webm_chunk muxer is broken since commit 18ac64235939c4c5c7656546a9545f68339affbe

[octop]

The commit 18ac64235939c4c5c7656546a9545f68339affbe introduced a regression, making the webm_chunk muxer cease to function properly.

libavformat/webm_chunk.c
@@ -127,12 +127,12 @@ static int webm_chunk_write_header(AVFormatContext *s)
     if (ret < 0)
         return ret;
     oc = wc->avf;
-    ret = get_chunk_filename(s, 1, oc->filename);
+    ret = get_chunk_filename(s, 1, oc->url);
     if (ret < 0)
         return ret;
     if (wc->http_method)
         av_dict_set(&options, "method", wc->http_method, 0);
-    ret = s->io_open(s, &oc->pb, oc->filename, AVIO_FLAG_WRITE, &options);
+    ret = s->io_open(s, &oc->pb, oc->url, AVIO_FLAG_WRITE, &options);
     av_dict_free(&options);
     if (ret < 0)
         return ret;

In my own testing, I discovered that oc->url in that code evaluates to null, making the muxer return prematurely.

To reproduce:
ffmpeg -re -f lavfi -i sine -map 0 -strict -2 -c:a opus -f webm_chunk -header webm_live_sine.hdr webm_live_sine_%d.chk

[Marton Balint]

Could you try the attached patch?

[octop]

The muxer works again after applying your patch, thank you. So the issue was caused by trying to "dynamically" create the nonexistent url field in that AVFormatContext *oc structure?

[Marton Balint]

Yeah, the URL field (which is a replacement for the deprecated filename field) must be dynamically allocated, so it cannot be used as a static (fixed-length) buffer. I will send this patch to ffmpeg-devel and apply and backport it in a few days.

[Marton Balint]

Fixed in 2dbe936bf7f9e0fe7e8f05e5c3b78fb1afbff164.