Recherche avancée

Sur d’autres sites (13017)

• avformat_seek_file timestamps not using the correct time base

  19 juin 2021, par Charlie

  I am in the process of creating a memory loader for ffmpeg to add more functionality. I have audio playing and working, but am having an issue with avformat_seek_file timestamps using the wrong format.

  


  avformat.avformat_seek_file(file.context, -1, 0, timestamp, timestamp, 0)


  


  From looking at the docs it says if the stream index is -1 that the time should be based on AV_TIME_BASE. When I load the file through avformat_open_input with a null AVFormatContext and a filename, this works as expected.

  


  However when I create my own AVIOContext and AVFormatContext through avio_alloc_context and avformat_alloc_context respectively, the timestamps are no longer based on AV_TIME_BASE. When testing I received an access violation when I first tried seeking, and upon investigating, it seems that the timestamps are based on actual seconds now. How can I make these custom contexts time based on AV_TIME_BASE ?

  


  The only difference between the two are the custom loading of AVIOContext and AVFormatContext :

  


      data = fileobject.read()

    ld = len(data)

    buf = libavutil.avutil.av_malloc(ld)
    ptr_buf = cast(buf, c_char_p)

    ptr = ctypes.create_string_buffer(ld)
    memmove(ptr, data, ld)

    seeker = libavformat.ffmpeg_seek_func(seek_data)
    reader = libavformat.ffmpeg_read_func(read_data)
    writer = libavformat.ffmpeg_read_func(write_data)

    format = libavformat.avformat.avio_alloc_context(ptr_buf, buf_size, 0,
                                                     ptr_data,
                                                     reader,
                                                     writer,
                                                     seeker
                                                     )

    file.context = libavformat.avformat.avformat_alloc_context()
    file.context.contents.pb = format
    file.context.contents.flags |= AVFMT_FLAG_CUSTOM_IO

    result = avformat.avformat_open_input(byref(file.context),
                                          b"",
                                          None,
                                          None)

    if result != 0:
        raise FFmpegException('avformat_open_input in ffmpeg_open_filename returned an error opening file '
                              + filename.decode("utf8")
                              + ' Error code: ' + str(result))

    result = avformat.avformat_find_stream_info(file.context, None)
    if result < 0:
        raise FFmpegException('Could not find stream info')

    return file



  


  Here is the filename code that does work :

  


      result = avformat.avformat_open_input(byref(file.context),
                                          filename,
                                          None,
                                          None)
    if result != 0:
        raise FFmpegException('avformat_open_input in ffmpeg_open_filename returned an error opening file '
                              + filename.decode("utf8")
                              + ' Error code: ' + str(result))

    result = avformat.avformat_find_stream_info(file.context, None)
    if result < 0:
        raise FFmpegException('Could not find stream info')

    return file


  


  I am new to ffmpeg, but any help fixing this discrepancy is greatly appreciated.

  


• Back on the Salty Track

  12 juin 2011, par Multimedia Mike — General

  After I posted about my initial encounter and frustration with Google’s Native Client (NaCl) SDK and took a deep breath, I realized that I achieved an important proof of concept— I successfully played music using the NaCl SDK audio output interface. Then I started taking a closer read through the (C-based set of) header files and realized I might be able to make a go of it after all. I had much better luck this time and managed to create a proper Native Client interface that allows for controlling playback, presenting metadata, and toggling individual voices (a fascinating tool for studying classic game music).

  I haven’t bothered to post the actual plugin because, really, what’s the point ? I started with NaCl SDK 0.3 which requires Chrome 12, which means terribly limited reach, even among Chrome users. At least, that was true when I restarted this little project. Chrome 12 was formally released this past week. Chrome development really does move at breakneck pace.

  Anyway, here is a static screenshot of what the plugin currently looks like :

  
  
  

  Not pretty, but it does the job.

