On Tue, 20 May 2025, Timothy Allen via ffmpeg-devel wrote:
Good day I wanted to offer a discussion of the referenced patch. I have found that, when a link in an extended M3U file (as used by HLS) includes a colon, FFmpeg will fail to load the file. The bug has already been reported: https://trac.ffmpeg.org/ticket/10679 The error reads: [hls @ 0x78dea4000c80] Failed to open segment 0 of playlist 0 [hls @ 0x78dea4000c80] Segment 0 of playlist 0 failed too many times, skipping The referenced patch fixes the issue. However, it is worth noting that the patch changes the behavior of one of the unit tests: - http://a/b/c/d;p?q g:h => g:h + http://a:b/c/d;p?q e => http://a:b/c/e The original unit test derives from the following two trac tickets: https://trac.ffmpeg.org/ticket/8813 https://trac.ffmpeg.org/ticket/8814 This is a breaking change, and, in particular, violates one specific example given in the rfc at https://tools.ietf.org/html/rfc3986#section-5.4 (the first example). In particular, it will affect cases where a URL consists only of host:port, with no scheme or path, and the base URL links to an unrelated host. Where before, the new link would take the form of "host:port", the new link will now use "host:port" as the last element of the path. I believe that this behaviour is more intuitive given modern use of URIs (and is replicated in many browsers), but I recognise this is a matter of taste.
Sorry, but I'd rather not break RFC compliant parsing because of problematic files. Feel free however to add some more context to the ticket, like which browsers and/or media players support such a file, which don't. And how the ones which do support it distinguish between an URL scheme and a relative URL with a colon in the first path component, e.g.
video:first.ts or http:first.ts Thanks, Marton _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
