Experimental Physics and
Industrial Control System

Hi John,

 

I have not been able to track down and fix that problem.  However, there is a trivial workaround.

 

The constructor syntax is:

 

pointGrey(const char *portName, int cameraId, int traceMask, int memoryChannel,

                 int maxBuffers, size_t maxMemory,

                 int priority, int stackSize);

 

These are the arguments that are passed via the iocsh pointGreyConfig command.

 

The “memoryChannel” argument specifies which set of parameters stored in non-volatile camera memory should be loaded at startup.  If this argument is zero then nothing is restored from non-volatile memory.  If it is greater than zero then that set of parameters is loaded.  The workaround is to simply use 1 for this argument, so that the first thing that is done is to load the first set of non-volatile parameters into the camera.

 

Here are the symptoms of the problem.

 

-          If memoryChannel 0 then the first time the IOC is run after power-cycling the camera it works fine.

-          The second time the IOC is run the driver fails to communicate with the camera part way through the initialization.

-          This problem only happens with the BlackFly camera, not the Grasshopper3 or the Flea3.

-          The problem only happens on Linux, not on Windows.

 

It seems that what is happening is that when the IOC exits on Linux it leaves the camera in a “bad state”.  Restoring the state from non-volatile memory in the constructor fixes the problem.  But why does it only happen on Linux?  I have written a test program that makes exactly the same calls to their library that the IOC does, and it does not have the problem.  But the test program is single-threaded.  Is there some code in their library that is not thread-safe?  But then why does it work on Windows?

 

There is really no down-side to the workaround, since the IOC typically overwrites all of the camera parameters from save/restore during iocInit.  So the fact that they were loaded from non-volatile memory should not matter.

 

Note that compared to the AVT (Prosilica) cameras I have found that the Point Grey GigE cameras tend to drop many more frames under the same network load and bandwidth usage.

 

Mark