Re: Socket's close() doesn't close connection (old bug?)

[Martin Egholm Nielsen <martin at egholm-nielsen dot dk>]

Just continuing in my own personal forum...

> > > In my eager to reproduce another bug I've stumbled across yet another 
> > > strange issue - namely that Socket's close() doesn't close underlying 
> > > socket (with no traffic), nor "wake up" a blocked read-thread (as a 
> > > consequence)...
> > Blah blah blah, this turned out to be reported already:
> >
> > http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15430
> >
> > Unfortunately still open ;-)

> It seems that if you set the SoTimeout value, so that read() returns, 
> the socket will be closed after returning the blocking thread.
> So I guess it's only a matter of interrupting any blocking thread when 
> close() is invoked - then everything will be good?!
>
> (My naive thoughts - I'll look into the implementation in a moment to 
> see if this is possible in any easy matter...)

Well, it seems like one can do the following:

1) Register the thread invoking read() (and its siblings) in 
"natPlainSocketImplPosix.cc".
2) When closing the socket (also in natPlainSocketImplPosix.cc), a flag 
should be set true, and the thread cached from before should be interrupted.
3) Back in read(), the flag is examined for why the read was 
interrupted, returning a correct exception to the user.

That's a small workaround...

// Martin