[FFmpeg-devel] [PATCH] print message if probe failed due to probe conflict

[Reimar Döffinger]

On Fri, Sep 25, 2009 at 02:55:49PM +0200, Michael Niedermayer wrote:
> On Fri, Sep 25, 2009 at 02:41:40PM +0200, Reimar D?ffinger wrote:
> > On Fri, Sep 25, 2009 at 02:33:29PM +0200, Michael Niedermayer wrote:
> > > On Fri, Sep 25, 2009 at 02:11:38PM +0200, Reimar D?ffinger wrote:
> > > > If two probe functions return the same value, it is a bit hard to find
> > > > out which ones were involved and broke things.
> > > > So I propose this:
> > > 
> > > does your code work?
> > > it looks to me like it would print a warning if 3 demuxers return
> > > 1
> > > 1
> > > 90
> > > but not if they return 
> > > 1
> > > 90
> > > 1
> > > 
> > > simply printing all scores and piping through sort -n seems more usefull
> > > to me in this case ...
> > 
> > It was not supposed to print anything in both cases, if forgot to check
> > for fmt == NULL. I don't mean it as a test tool for conflicting probe
> > values, but as an actually useful message for users, only for the case
> > where probing actually fails.
> 
> Your code still does not look like it would work
> especially as a user understandable warning because when probe fails the
> probe buffer is increased until some limit, conflicts at > 25 can happen
> and trigger your warning while in the next round probe can succeed without
> conflict as one or more probes are than able to return more accurate
> scores.

That would only leave checking one level higher up, though that gives
hardly any useful information:
Index: libavformat/utils.c
===================================================================
--- libavformat/utils.c (revision 20013)
+++ libavformat/utils.c (working copy)
@@ -453,6 +453,7 @@
     /* Do not open file if the format does not need it. XXX: specific
        hack needed to handle RTSP/TCP */
     if (!fmt || !(fmt->flags & AVFMT_NOFILE)) {
+        int score;
         /* if no file needed do not try to open one */
         if ((err=url_fopen(&pb, filename, URL_RDONLY)) < 0) {
             goto fail;
@@ -462,7 +463,7 @@
         }
 
         for(probe_size= PROBE_BUF_MIN; probe_size<=PROBE_BUF_MAX && !fmt; probe_size<<=1){
-            int score= probe_size < PROBE_BUF_MAX ? AVPROBE_SCORE_MAX/4 : 0;
+            score= probe_size < PROBE_BUF_MAX ? AVPROBE_SCORE_MAX/4 : 0;
             /* read probe data */
             pd->buf= av_realloc(pd->buf, probe_size + AVPROBE_PADDING_SIZE);
             pd->buf_size = get_buffer(pb, pd->buf, probe_size);
@@ -484,6 +485,9 @@
                     av_log(logctx, AV_LOG_DEBUG, "Probed with size=%d and score=%d\n", probe_size, score);
             }
         }
+        if (!fmt && score > 0)
+            av_log(logctx, AV_LOG_WARNING, "No format match due to probe conflict with score %i\n", score);
+
         av_freep(&pd->buf);
     }
 

However maybe my previous patch as AV_LOG_DEBUG only as debugging help
would be better?
It would not catch all probe conflicts, only those that lead to probe
failure at a given buffer size, and it would "spam" in cases where probe
succeeds with larger probe buffer, but seems better than nothing.