| Message ID | 20231130155323.13259-1-jaslo@ziska.de |
|---|---|
| Headers | show |
| Series |
|
| Related | show |
Hi Jaslo, Le jeudi 30 novembre 2023 à 16:43 +0100, Jaslo Ziska via libcamera-devel a écrit : > Hi, > > in this revision I mostly implemented Nicolas' suggestions namely > writing a comment on the locking behaviour for > gst_libcamera_src_negotiate and using the correct types and putting > the addition of the clearRequests function into a seperate commit. > Also the gst_libcamera_src_negotiate function now clears the > reconfigure flag and thus not all pads need to be iterated when > checking for this flag in the running task. > > Another thing I changed is that in processRequest in the > GST_FLOW_NOT_NEGOTIATED case it is now checked whether any pad > actually requested a renegotiation and if not an error is returned. Very nice work. For the entire series: Tested-by: Nicolas Dufresne <nicolas.dufresne@collabora.com> The test I have conducted is fairly minimal. I have tested the simple pipeline with UVC camera, and a single pad for now. For this purpose, I have patched GStreamer v4l2src test software: diff --git a/subprojects/gst-plugins-good/tests/examples/v4l2/v4l2src- renegotiate.c b/subprojects/gst-plugins-good/tests/examples/v4l2/v4l2src- renegotiate.c index cae28d2d66..d31fe08a1f 100644 --- a/subprojects/gst-plugins-good/tests/examples/v4l2/v4l2src-renegotiate.c +++ b/subprojects/gst-plugins-good/tests/examples/v4l2/v4l2src-renegotiate.c @@ -136,8 +136,8 @@ main (gint argc, gchar ** argv) loop = g_main_loop_new (NULL, FALSE); - desc = g_strdup_printf ("v4l2src name=src device=\"%s\" io-mode=\"%s\" " - "! capsfilter name=cf ! %s", device, io_mode, videosink); + desc = g_strdup_printf ("libcamerasrc name=src " + "! capsfilter name=cf ! %s", videosink); pipeline = gst_parse_launch (desc, &error); g_free (desc); if (!pipeline) { With good visual output and serial showing: $> v4l2src-renegotiate -s glimagesink Setting resolution to '320x240' [122:15:30.282196360] [691281] INFO IPAManager ipa_manager.cpp:143 libcamera is not installed. Adding '/home/nicolas/Sources/libcamera/build/src/ipa' to the IPA search path [122:15:30.282921702] [691281] INFO Camera camera_manager.cpp:284 libcamera v0.1.0+130-be11e882-dirty (2023-12-04T16:20:46-05:00) [122:15:30.803772071] [691293] INFO Camera camera.cpp:1183 configuring streams: (0) 320x240-YUYV Setting resolution to '1280x720' [122:15:34.232874025] [691293] INFO Camera camera.cpp:1183 configuring streams: (0) 1280x720-YUYV Setting resolution to '640x480' [122:15:37.238145227] [691293] INFO Camera camera.cpp:1183 configuring streams: (0) 640x480-YUYV [122:15:37.920471727] [691292] INFO V4L2 v4l2_videodevice.cpp:1820 /dev/video2[20:cap]: Zero sequence expected for first frame (got 2) We probably want a similar tool or test in libcamera itself though. It just a little difficult, since it so tied to the capabilities of the underneath hardware. If anyone have suggestions, this would be welcome, I would probably not block this meanwhile. If C is not a problem to folks here, I'd be happy to just copy over that modified example. Or I suppose its simple enough that it could be ported to C++. regards, Nicolas > > Regards, > > Jaslo > > Jaslo Ziska (3): > gstreamer: Move negotiation logic to separate function > gstreamer: Add GstLibcameraSrcState::clearRequests method > gstreamer: Implement renegotiation > > src/gstreamer/gstlibcamerasrc.cpp | 263 ++++++++++++++++++------------ > 1 file changed, 158 insertions(+), 105 deletions(-) > > -- > 2.43.0