3 ### Convert a directory tree of audio files
5 ### (c) 2010 Mark Wooding
8 ###----- Licensing notice ---------------------------------------------------
10 ### This file is part of the `autoys' audio tools collection.
12 ### `autoys' is free software; you can redistribute it and/or modify
13 ### it under the terms of the GNU General Public License as published by
14 ### the Free Software Foundation; either version 2 of the License, or
15 ### (at your option) any later version.
17 ### `autoys' is distributed in the hope that it will be useful,
18 ### but WITHOUT ANY WARRANTY; without even the implied warranty of
19 ### MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
20 ### GNU General Public License for more details.
22 ### You should have received a copy of the GNU General Public License
23 ### along with `autoys'; if not, write to the Free Software Foundation,
24 ### Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
26 ###--------------------------------------------------------------------------
27 ### External dependencies.
30 from __future__ import with_statement
32 ## Standard Python libraries.
37 import unicodedata as UD
42 import threading as TH
44 from math import sqrt, ceil
45 from contextlib import contextmanager
47 ## eyeD3 tag fettling.
52 gi.require_version('GLib', '2.0'); from gi.repository import GLib as G
53 gi.require_version('Gio', '2.0'); from gi.repository import Gio as GIO
54 gi.require_version('Gst', '1.0'); from gi.repository import Gst as GS
58 from PIL import Image as I
63 ###--------------------------------------------------------------------------
64 ### Special initialization.
71 ###--------------------------------------------------------------------------
72 ### Eyecandy progress reports.
76 Return the width of S, in characters.
78 Specifically, this is the number of backspace characters required to
79 overprint the string S. If the current encoding for `stdout' appears to be
80 Unicode then do a complicated Unicode thing; otherwise assume that
81 characters take up one cell each.
83 None of this handles tab characters in any kind of useful way. Sorry.
86 ## If there's no encoding for stdout then we're doing something stupid.
87 if SYS.stdout.encoding is None: return len(s)
89 ## Turn the string into Unicode so we can hack on it properly. Maybe that
90 ## won't work out, in which case fall back to being stupid.
91 try: u = s.decode(SYS.stdout.encoding)
92 except UnicodeError: return len(s)
94 ## Our main problem is combining characters, but we should also try to
95 ## handle wide (mostly Asian) characters, and zero-width ones. This hack
96 ## is taken mostly from http://www.cl.cam.ac.uk/~mgk25/ucs/wcwidth.c
100 if UD.category(ch) in ['Cf', 'Me', 'Mn'] or \
101 0x1160 <= cd <= 0x11ff: pass
102 elif UD.east_asian_width(ch) in ['F', 'W']: w += 2
108 class StatusLine (object):
110 Maintains a status line containing ephemeral progress information.
112 The status line isn't especially important, but it keeps interactive users
115 There should be only one status line object in your program; otherwise
116 they'll interfere with each other and get confused.
118 The update algorithm (in `set') is fairly careful to do the right thing
119 with long status `lines', and to work properly in an Emacs `shell' buffer.
123 "Initialize the status line."
126 me.eyecandyp = OS.isatty(SYS.stdout.fileno())
130 Set the status line contents to LINE, replacing what was there before.
132 This only produces actual output if stdout is interactive.
139 ## If the old line was longer, we need to clobber its tail, so work out
140 ## what that involves.
142 b = charwidth(me._last[n:])
147 ## Now figure out the length of the common prefix between what we had
148 ## before and what we have now. This reduces the amount of I/O done,
149 ## which keeps network traffic down on SSH links, and keeps down the
150 ## amount of work slow terminal emulators like Emacs have to do.
152 m = min(n, me._lastlen)
153 while i < m and line[i] == me._last[i]:
156 ## Actually do the output, all in one syscall.
157 b = charwidth(me._last[i:])
158 SYS.stdout.write(pre + '\b'*b + line[i:])
161 ## Update our idea of what's gone on.
166 "Clear the status line. Just like set('')."
