home *** CD-ROM | disk | FTP | other *** search
- # -*- Mode: Python -*-
- # vi:si:et:sw=4:sts=4:ts=4
-
- # discoverer.py
- # (c) 2005-2008 Edward Hervey <bilboed at bilboed dot com>
- # Discovers multimedia information on files
-
- # This library is free software; you can redistribute it and/or
- # modify it under the terms of the GNU Lesser General Public
- # License as published by the Free Software Foundation; either
- # version 2.1 of the License, or (at your option) any later version.
- #
- # This library is distributed in the hope that it will be useful,
- # but WITHOUT ANY WARRANTY; without even the implied warranty of
- # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
- # Lesser General Public License for more details.
- #
- # You should have received a copy of the GNU Lesser General Public
- # License along with this library; if not, write to the Free Software
- # Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
-
- """
- Class and functions for getting multimedia information about files
- """
-
- import os.path
-
- import gobject
-
- import gst
-
- from gst.extend.pygobject import gsignal
-
- class Discoverer(gst.Pipeline):
- """
- Discovers information about files.
- This class is event-based and needs a mainloop to work properly.
- Emits the 'discovered' signal when discovery is finished.
-
- The 'discovered' callback has one boolean argument, which is True if the
- file contains decodable multimedia streams.
- """
- __gsignals__ = {
- 'discovered' : (gobject.SIGNAL_RUN_FIRST,
- None,
- (gobject.TYPE_BOOLEAN, ))
- }
-
- mimetype = None
-
- audiocaps = {}
- videocaps = {}
-
- videowidth = 0
- videoheight = 0
- videorate = 0
-
- audiofloat = False
- audiorate = 0
- audiodepth = 0
- audiowidth = 0
- audiochannels = 0
-
- audiolength = 0L
- videolength = 0L
-
- is_video = False
- is_audio = False
-
- otherstreams = []
-
- finished = False
- sinknumber = 0
- tags = {}
-
-
- def __init__(self, filename, max_interleave=1.0, timeout=3000):
- """
- filename: str; absolute path of the file to be discovered.
- max_interleave: int or float; the maximum frame interleave in seconds.
- The value must be greater than the input file frame interleave
- or the discoverer may not find out all input file's streams.
- The default value is 1 second and you shouldn't have to change it,
- changing it mean larger discovering time and bigger memory usage.
- timeout: int; duration in ms for the discovery to complete.
- """
- gobject.GObject.__init__(self)
-
- self.mimetype = None
-
- self.audiocaps = {}
- self.videocaps = {}
-
- self.videowidth = 0
- self.videoheight = 0
- self.videorate = gst.Fraction(0,1)
-
- self.audiofloat = False
- self.audiorate = 0
- self.audiodepth = 0
- self.audiowidth = 0
- self.audiochannels = 0
-
- self.audiolength = 0L
- self.videolength = 0L
-
- self.is_video = False
- self.is_audio = False
-
- self.otherstreams = []
-
- self.finished = False
- self.tags = {}
- self._success = False
- self._nomorepads = False
-
- self._timeoutid = 0
- self._timeout = timeout
- self._max_interleave = max_interleave
-
- if not os.path.isfile(filename):
- self.debug("File '%s' does not exist, finished" % filename)
- self.finished = True
- return
-
- # the initial elements of the pipeline
- self.src = gst.element_factory_make("filesrc")
- self.src.set_property("location", filename)
- self.src.set_property("blocksize", 1000000)
- self.dbin = gst.element_factory_make("decodebin")
- self.add(self.src, self.dbin)
- self.src.link(self.dbin)
- self.typefind = self.dbin.get_by_name("typefind")
-
- # callbacks
- self.typefind.connect("have-type", self._have_type_cb)
