Re: devSNMP: Thread error?

[Andrew Johnson <anj@aps.anl.gov>]

Albert,

Kagarmanov, Albert wrote:
> 	This is not a real error, it's warning from
> pthread_attr_setstacksize system-call, which invoked in devSnmp.c in
> function 
> 		int epicsSnmpInit(int param)

Actually it is an error, in the epicsSnmpInit() code.

> It looks like 10000 stack-size is small enough for your Linux (See
> http://docs.sun.com/app/docs/doc/816-5137/6mba5vpja?a=view for detail)
>
> We've changed this size till 128K as it recommended in
> $EPICS/base/documentation/RELEASE_NOTES.html and don't have this
> warning.
> See our http://www-mks2.desy.de/content/e4/e40/e41/e12212/ for detail.

The stackSize parameter to the epicsThreadCreate() call should not be a 
number as the sizes needed can be different on different architectures. 
 The epicsSnmpInit() function should be passing 
epicsThreadGetStackSize() one of three values epicsThreadStackSmall, 
epicsThreadStackMedium or epicsThreadStackBig according to how much 
stack the thread is expected to need, and using the return value from 
epicsThreadGetStackSize() as the parameter to epicsThreadCreate(). As an 
example, here's an excerpt from base/src/db/callback.c:

    callbackTaskId[ind] = epicsThreadCreate(taskName, priority,
        epicsThreadGetStackSize(epicsThreadStackBig),
        (EPICSTHREADFUNC)callbackTask, &priorityValue[ind]);

By using this approach, the problem should not arise in the future as a 
new architecture with a bigger or smaller stack requirement that gets 
added to Base can adjust the absolute amount of memory allocated to the 
three sizes.

My appologies if the wording in RELEASE_NOTES.html was confusing, and 
the Application Developers Guide entry for epicsThreadCreate() probably 
didn't help much either; I have just revised both documents.

- Andrew
--
There is considerable overlap between the intelligence of the smartest
bears and the dumbest tourists -- Yosemite National Park Ranger