169 def commit(me, line = None):
171 Commit the current status line, and maybe the string LINE.
173 If the current status line is nonempty, then commit it to the transcript.
174 If LINE is not None, then commit that to the transcript too.
176 After all of this, we clear the status line to get back to a clean state.
180 SYS.stdout.write('\n')
182 SYS.stdout.write(me._last + '\n')
184 SYS.stdout.write(line + '\n')
188 STATUS = StatusLine()
190 def filestatus(file, status):
191 return '%s%s: %s' % (' '*8, OS.path.basename(file), status)
193 class ProgressEyecandy (object):
195 Provide amusement while something big and complicated is happening.
197 This is an abstract class. Subclasses must provide a method `progress'
198 returning a pair (CURRENT, MAX) indicating the current progress through the
202 def __init__(me, what, silentp = False):
204 Initialize a progress meter.
206 WHAT is a prefix string to be written before the progress eyecandy
210 me._silentp = silentp
214 def _fmt_time(me, t):
215 "Format T as a time, in (maybe hours) minutes and seconds."
216 s, t = t % 60, int(t/60)
217 m, h = t % 60, int(t/60)
219 return '%d:%02d:%02d' % (h, m, s)
221 return '%02d:%02d' % (m, s)
224 "Show the current level of progress."
226 ## If we're not showing pointless frippery, don't bother at all.
227 if not STATUS.eyecandyp:
230 ## Update the spinner index.
231 me._spinner = (me._spinner + 1)%4
233 ## Fetch the current progress information. Note that we always fetch
234 ## both the current and maximum levels, because both might change if an
235 ## operation revises its idea of how much work needs doing.
236 cur, max = me.progress()
238 ## If we couldn't get progress information, display something vaguely
240 if cur is None or max is None:
241 STATUS.set('%s %c [unknown progress]' %
242 (me._what, r'/-\|'[me._spinner]))
245 ## Work out -- well, guess -- the time remaining.
248 eta = me._fmt_time(ceil((t - me._start)*(max - cur)/cur))
252 ## Set the status bar.
254 STATUS.set('%s %c [%s%s] %3d%% (%s)' % \
256 r'/-\|'[me._spinner],
261 def done(me, win = True):
262 "Show a completion notice, or a failure if WIN is false."
264 STATUS.set('%s FAILED!' % me._what)
265 elif not me._silentp:
266 STATUS.set('%s done (%s)' %
268 me._fmt_time(T.time() - me._start)))
273 ###--------------------------------------------------------------------------
274 ### Timeout handling.
276 KILLSWITCH = TH.Event()
282 moan('dying messily due to timeout')
285 ###--------------------------------------------------------------------------
286 ### Parsing utilities.
288 ## Allow hyphens in identifiers.
289 IDCHARS = P.alphanums + '-_'
290 P.Keyword.setDefaultKeywordChars(IDCHARS)
292 ## Some common kinds of tokens.
293 Name = P.Word(IDCHARS)
294 Num = P.Word(P.nums).setParseAction(lambda toks: map(int, toks))
295 String = P.QuotedString('"', '\\')
297 ## Handy abbreviations for constructed parser elements.
298 def K(k): return P.Keyword(k).suppress()
299 def D(d): return P.Literal(d).suppress()
300 def R(p): return P.ZeroOrMore(p).setParseAction(lambda s, l, t: [t])
303 ###--------------------------------------------------------------------------
304 ### Format identification and conversion.
306 class IdentificationFailure (Exception):
309 class FileCategory (object):
311 A FileCategory represents a class of files.
313 For example, it's sensible to consider audio, or image files as a
314 category. A file category knows how to recognize member files from
318 def __init__(me, name, mime_pats, ident):
320 Construct a new category.
322 The PATS are a list of `fnmatch' patterns to be compared with a MIME
323 type. The IDENT is a function which produces an identification object
324 given a file's name and first-guess MIME type. The object is passed to a
325 Format's `check' method to see whether a file needs re-encoding, and to
326 `convert' to assist with the conversion.
