gstreamer/mediainfo/README

= Media detail browser =
The package contains a UI for browsing media files and showing details of the
files.

= Browsing =
- in the preview hook of the file browser we run discover and update the details
- if one has grillo installed we could use grillo sources in addition to local files
- there should be a "open url" menu entry in addition

= Details =
- show container, video-tracks, audio-track, subtitle-trackes, metadata and
  parsing messages
- auto-play after updating the details
  have a menu command to turn auto-play off
- offer several commands in addition (see below)

example output from commandline tool:
> gst-discoverer -v /home/ensonic/Musik/xotox-hypnocat.mp3

Topology:
  audio: audio/mpeg, mpegversion=(int)1, mpegaudioversion=(int)1, layer=(int)3, rate=(int)44100, channels=(int)2, parsed=(boolean)true
    Codec:
      audio/mpeg, mpegversion=(int)1, mpegaudioversion=(int)1, layer=(int)3, rate=(int)44100, channels=(int)2, parsed=(boolean)true
    Additional info:
      None
    Channels: 2
    Sample rate: 44100
    Depth: 16
    Bitrate: 128000
    Max bitrate: 0
    Tags:
      taglist, audio-codec=(string)"MPEG\ 1\ Audio\,\ Layer\ 3\ \(MP3\)", has-crc=(boolean)false, channel-mode=(string)stereo, bitrate=(uint)128000;

Properties:
  Duration: 0:04:00.039125000
  Seekable: yes


== compare ==
- dup detail pane and allow to pick a new file
- show differing entries in red
- add a narrow vertical bar between the two details panes and mark blocks with a
  delta red
- diffing one (broken) file against several known working files would be nice
  - it would build ranges for the group of files (e.g. for the video width)
  - it would highlight fields that the are different to any of the working files

== analyse ==
- play the file by using fakesinks and gather statistics:
  - bitrate profiles for each a/mediainfo/v track
  - media specific:
    - audio: volume profile
    - video: contrast profile
  - show disconts as vertical bars

== structural view/hex view ==
- make parsing elements post structural messages (if enabled via property)
  - message would contain:
    - stream offset in bytes
    - block identifier (e.g. fourcc)
    - human readable description (if available)
    - flags (e.g. if the block is used or skipped)
    - indentation depth
  - we would need a way to indicate that e.g. h264parse would be indented below
    the containing container block
- a structured hex view would be nice
  (somehow align the parse tree with the hexdump)

== unsorted ==
- show named audio channel configurations instead only numbers
  e.g. "mono", "stereo", "5.1"
- handle chapters (once the patches are merged)
- tag lists
  - if there is a "language-code" in the tags (or subtitles?) use flag icons
    - deluge installs some under: /usr/share/pyshared/deluge/data/mediainfo/pixmaps/flags/
    - famfamfam-flag-png: locale/usr/share/flags/countries/
  - geo-tags: map-widget?, link to google-maps?
  - artist: links to {last.fm,wikipedia}
  - format dates nicely
- show level meters for audio next or below video drawable

== deep scan mode ==
- update fields when playing
  - listen for duration messages on the bus
- get bit-rate over time
  - specify window size, get min,max,avg bitrate for each window
  - gst-mi can draw them as a graph
- get key-frame statistics (using gst-index)
  - number of keyframes
  - min,max,avg keyframe interval
- disconts
- raw data statistics
  - audio: level, ...
  - video: histogram, ...

= TODO for gstreamer =
- file/stream layout
  - from every element we'd like to know what data is processes, what is pushed
    further and some metadata about it:
    layout {
      gsize offset;         // in bytes
      gsize length;         // in bytes
      gboolean known;       // or an enum: handled, skipped, unknown
      gchar *name;          // e.g. atom/chunk name
      gchar *description;   // long description or NULL
      enum block_type type; // meta, audio, video, text, ...
    };
    - offset is not neccesarily easy to determine for later elements, not sure
      if we can make it relative
  - elements could emit messages with this info
    - need a common way to enable it ("post-stream-layout" property)
  - we would need a cairo custom widget to draw a table
    - one row per element
    - each row contains colored segments

= TODO for discoverer =
- add deep-scan mode (see above)
  - add a mode property: quick-scan, deep-scan
   (or just a boolean for deep-scan that is false by default)
  - gst-discoverer will use "-d" for deep-scan/details
    (-a for anayze or -s for scan are used otherwise already)
  - in deep-scan, don't stop on handle_message::GST_MESSAGE_ASYNC_DONE
    - in deep-scan we need pad-probes for encoded data pads (to get bitrates)
      - uridecodebin_pad_added_cb() has raw pads :/
      - look at uridecodebin_element_added_cb()
    - the pads on uri-decodebin are sufficient to get keyframes and disconts
    - qtdemux could return stsz table to get bitrate profile
      http://wiki.multimedia.cx/index.php?title=QuickTime_container#stsz
- get duration per stream
  - this would need individual queries on the demuxer src pads
  - or duration as part of per stream tags
- errors/warnings about files/stream processing
  - we'd like to know about fixable, unfixable issues in the file/stream
  - many elements do this already (178 uses in 89 files)
  - the pipeline and thus the bus is internal to discoverer, so it would be nice
    if it could gather the messages and offer them
  - having them globaly should be enough
- CBR/VBR type for each stream
  - can we derive that from the bitrate tags (no min/max bitrate set)?
  find . -name "*.c" -exec egrep -Hn "GST_TAG[A-Z_]*_BITRATE" {} \;
  - formats
    - all raw files are cbr
    - some codecs are cbr only (some speech codecs)
    - audio
      - in an MP3 file, if there is no VBRI or Xing,
        it probably is not a vbr stream
      -
- lossy/lossless compression
  - all raw formats are lossless (well rgb->yuv is lossy, but ...)
  - some audio formats are lossless (flac, wavpack, ...)
  - some video formats can be lossless (dirac,

= discoverer workflow =
== sync quick-scan ==
app                   disco
---                   -----
 |    discover_uri()    |
 |--------------------->|
 |         done         |
 |<---------------------|
 |

== async quick-scan ==
app                   disco
---                   -----
 | discover_uri_async() |
 |--------------------->|
 |         ...          |
 | discover_uri_async() |
 |--------------------->|
 |       start()        |
 |--------------------->|
 |     ::discovered     |
 |<---------------------|
 |         ...          |
 |     ::discovered     |
 |<---------------------|
 |     ::finisheded     |
 |<---------------------|
 |

== sync deep-scan ==
- same as "sync quick-scan", no intermediate result

== async deep-scan ==
- same as "async quick-scan", but each discovered signal is followed by an
  "analyzed" signal with detailed information
  - we could also just emit discovered twice.

app                   disco
---                   -----
 | discover_uri_async() |
 |--------------------->|
 |         ...          |
 | discover_uri_async() |
 |--------------------->|
 |       start()        |
 |--------------------->|
 |     ::discovered     |
 |<---------------------|
 |     ::analyzed       |
 |<---------------------|
 |         ...          |
 |     ::discovered     |
 |<---------------------|
 |     ::analyzed       |
 |<---------------------|
 |     ::finisheded     |
 |<---------------------|
 |