Issue 51 - release GIL while calling ndiCommand - #52
Conversation
|
I just ran this on our SmartLiver clinical system. Without this fix, when I set the tracking thread to 60fps, the UI thread grinds down to about 4 fps. When I have this fix, the UI thread is stable at about 16-17 fps (on a Qt timer set to 20 fps, so - about what I would expect). So, while I didnt check that tracking transforms had changed or not, it did improve concurrency. |
|
Looks straightforward, but I'm not familiar with threading in python. Should this be an #ifdef compile option in case it breaks things for other users? If you're able to test/confirm that tracking is unaffected, that would be ideal. |
| } | ||
| else | ||
| { | ||
| Py_DECREF(initial); |
There was a problem hiding this comment.
Again, not my expertise, but why are these two reference decreases repeated? Is this a separate bug being fixed?
There was a problem hiding this comment.
I have to be honest and say it was suggested by AI as a separate bug, so I left it in there. I don't really know how to test it in a reasonable time frame? Is there a simple test harness, where we could output the ref count while tracking?
OK. My next opportunity is Thursday. |
2 participants
Hi @thompson318, @Sunderlandkyl, @adamrankin.
Who's best placed to look at this?
In terms of testing - i just ran it on an NDI Vega. It seemed to still track, but I did not have full access to an application to test it on. I'll try to get access to a Spectra shortly.
Kind regards
Matt