[Quickfix-developers] Acceptor cleanup problem when port is busy (QuickFIX 1.3.2)

[Gene G. <mus...@ya...>]

Quickfix 1.3.2 has a following problem:
when the port specified for acceptor is already taken
(for example FIX server is started erroneously twice),
acceptor fails, but gives no indication to of that to
the host program.  Acceptor (both ST and MT) "start"
method still returns true as if application has shut
down normally, and no Application method is called to
indicate that something is wrong. 

Since thread that executes Application->onRun does not
exit on user action (as is normally the case),  and no
indication is given to program that something is
wrong, this behavior usually results in an unpleasant
situtaion such as deadlock (when thread_join is
attempted with never-exiting onRun), or a crash (on my
Linux due to a libpthread bug thread_join returns
fine, but onRun is still executing and crashes when
application class is cleaned up)

Although there is nothing that FIX engine can do to
stop onRun thread by itself, it should somehow
indicate to the host program that the acceptor failed
to initiate the accepting socket. Returning "false"
from the start method would allow the developers to
clean up onRun themselves. 

Fixing return value of the start method will require
small modifications to SocketAccepor::onStart, and
ThreadedSocketAcceptor::onStart; if the blocking call
(SocketServer->block or socket_accept) fails, the
whole method should return false.

Gene

__________________________________________________
Do you Yahoo!?
Yahoo! Mail Plus - Powerful. Affordable. Sign up now.
http://mailplus.yahoo.com