  Dev Journal
  Various notes based on this outing :

  • Portability : I tested my plugin using Chrome 12 on 64-bit Windows, Mac, and Linux. Mac and Linux both work ; Windows does not.
  • Build System : SDK 0.3 is still lacking in its ability to compile .cpp files (instead of .cc files) ; necessary because libgme is C++ using .cpp files. This requires some build system modification.
  • Getting the interfaces : This is where I got tripped up the first time around. get_browser_interface() from their example actually refers to a parameter passed in through the PPP_InitializeModule() function. The SDK’s template generator renames this to get_browser().
  • Debugging : I feel unstoppable once I have a printf() mechanism available to me during development. To that end, console.log() from JavaScript outputs to Chrome’s built-in JavaScript console log while putting printf() statements in the actual NaCl plugin causes the messages to show up in /.xsession-errors on Linux/X.
  • Size Matters : The binaries generated with the NaCl 0.3 SDK are ridiculously huge. The basic "Hello World" example in C compiles to binaries that are 6.7 MB and 7.8 MB for the 32- and 64-bit builds, respectively. This made me apprehensive to build a full version of SaltyGME that contains all the bells and whistles offered by the library. However, all of the GME code compiled into the binary adds very little size. Curiously, the C++ version of "Hello World" only ranges from 1.8-2.0 MB for 32- and 64-bit. Is there some kind of C tax happening here ? Note that running ’strip’ on the resulting .nexe files (they’re ELF files, after all) brings the sizes down into the C++ range, but at the cost of causing them to not work (more specifically, not even load).
  • No Messaging : The NaCl SDK is supposed to have a messaging interface which allows the NaCl plugin to send asynchronous messages up to the hosting page. When I try to instantiate it, I get a NULL. I’m stuck with the alternative of polling from the JavaScript side to, e.g., determine when a song has finished loading via the network.

  That’s all I can think of for now. I may work on this a little more (I’d like to at least see some audio visualization). Maybe Google will enable NaCl per default sometime around Chrome 21 and this program will be ready for prime time by then.

  See Also :

  • My first attempt at this same project

• ffmpeg in child process doesn't exit when parent closes pipe

  28 mars 2013, par user2221129

  This code snippet is from a pthread process. It is responsible for reading configuration/options to pass to ffmpeg. The data piped to ffmpeg is coming in on a ring buffer of video frames (as a proof of concept, in the final implementation these would be coming from a camera device). The parent writes to the child via a pipe established with dup2. The problem I'm having is that most of the time, the ffmpeg process doesn't seem to recognize the pipe has been closed by the parent (as strace shows the ffmpeg is in read). There are probably some unhandled mutex situations.

  #define CHILD_READ  writepipe[0]
  
  #define PARENT_WRITE    writepipe[1]
  
  #define MAX_VIDEO_BUFFERS 30
  
  
  
  static int last_frame_written = -1;
  
  static int buffer_size = -1;
  
  static void *video_buffer[MAX_VIDEO_BUFFERS];
  
  
  
  #define MAXTHREADS 2
  
  static pthread_t thread[MAXTHREADS];
  
  static int sigusr[MAXTHREADS];
  
  static int sigusr_thread[MAXTHREADS];
  
  
  
  #define MAXPARAMETERS 100
  
  
  
  void* encoder_thread(void *ptr)
  
  {
  
      int param;
  
      const char **parameters = new const char* [MAXPARAMETERS];
  
      ssize_t bytes_written_debug = 0;
  
  
  
      int writepipe[2] = {-1,-1};
  
      int last_frame_read = -1;
  
      pid_t   childpid;
  
  
  
      // xxx
  
      if ( 0 != pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL) )
  
      {
  
          fprintf(stderr,"pthread_setcancelstate could not be set\n");
  
      }
  
  
  
      param = *(int*)ptr;
  
      fprintf(stderr,"param = %d\n", param);
  
  
  
      switch (param)
  
      {
  
          ...read config file and populate parameters for child call...
  
      }
  
  
  
      while ( last_frame_written == -1 )
  
      {
  
          fprintf(stderr,"ENCODER THREAD WAITING\n");
  
          sleep(1);
  
      }
  
  
  
      if ( pipe(writepipe) < 0 )
  
      {
  
          ...handle error case...
  
