pq20445 - Unhandled sig_alrm causes packaged application to get a Alarm Clock message and melt.
----------------------------------------------------------------

Affected Platforms:

 AIX


Affected Features:

 BaseSmalltalk


Prerequisites:

 None


Symptom:

An alarm clock message is being generated when at
tempting to perform tcpip socket communications 
in a packaged application.  The error message 
occurs when SciSocketManager>>#getServByName:
protocol: is called.  This is the first async 
call in the app.  The problem appears to be 
related to the async call rather then specifically
 to that method.  The problem occurs only in a 
 packaged image.  It was packaged as an EpHeadlessRuntime 
 app but I believe the same error occurs as a VA 
 Reduced Runtime image.  The runtime image melts 
 immediately after the Alarm Clock message.  The 
 client component of the app is running on AIX, 
 the server is on win/nt 4.0 (the server could be
  running as either a development or runtime image).


 Workaround:

Interrupts are not set at the time that the os 
receives the sigalarm and so the vm dies. 
Specifically, the timer interrupt is the culprit.
 So the solution is to force the timer interrupt 
 to be installed by simply saying:

(Delay forMilliseconds: 1) wait

prior to the first socket activity. This need only
 be done once at startup.  In truth, the SIGALARM
  problem will happen only on calls that really block 
  (connect or accept) -- so on win3.1 and any polled 
  platform, this error shouldn't occur.