Add support for videoconformancesink

[ndufresne]

This PR is in preparation for GStreamer 1.20 and the new videoconformancesink being accepted. This is sent early, so this can get more attention and testing. The patch will check GStreamer version and use the conformance sink if GStreamer is new enough. With the new videoconformancesink, GStreamer padding is ignored, that padding was breaking some V8 and VP9 tests.

Pending GStremaer MR: https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2287

[pamarcos]

Thanks a lot for the PR :)
LGTM. I added a little and fussy comment, though.

[ndufresne]

Ok, I believe I fixed the first round of review. We settle to videocodectestsink upstream, so I updated the code accordingly also.

[pamarcos]

LGTM, apart of the fussy comment. Nice work 👍
The gst_element_exists you implemented with lru_cache is exactly what I meant

[ndufresne]

@pamarcos ok, added the "safety" return, and simplified the commit message a bit with only what's relevant to know from a testing point of view.

[pamarcos]

Perfect @ndufresne! Thanks a mill for your contribution 💪

[ndufresne]

My next goal is to setup upstream GStreamer CI for software VP8 decode, as this one is passing, I should be able to setup CI for VP9 as well after figuring the one remaining issue, for H.264/H.265, we'll need to think of a clean way to mark as known to fail some of the tests, as it seems we'll never fully pass these, and for HW, we have to deal with HW errors, so such system could be reuse, we basically want to catch regression (or improvement).

[pamarcos]

My next goal is to setup upstream GStreamer CI for software VP8 decode, as this one is passing, I should be able to setup CI for VP9 as well after figuring the one remaining issue, for H.264/H.265, we'll need to think of a clean way to mark as known to fail some of the tests, as it seems we'll never fully pass these, and for HW, we have to deal with HW errors, so such system could be reuse, we basically want to catch regression (or improvement).

What we use for our CI are the --threshold to ensure a minimum amount of tests and the --time-threshold to ensure no performance regressions. Finally, we have stored known fluster results for specifc nodes in a repo. Then, our CI checks that there is no diff with them, except for known racy/fluke tests. Something like:

// Check for differences, ignoring extra blank lines, CRLF line endings and the test name,
// because we reuse the same output for different decoders in H.265
def diffParams = '-uB -I"|Test|" --strip-trailing-cr'
diffParams += ' -I"|FM1"' // TODO: remove once OCP-1853 is fixed
sh "diff $diffParams expected.out ${this.repoDir}/summary.log"

The expected.out taken from the repo is the output of fluster for that particular decoder when generating the summary. When comparing, we only take into account changes in the result of tests in the summary. This has helped us so far to achieve stability, catching regressions earlier in the dev process.

[ndufresne]

This is gold, thanks.