Article 155893 of comp.os.vms:

In article <3249923D.2BDD@nvt.novit.telemax.no>, Hans Jorgen Helgesen <hans.helgesen@nvt.novit.telemax.no> writes:
:OS: VAX/VMS V5.5, programming language "C".
:
:Problem: I have two processes communicating with each other via VMS 
:mailboxes and SYS$QIOW() calls. The sending process first calls 
:SYS$QIO(..IO$_READVBLK..) to set up an async. read. Then it calls 
:SYS$QIOW (...IO$_WRITEBLK..) to send a message to the receiving process' 
:mailbox. When the write-io has completed, the sender waits for the async. 
:read to complete.
:
:This works fine almost always, but from time to time (this is a heavy 
:loaded system), the write-call hangs. This might happen if the receiver 
:doesn't read the message, BUT IT SURE DOES. 

  Personnal recommendations:

	o The specification of the IOSB argument is `required'
	  whenever a call `permits' the specification -- assume
	  the IOSB argument is required, not optional.
	o Confirm that the IOSB storage allocation is made in
	  non-volatile storage, and that the storage will be valid
	  over the life of the asynchronous (sys$qio, etc) call.
	  Be very wary of any IOSB that is declared/allocated from
	  within the stack frame of a routine that might `return'
	  while the sys$qio or other synchronous call -- a call
	  that was issued from in that stack frame or an inner
	  stack frame -- may be (is) still outstanding.
        o Assume the specification of an explicit (non-zero) event
	  flag is required, whenever a call `permits' same.  (The
	  potential for contention on EF0 -- the default EF -- is
	  rather large.)
	o Use unique (non-zero) event flags across each (and all)
	  asynchronous calls that may be outstanding at any given
	  time.  (This technique avoids `spurious' IOSB checks, when
	  another thread happens to set a (shared) event flag.)
	o synchronize completion via the sys$synch call, or by
	  checking *both* for the event flag *and* for a non-zero
	  IOSB.
	o Unique event flags can (and should) be allocated via calls
	  to lib$get_ef.

  -------------------------- pure personal opinion ---------------------------
   Stephen 'Hoff' Hoffman   OpenVMS Engineering   hoffman@xdelta.enet.dec.com
   <Headers deliberately munged to avoid automated junkmail -- the spam-bots.>
    OpenVMS Frequently Asked Questions (FAQ): http://www.openvms.digital.com