328 An identification object must have an attribute `mime' which is a set of
329 possible MIME types accumulated for the object.
332 me._mime_pats = mime_pats
334 CATEGORYMAP[name] = me
336 def identify(me, file, mime):
338 Attempt to identify FILE, given its apparent MIME type.
340 If identification succeeds, return an identification object which can be
341 used by associated file formats; otherwise return None.
343 for p in me._mime_pats:
344 if not FN.fnmatchcase(mime, p):
347 return me._ident(file, mime)
348 except IdentificationFailure:
352 class BaseFormat (object):
354 A BaseFormat object represents a particular encoding and parameters.
356 The object can verify (the `check' method) whether a particular file
357 matches its requirements, and if necessary (`encode') re-encode a file.
359 Subclasses should define the following methods.
362 Answer whether the file identified by ID is acceptable according to
363 the receiver's parameters.
365 convert(MASTER, ID, TARGET)
366 Convert the file MASTER, which has been identified as ID, according
367 to the receiver's parameters, writing the output to TARGET.
369 Subclasses should also provide these attributes.
372 A FileCategory object for the category of files that this format
375 EXT A file extension to be applied to encoded output files.
377 NAME A user-facing name for the format.
379 PROPS A parser element to parse a property definition. It should produce
380 a pair NAME, VALUE to be stored in a dictionary.
382 Subclasses for different kinds of file may introduce more subclass
387 """Post-encoding fixups."""
393 def defformat(name, cls):
394 "Define a format NAME using class CLS."
395 if not hasattr(cls, 'NAME'):
396 raise ValueError, 'abstract class'
397 if not hasattr(cls, 'CATEGORY'):
398 raise ValueError, 'no category'
399 FORMATMAP[name] = cls
401 class FormatParser (P.ParserElement):
403 Parse a format specifier:
405 format-spec ::= string [format-properties]
406 format-properties ::= `{' format-property (`,' format-property)* `}'
408 The syntax of a format-property is determined by the PROPS attribute on the
409 named format and its superclasses.
414 ## We cache the parser elements we generate to avoid enormous consing.
417 def parseImpl(me, s, loc, actp = True):
419 ## Firstly, determine the format name.
420 loc, r = Name._parse(s, loc, actp)
423 ## Look up the format class.
424 try: fcls = FORMATMAP[fmt]
426 raise P.ParseException(s, loc, "Unknown format `%s'" % fmt)
428 ## Fetch the property-list parser from the cache, if possible; else
437 except AttributeError: continue
438 if p in seen: continue
439 if prop is None: prop = p
443 pp = me.CACHE[fmt] = None
445 props = P.delimitedList(prop)
446 props.setParseAction(lambda s, l, t: dict(t.asList()))
447 pp = me.CACHE[fmt] = O(D('{') - props - D('}'))
449 ## Parse the properties.
453 loc, r = pp._parse(s, loc, actp)
457 ## Construct the format object and return it.
458 return loc, fcls(**pd)
460 Format = FormatParser()
462 def prop(kw, pval, tag = None):
463 if tag is None: tag = kw
466 p.setParseAction(lambda s, l, t: (tag, True))
468 p = K(kw) + D('=') + pval
469 p.setParseAction(lambda s, l, t: (tag, t[0]))
472 ###--------------------------------------------------------------------------
473 ### Policies and actions.
475 class Action (object):
477 An Action object represents a conversion action to be performed.
479 This class isn't intended to be instantiated directly. It exists to define
480 some protocol common to all Action objects.
482 Action objects have the following attributes.
484 master The name of the master (source) file.
486 target The name of the target (destination) file.
488 PRIORITY The priority of the action, for deciding which of two actions
489 to perform. Higher priorities are more likely to win.
491 Converting an Action to a string describes the action in a simple
492 user-readable manner. The `perform' method actually carries the action
498 def __init__(me, master):
499 "Stash the MASTER file name for later."
