[prev in list] [next in list] [prev in thread] [next in thread] 

List:       cygwin
Subject:    Re: Postgres Backend doesn't catch the next command, after SIGUSR2
From:       Patrick Samson <p_samson () yahoo ! com>
Date:       2004-03-30 10:30:57
Message-ID: 20040330103057.32104.qmail () web60302 ! mail ! yahoo ! com
[Download RAW message or body]


--- Corinna Vinschen wrote:
> On Mar 30 01:03, Patrick Samson wrote:
> > >From msdn:
> > "All I/O operations that are canceled will
> complete
> > with the error ERROR_OPERATION_ABORTED. All
> completion
> > notifications for the I/O operations will occur
> > normally."
> 
> Urgh, I mised that.
> 
> > What's your feeling about:
> > if (!CancelIo ((HANDLE) socket))
> >   {...}
> > else
> >   {
> >    if (WSAGetOverlappedResult (socket, &ovr, &len,
> >  FALSE, flags) && len != 0)
> >     ret = (int) len;
> >    else WSASetLastError (WSAEINTR);
> 
> Did you try it?

No. Never ever build the dll yet.
I guess I should one day ;)

> Yesterday I changed Cygwin to use
> asynchronous I/O
> instead of overlapped I/O so it now can do without
> CancelIo.
> However, two people reported hangs which don't occur
> for me.  If
> if takes too long to track down, I guess I'll revert
> to overlapped
> I/O plus your patch.  But I would be more happy with
> a working
> async I/O solution.

Yesterday, I thought that mixing interrupt signals
and overlapped operations had some design flaw,
because of (from WS2 API):
"There is no way to cancel individual overlapped
operations pending on a given socket, however, the
closesocket() function can be called to close the
socket and eventually discontinue all pending
operations."
So non overlapped I/O would be better for my
personal case (only mime I agree). But I didn't even
think you could change to another behaviour.
Async I/O would probably saves my life, if the hangs
are solved.

Today, I think differently. If it's not possible
to cancel one *individual* op, I understand that
it's possible to cancel *all* ops in a clean way.
closesocket is one way, but not suitable in that
case. If cancelIO also is a clean way, then playing
with signal and overlapped I/O remains workable.



__________________________________
Do you Yahoo!?
Yahoo! Finance Tax Center - File online. File on time.
http://taxes.yahoo.com/filing.html

--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Problem reports:       http://cygwin.com/problems.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/

[prev in list] [next in list] [prev in thread] [next in thread]