[tolj@wapme-systems.de: Re: Problem with accept?!!

[Thomas Pfaff]

Corinna Vinschen wrote:
> Thomas,
> 
> On Tue, Jun 03, 2003 at 12:48:41PM +0200, Corinna Vinschen wrote:
> 
>>On Tue, Jun 03, 2003 at 11:44:13AM +0200, Thomas Pfaff wrote:
>>
>>>Attached are my changes so far, i will do some further testing.
>>
>>Cool!  I didn't know that the WinSock error is thread specific.  With
>>your changes it looks way better.  Are you going to check that in?
> 
> 
> btw., there's a problem left.  When signal_arrived is triggered, the
> thread is terminated using TerminateThread().  As mentioned before, the
> problem with this approach is that TerminateThread() leaves dangling
> handles which are only removed by exiting the application.
> 
> I was already experimenting with the old WSASetBlockingHook stuff but
> it doesn't work well.  So it just occured to me that we could do sth.
> similar as in cygwin_select.  What about creating *another* socket in
> the main function which only function is to be able to be select()ed.
> Then, as soon as signal_arrived is triggered, this socket gets closed
> by the main thread.  The result would be that select() returns and we
> can leave the accept/connect_thread gracefully instead of terminating
> it.
> 
> Good?  Bad?  Ugly?

I do not know if i understand you right but:

If you see a chance to use cygwin_select instead of winsock_select than 
there is no need to terminate the threads since they must not be used at 
all, the accept/connect thread stuff can be moved in the mainthread 
since cygwin_select is already interruptible.

Thomas