Add a GenTL consumer interface

[xiaoqiangwang]

It makes aravis a GenTL consumer and work with GenICam cameras with non USB/Ethernet interfaces .

One breaking change is that GenTL requires its own start/stop acquisition calls, which have to be synched with ArvCamera object, i.e.

# prepare parameters
...
# start
arv_stream_start_acquisition
arv_camera_start_acquisition

# stop
arv_camera_stop_acquisition
arv_stream_stop_acquisition

I have tested the implementation with a Basler acA1300-200uc USB3 camera with Pylon SDK 7.3 on a Debian 12 x86_64 system. Both arv-tool and arv-viewer work normally. /~https://github.com/areaDetector/ADAravis can also work with this new implementation.

[xiaoqiangwang]

One more thing to note is arv-viewer crashes when loading Pylon 7.4 ProducerU3V.cti or ProducerGEV.cti

(gdb) bt
#0  0x00007ffff6cfcd3c in ?? () from /lib/x86_64-linux-gnu/libc.so.6
#1  0x00007ffff6cadf32 in raise () from /lib/x86_64-linux-gnu/libc.so.6
#2  0x00007ffff6c98472 in abort () from /lib/x86_64-linux-gnu/libc.so.6
#3  0x00007ffff5aba3ab in ?? () from /lib/x86_64-linux-gnu/libgcc_s.so.1
#4  0x00007ffff5acdff1 in _Unwind_RaiseException () from /lib/x86_64-linux-gnu/libgcc_s.so.1
BFD: reopening /tmp/shared.uxJxTn: No such file or directory
warning: Can't read data for section '.eh_frame' in file '/tmp/shared.uxJxTn'
#5  0x00007fffd557ea7c in ?? () from /tmp/shared.uxJxTn
#6  0x00007fffffffbea0 in ?? ()
#7  0x00007fffffffbee0 in ?? ()
#8  0x0000555555980780 in ?? ()
#9  0x00007fffd54ad2b0 in ?? () from /tmp/shared.uxJxTn
#10 0x00007fffd56547f0 in ?? () from /tmp/shared.uxJxTn
#11 0x0000555555bb7868 in ?? ()
#12 0x0000000000000000 in ?? ()

arv-tool and ADVimba do not have this problem. I did not dig into the exact cause yet. After a bit checking, a workaround is to preload any other pylon libraries. e.g.

$ LD_PRELOAD=/opt/pylon/lib/libpylonbase.so ./viewer/arv-viewer-0.8

A speculation is that, many pylon libraries have STL (std::string, std::thread ...) symbols builtin, but the GenTL producer libraries use dynamic linking. Maybe there are conflicts somewhere.

[EmmanuelP]

Hi @xiaoqiangwang ,

Thanks a lot. I just ran a quick test that worked fine using Basler U3VProvider. Nice work !

One breaking change is that GenTL requires its own start/stop acquisition calls, which have to be synched with ArvCamera object, i.e.

Question about that: is the execution of the Genicam feature AcquisitionStart supposed to start the acquisition on all the device datastream ? If so, we can keep a track off all created streams, and call stream_start_acquisition internally. You may use g_object_weak_ref from the ArvDevice to keep this list.

Also, it would be nice to add a arv_stream_create_and_push_buffers that will create the buffers knowing the datastream instance, and possibly help to avoid the memory copy. We will also need a arv_buffer_new_with_user_data for that purpose.

[xiaoqiangwang]

Question about that: is the execution of the Genicam feature AcquisitionStart supposed to start the acquisition on all the device datastream ?

I have not see the document says otherwise, but I assume it is true, because there is no knowing of datastreams in GenICam interface.

If so, we can keep a track off all created streams, and call stream_start_acquisition internally. You may use g_object_weak_ref from the ArvDevice to keep this list.

If I create two news ArvDevice methods, device_start/stop_acquisition, these two relay camera_start/stop_acquisition calls to underlying stream_start/stop_acquisition. Would that be Ok?

Also, it would be nice to add a arv_stream_create_and_push_buffers that will create the buffers knowing the datastream instance, and possibly help to avoid the memory copy. We will also need an arv_buffer_new_with_user_data for that purpose.

I understand this need but have no confidence in designing this API.

[EmmanuelP]

If so, we can keep a track off all created streams, and call stream_start_acquisition internally. You may use g_object_weak_ref from the ArvDevice to keep this list.

If I create two news ArvDevice methods, device_start/stop_acquisition, these two relay camera_start/stop_acquisition calls to underlying stream_start/stop_acquisition. Would that be Ok?

Yes, seems fine.

[EmmanuelP]

Also, it would be nice to add a arv_stream_create_and_push_buffers that will create the buffers knowing the datastream instance, and possibly help to avoid the memory copy. We will also need an arv_buffer_new_with_user_data for that purpose.

I understand this need but have no confidence in designing this API.

No problem. Once your work is merged, we can iterate and fix these sort of issues.

[xiaoqiangwang]

The change to start/stop_acquisition has been made. Since the relay is internalised, user API is unaffected. arvviewer.c and arvtest.c are unmodified and the tests pass. Thanks for the suggestion.

I have briefly tested the idea of adding buffers directly to GenTL producers, /~https://github.com/xiaoqiangwang/aravis/tree/gentl_buffer. But it turns out that Basler GenTL producers do not support adding buffers during acquisition. And GenTL standard does not make it mandatory.

For applications with a static set of buffers, that may work. But in this case, it constantly passes on filled buffers and created and queue new ones.

[EmmanuelP]

Do you still have pending work on this pull request ? Any known crash or bug ?

[xiaoqiangwang]

I am testing the event notification. The reading of event data is there. For the user API, I imagine a signal "device-event" on the device level.

We can also make the event notification another PR.

[xiaoqiangwang]

I have pushed the event notification part. arv-test has a new Event option for testing.

One problem with the event data nodes is that they stopped reading after the first event update. Is there a way to invalidate the node or force a new read?

[xiaoqiangwang]

@EmmanuelP just to be clear that this PR is ready for review. It passes the arv-test program and runs stable with ADAravis.

[EmmanuelP]

Ok. I will try to merge your work next week. Thanks again.

[EmmanuelP]

Hi @xiaoqiangwang

I have commited most of your changes in main. Events are disabled for now (the code is in, but commented out), I have still to think about the API. The changes in arv-test are not commited, the addition of the event testing in the transfer test code makes thing a bit too complicated to my taste.

Thanks again, nice piece of work !