Re: Socket::isPending () issue

[David Sugar]

We had been a bit busy with the Common C++ "2" release, so I had not had 
a chance to think about this question until now...

Interuppted system calls can happen from signal handlers that have 
"sa_interrupt" set.  If one resumes the poll and there is data pending 
as well, the poll should then drop back out immediately and not keep 
looping.  However, there is no harm in examining the returned struct for 
pending data rather than looping right back.

In the case of select, I do not know why we did not do the same, but I 
think the immediate fallback loop in poll might be a mistake, as you 
point out.

Pavel Tsekov wrote:

> Hello :)
>
> I want to raise another issue witch have made be go crazy for the
> last few days :) On unices, based on your system a decision is made
> if poll () or select () will be used to check if network data is
> available for reading/writing. One would assume that using either of
> these will give him the same result - but not. The version of Socket::isPending
> () which uses poll () has an additional check - if the return value of
> poll is -1 then a check is made if errno is set to EINTR and if
> positive poll () is called again. This may result in an infinite loop.
>
> I have a code which sends data through a tcpstream instance. Before
> each write I do a Readline() with a timeout of 1000 milliseconds
> to see if the other end wants to inform me of something. This seemed
> to work fine though if I send a large amount of data (more than 5 Mb)
> I got struck buy this issue - at some point the code blocks in the
> isPending () and cannot get out of there. I solved the issue by using
> the version of isPending () which uses select ().
>
> Now my question is - what's the purpose for this difference in the
> behaviour of isPending ().
>
> Thanks ;)
>
> P.S. The system is: i386 runnning RedHat Linux 7.2 - everything is stock, no
> installation of custom packages.
>
>
> _______________________________________________
> Bug-commoncpp mailing list
> address@hidden
> http://mail.gnu.org/mailman/listinfo/bug-commoncpp