0 size buffers in processBuffer() when capturing 'Single' or 'Multiple'

[daykin]

Discovered by testing fix for #18 (not related at all, to be clear):

Environment:
Arch: linux-x86_64-debug
EPICS Base 7.0.7
ADGenICam R1-8, master commit 639aa4f
ADAravis R2-3, master commit ebca7d0
Camera: The Imaging Source Europe DMK 33GX174, via gigabit ethernet
Additional Plugins: Only NDStdArrays Image plugin for viewing output array in CS-Studio

Steps to reproduce:

1. Generate xml and database from arv-tool-0.8 genicam per the documentation
2. Point build system to appropriate EPICS_BASE, ADCORE, GENICAM modules in configure/RELEASE
3. Add in the ADBase and NDFile settings .req files to aravisApp/Db/aravisCamera_settings.req
4. Build everything
5. Start up an IOC, invoking aravisConfig(portName, aravisCameraName)
6. load in aravisCamera.template (and, by extension, ADGenICam.template and ADBase.template)
7. load in camera-specific records generated in step 1
8. Change $(P)$(R):DataType to UInt16
9. RegionSize and PixelFormat remain Mono16, no conversion or shift
10. GainAuto and ExposureAuto set to 'Continuous'
11. AcquirePeriod set to 1 second
12. NumImages set to 100
13. ImageMode set to 'Multiple'
14. Acquire

Result:
ArraySize_RBV Remains zero. In IOC, the following output occurs:
2023/03/21 15:30:00.404 ADAravis:processBuffer: w: 1920, h: 1200, size: 0, expected_size: 4608000

If ImageMode is set to 'Continuous' the error goes away.

[daykin]

Additional info:
Upon starting acquisition in 'single' or 'multiple', the following error occurs:
** (process:2297535): CRITICAL **: 16:02:20.339: arv_gc_integer_get_value: assertion 'error == NULL || *error == NULL' failed

[MarkRivers]

That sounds to me like your camera may not support "single" or "multiple" modes. Open the camera-specific screens and look for the GenICam feature called AcquisitionMode. What choices do you see there?

[daykin]

Ha, I was literally in the process of typing that: This camera does not support any mode besides AcquisitionMode=Continuous. would it be worthwhile to try to add a better check for that?

[MarkRivers]

Returning an error if one attempts to set a non-supported AcquisitionMode and leaving the mode unchanged would be a good idea. There is already some logic that converts Multiple to Continuous for cameras that don't support Multiple:

https://github.com/areaDetector/ADGenICam/blob/639aa4f51fe233e311f22c8f61ed6138b0d5a442/GenICamApp/src/GenICamFeature.cpp#L546

That section of code could be enhanced to do something similar if Single or Multiple or both are unsupported.

I have no way to test that, but I am happy to accept a PR if you want to test it.

[jeinstei]

I'm with @MarkRivers on this, instead of automatically switching everything. This is a low-level error as the buffer size mismatch is throwing an error. To do this right, wouldn't we want a "bad buffer" or "error count" readback instead? Or a "last error" string PV?