      }
  
  
  
  
  
      if ( (childpid = fork()) < 0)
  
      {
  
          fprintf(stderr,"child failed\n");
  
      }
  
      else if ( 0 == childpid )   /* in the child */
  
      {
  
          dup2(CHILD_READ,  STDIN_FILENO);
  
          //close(CHILD_READ); // doesn't seem to matter
  
          close(PARENT_WRITE);
  
  
  
          execv("/usr/bin/ffmpeg",(char **)parameters);
  
          return;
  
      }
  
      else                /* in the parent */
  
      {
  
          fprintf(stderr,"THREAD CHILD PID: %d\n",childpid);
  
          close(CHILD_READ);
  
  
  
          while ( last_frame_written > -1 && 0==sigusr_thread[param] )
  
          {
  
              if ( last_frame_read != last_frame_written )
  
              {
  
                  // send frame to child (ffmpeg)
  
                  last_frame_read = last_frame_written;
  
                  bytes_written_debug = write(PARENT_WRITE,video_buffer[last_frame_read], buffer_size);
  
                  if ( bytes_written_debug < 0 )
  
                  {
  
                      fprintf(stderr, "write error\n");
  
                      break;
  
                  }
  
              }
  
  
  
              usleep(10000); // assume ~100Hz rate
  
          }
  
  
  
  /// Problems begin now: no more data from main process,
  
  /// Shouldn't this close "close" the pipe created with dup2 and hence EOF ffmpeg?
  
          if ( close(PARENT_WRITE) )
  
          {
  
              fprintf(stderr,"\n\nclose failure! wtf?\n\n\n");
  
          }
  
  
  
  // debug sleep:
  
  // waiting 10 sec doesn't seem to help...
  
  // trying to give ffmpeg some time if it needs it
  
          sleep(10);
  
  
  
  // kill never fails here:
  
          if ( !kill(childpid,SIGINT) )
  
          {
  
              fprintf(stderr, "\n\nkill child %d\n", childpid);
  
          }
  
          else
  
          {
  
              fprintf(stderr, "\n\nkill failed for child %d\n", childpid);
  
          }
  
  
  
          fprintf(stderr,"\n\nwaiting\n\n\n");
  
          int status = 0;
  
  
  
  // wait forever (in most cases); ps -axw shows all the ffmpeg's
  
          wait(&status);
  
      }
  
  }
  
  
  
  
  
  
  
  
  
  int main()
  
  {
  
      ...
  
      for ( int i = 0; i < MAXTHREADS; ++i)
  
      {
  
          sigusr[i] = 0;
  
          sigusr_thread[i] = 0;
  
          param[i] = i;
  
          iret = pthread_create( &thread[i], NULL, encoder_thread, (void*) &param[i] );
  
      }
  
      ...
  
  
  
      // Maybe this is important to the pthread signaling?
  
      // Don't think so because thread's SIGINT isn't picked up by signal handler (not shown)
  
      sigemptyset(&set);
  
      sigaddset(&set, SIGHUP);
  
      sigaddset(&set, SIGINT);
  
      sigaddset(&set, SIGUSR1);
  
      sigaddset(&set, SIGUSR2);
  
      sigaddset(&set, SIGALRM);
  
  
  
      /* block out these signals */
  
      sigprocmask(SIG_BLOCK, &set, NULL);
  
  
  
      ...read file/populate frame buffer until file exhausted...
  
  
  
      fprintf(stderr, "waiting for threads to exit\n");
  
  
  
      for ( int i = 0; i < MAXTHREADS; ++i)
  
      {
  
          sigusr_thread[i] = 1;
  
          pthread_join( thread[i], NULL);
  
      }
  
      fprintf(stderr, "done waiting for threads to exit\n");
  
  
  
      ...
  
  
  
      return 0;
  
  }

  I've embedded comments and questions in the thread's parent code after the frame buffer read/write loop.

  Hope I've provided enough detail - stackoverflow posting noob ! Thank you.