sdp-testcase: reproducer for pt + profile-level-id parsing bug

[tmatth]

To run, simply build janus and:
./sdp-testcase

DO NOT MERGE

[lminiero]

Whoops, sorry, I pressed the wrong button and marked it ready to review. How to make it a draft again?

[lminiero]

Done, sorry for the noise 😄

[lminiero]

The issue is that there are multiple rtpmap attributes for the same codec, e.g.:

a=rtpmap:102 H264/90000
a=rtpmap:127 H264/90000
a=rtpmap:125 H264/90000
a=rtpmap:108 H264/90000

Since we parse the attributes in order, any time we find a matching rtpmap attribute we update the pt property: when looking for a H.264 profile, the last value of pt is 108, so that's the payload type we check when looking for fmtp attributes. That's also why you sometimes get the wrong pt back (there's indeed such a profile for that payload type) or -1 (no such profile for that specific payload type).

I'll have to think about the best way to address this. Rather than only keeping track of a single pt integer as we do now, we may have to start using a list of some sort (e.g., a GList of GINT_TO_POINTER elements) in janus_sdp_get_codec_pt_full, so that if, for instance, H.264 has 4 different payload types, we keep track of them all, and when we have to check fmtp, we accept any of those for our validation process. An alternative may be a GHashTable but that sounds heavier. I'll play with this a bit and keep you posted: in case I manage to get something that "fixes" your testcase, I'll prepare a PR that can be used to ensure it doesn't mess with anything else that was still working instead.

[lminiero]

Closing as I have your tester locally, and this doesn't really need to be merged. Thanks!