site some time tomorrow.
>
>> the way to handle that properly is to check for any pending signals
>> before you enter the select call (while blocking any from coming in) and
>> then checking after the select returns. this can only be done correctly
>> in C.
>
>There's no way in perl or C that I know of to atomically set a signal
>mask and enter a select call. I've seen rumors of a pselect library
>call available on some systems that can do this. A sigprocmask call
>followed by a select call has a race condition, because a signal can
>come in between the two statements. How do you solve this in C? I've
>seen a klugey solution that requires a setjmp/longjmp.
The way I have seen this fixed is to have signal handler write()
to a pipe that is in the select() list rather than just set a flag.
--
Nick Ing-Simmons
http://www.ni-s.u-net.com/