- self.dbin.connect("new-decoded-pad", self._new_decoded_pad_cb)
- self.dbin.connect("no-more-pads", self._no_more_pads_cb)
- self.dbin.connect("unknown-type", self._unknown_type_cb)
-
- def _timed_out_or_eos(self):
- if (not self.is_audio and not self.is_video) or \
- (self.is_audio and not self.audiocaps) or \
- (self.is_video and not self.videocaps):
- self._finished(False)
- else:
- self._finished(True)
-
- def _finished(self, success=False):
- self.debug("success:%d" % success)
- self._success = success
- self.bus.remove_signal_watch()
- if self._timeoutid:
- gobject.source_remove(self._timeoutid)
- self._timeoutid = 0
- gobject.idle_add(self._stop)
- return False
-
- def _stop(self):
- self.debug("success:%d" % self._success)
- self.finished = True
- self.set_state(gst.STATE_READY)
- self.debug("about to emit signal")
- self.emit('discovered', self._success)
-
- def _bus_message_cb(self, bus, message):
- if message.type == gst.MESSAGE_EOS:
- self.debug("Got EOS")
- self._timed_out_or_eos()
- elif message.type == gst.MESSAGE_TAG:
- for key in message.parse_tag().keys():
- self.tags[key] = message.structure[key]
- elif message.type == gst.MESSAGE_ERROR:
- self.debug("Got error")
- self._finished()
-
- def discover(self):
- """Find the information on the given file asynchronously"""
- self.debug("starting discovery")
- if self.finished:
- self.emit('discovered', False)
- return
-
- self.bus = self.get_bus()
- self.bus.add_signal_watch()
- self.bus.connect("message", self._bus_message_cb)
-
- # 3s timeout
- self._timeoutid = gobject.timeout_add(self._timeout, self._timed_out_or_eos)
-
- self.info("setting to PLAY")
- if not self.set_state(gst.STATE_PLAYING):
- self._finished()
-
- def _time_to_string(self, value):
- """
- transform a value in nanoseconds into a human-readable string
- """
- ms = value / gst.MSECOND
- sec = ms / 1000
- ms = ms % 1000
- min = sec / 60
- sec = sec % 60
- return "%2dm %2ds %3d" % (min, sec, ms)
-
- def print_info(self):
- """prints out the information on the given file"""
- if not self.finished:
- return
- if not self.mimetype:
- print "Unknown media type"
- return
- print "Mime Type :\t", self.mimetype
- if not self.is_video and not self.is_audio:
- return
- print "Length :\t", self._time_to_string(max(self.audiolength, self.videolength))
- print "\tAudio:", self._time_to_string(self.audiolength), "\tVideo:", self._time_to_string(self.videolength)
- if self.is_video and self.videorate:
- print "Video :"
- print "\t%d x %d @ %d/%d fps" % (self.videowidth,
- self.videoheight,
- self.videorate.num, self.videorate.denom)
- if self.tags.has_key("video-codec"):
- print "\tCodec :", self.tags.pop("video-codec")
- if self.is_audio:
- print "Audio :"
- if self.audiofloat:
- print "\t%d channels(s) : %dHz @ %dbits (float)" % (self.audiochannels,
- self.audiorate,
- self.audiowidth)
- else:
- print "\t%d channels(s) : %dHz @ %dbits (int)" % (self.audiochannels,
- self.audiorate,
- self.audiodepth)
- if self.tags.has_key("audio-codec"):
- print "\tCodec :", self.tags.pop("audio-codec")
- for stream in self.otherstreams:
- if not stream == self.mimetype:
- print "Other unsuported Multimedia stream :", stream
- if self.tags:
- print "Additional information :"
- for tag in self.tags.keys():
- print "%20s :\t" % tag, self.tags[tag]
-
- def _no_more_pads_cb(self, dbin):
- self.info("no more pads")
- self._nomorepads = True
-
- def _unknown_type_cb(self, dbin, pad, caps):
- self.debug("unknown type : %s" % caps.to_string())
- # if we get an unknown type and we don't already have an
- # audio or video pad, we are finished !
