Experimental Physics and
Industrial Control System

Marty Kraimer <[email protected]> · Wed, 16 Nov 2005 14:17:45 -0600

I talked to Jeff awhile ago and we agree that there is a big problem

with epicsAtExit.

I added an epicsAtExit handler to dbCa that stops the dbCaTask after

calling ca_context_destroy.

When this happened on linux using real time priorities an assert failure

occured when a db_post_event occured

for a record the was the target of a CP link.

After talking to Jeff I think I agree that we can do one of two things

1) Dont do any cleanup, i.e. just do what we did on 3.13 on vxWorks with

the exception of rebootHookAdd

2) Completely clean up the IOC database, i.e. make all device support

terminate or at the very least all soft support

and make the record stop

If we try for half way measures there will probably always be problems.

I commented out the code in epicsExitCallAtExits and tried exampleApp on

linux (with priorities) and

on solaris. I had the example sequence program runing. I terminated both

via exit and Ctl/C. The exampleApp always terminated properly.

So why did we invent epicsAtExit?
Does anyone remember?

Marty