- self.otherstreams.append(caps.to_string())
- if not self.is_video and not self.is_audio:
- self.finished = True
- self._finished()
-
- def _have_type_cb(self, typefind, prob, caps):
- self.mimetype = caps.to_string()
-
- def _notify_caps_cb(self, pad, args):
- caps = pad.get_negotiated_caps()
- if not caps:
- pad.info("no negotiated caps available")
- return
- pad.info("caps:%s" % caps.to_string())
- # the caps are fixed
- # We now get the total length of that stream
- q = gst.query_new_duration(gst.FORMAT_TIME)
- pad.info("sending duration query")
- if pad.get_peer().query(q):
- format, length = q.parse_duration()
- if format == gst.FORMAT_TIME:
- pad.info("got duration (time) : %s" % (gst.TIME_ARGS(length),))
- else:
- pad.info("got duration : %d [format:%d]" % (length, format))
- else:
- length = -1
- gst.warning("duration query failed")
-
- # We store the caps and length in the proper location
- if "audio" in caps.to_string():
- self.audiocaps = caps
- self.audiolength = length
- self.audiorate = caps[0]["rate"]
- self.audiowidth = caps[0]["width"]
- self.audiochannels = caps[0]["channels"]
- if "x-raw-float" in caps.to_string():
- self.audiofloat = True
- else:
- self.audiodepth = caps[0]["depth"]
- if self._nomorepads and ((not self.is_video) or self.videocaps):
- self._finished(True)
- elif "video" in caps.to_string():
- self.videocaps = caps
- self.videolength = length
- self.videowidth = caps[0]["width"]
- self.videoheight = caps[0]["height"]
- self.videorate = caps[0]["framerate"]
- if self._nomorepads and ((not self.is_audio) or self.audiocaps):
- self._finished(True)
-
- def _new_decoded_pad_cb(self, dbin, pad, is_last):
- # Does the file contain got audio or video ?
- caps = pad.get_caps()
- gst.info("caps:%s" % caps.to_string())
- if "audio" in caps.to_string():
- self.is_audio = True
- elif "video" in caps.to_string():
- self.is_video = True
- else:
- self.warning("got a different caps.. %s" % caps.to_string())
- return
- if is_last and not self.is_video and not self.is_audio:
- self.debug("is last, not video or audio")
- self._finished(False)
- return
- # we connect a fakesink to the new pad...
- pad.info("adding queue->fakesink")
- fakesink = gst.element_factory_make("fakesink", "fakesink%d-%s" %
- (self.sinknumber, "audio" in caps.to_string() and "audio" or "video"))
- self.sinknumber += 1
- queue = gst.element_factory_make("queue")
- # we want the queue to buffer up to the specified amount of data
- # before outputting. This enables us to cope with formats
- # that don't create their source pads straight away,
- # but instead wait for the first buffer of that stream.
- # The specified time must be greater than the input file
- # frame interleave for the discoverer to work properly.
- queue.props.min_threshold_time = int(self._max_interleave * gst.SECOND)
- queue.props.max_size_time = int(2 * self._max_interleave * gst.SECOND)
- queue.props.max_size_bytes = 0
-
- # If durations are bad on the buffers (common for video decoders), we'll
- # never reach the min_threshold_time or max_size_time. So, set a
- # max size in buffers, and if reached, disable the min_threshold_time.
- # This ensures we don't fail to discover with various ffmpeg
- # demuxers/decoders that provide bogus (or no) duration.
- queue.props.max_size_buffers = int(100 * self._max_interleave)
- def _disable_min_threshold_cb(queue):
- queue.props.min_threshold_time = 0
- queue.disconnect(signal_id)
- signal_id = queue.connect('overrun', _disable_min_threshold_cb)
-
- self.add(fakesink, queue)
- queue.link(fakesink)
- sinkpad = fakesink.get_pad("sink")
- queuepad = queue.get_pad("sink")
- # ... and connect a callback for when the caps are fixed
- sinkpad.connect("notify::caps", self._notify_caps_cb)
- if pad.link(queuepad):
- pad.warning("##### Couldn't link pad to queue")
- queue.set_state(gst.STATE_PLAYING)
- fakesink.set_state(gst.STATE_PLAYING)
- gst.info('finished here')
-