503 "Choose either ME or HIM and return one."
504 if him is None or me.PRIORITY > him.PRIORITY:
509 class CopyAction (Action):
511 An Action object for simply copying a file.
513 Actually we try to hardlink it first, falling back to a copy later. This
514 is both faster and more efficient with regard to disk space.
517 ## Copying is good. Linking is really good, but we can't tell the
518 ## difference at this stage.
521 def __init__(me, master, targetdir):
522 "Initialize a CopyAction, from MASTER to the TARGETDIR directory."
523 Action.__init__(me, master)
524 me.target = OS.path.join(targetdir, OS.path.basename(master))
530 "Actually perform a CopyAction."
532 STATUS.set(filestatus(me.master, 'link'))
533 OS.link(me.master, me.target)
535 if err.errno != E.EXDEV:
537 STATUS.set(filestatus(me.master, 'copy'))
538 new = me.target + '.new'
539 SH.copyfile(me.master, new)
540 OS.rename(new, me.target)
543 class ConvertAction (Action):
545 An Action object for converting a file to a given format.
547 Additional attributes:
549 id The identification object for the master file.
551 format The format to which we're meant to conver the master.
554 def __init__(me, master, targetdir, id, format):
555 "Initialize a ConvertAction."
556 Action.__init__(me, master)
557 stem, ext = OS.path.splitext(OS.path.basename(master))
558 me.target = OS.path.join(targetdir, stem + '.' + format.EXT)
563 return 'convert to %s' % me.format.NAME
566 "Acually perform a ConvertAction."
567 STATUS.set(filestatus(me.master, me))
568 me.format.convert(me.master, me.id, me.target)
572 class FormatPolicy (object):
574 A FormatPolicy object represents a set of rules for how to convert files.
576 Given a master file, the FormatPolicy will identify it and return a list of
577 actions to be performed. The methods required of a FormatPolicy are:
580 Store CAT as the policy's category. Check that this is consistent
581 with the policy as stored.
583 actions(MASTER, TARGETDIR, ID, COHORT)
584 Given a MASTER file, identified as ID, a target directory
585 TARGETDIR, and a list COHORT of (FILE, ID) pairs for other files
586 of the same category in the same directory, return a list of
587 actions to be performed to get the target directory into the right
588 form. The list might be empty if the policy object /rejects/ the
592 class AndPolicy (FormatPolicy):
594 A FormatPolicy which does the union of a bunch of other policies.
596 Each subsidiary policy is invoked in turn. The highest-priority action for
597 each target file is returned.
600 def __init__(me, policies):
601 me._policies = policies
603 def setcategory(me, cat):
605 for p in me._policies:
608 def actions(me, master, targetdir, id, cohort):
610 for p in me._policies:
611 for a in p.actions(master, targetdir, id, cohort):
613 tmap[a.target] = a.choose(tmap.get(a.target))
618 And = K('and') - D('{') - R(Policy) - D('}')
619 And.setParseAction(lambda s, l, t: AndPolicy(t[0]))
621 class OrPolicy (FormatPolicy):
623 A FormatPolicy which tries other policies and uses the first that accepts.
625 Each subsidiary policy is invoked in turn. If any accepts, the actions it
626 proposes are turned and no further policies are invoked. If none accepts
627 then the file is rejected.
630 def __init__(me, policies):
631 me._policies = policies
633 def setcategory(me, cat):
635 for p in me._policies:
638 def actions(me, master, targetdir, id, cohort):
639 for p in me._policies:
640 aa = p.actions(master, targetdir, id, cohort)
646 Or = K('or') - D('{') - R(Policy) - D('}')
647 Or.setParseAction(lambda s, l, t: OrPolicy(t[0]))
649 class AcceptPolicy (FormatPolicy):
651 A FormatPolicy which copies files in a particular format.
653 If all of the files in a cohort are recognized as being in a particular
654 format (including this one), then accept it with a CopyAction; otherwise
658 def __init__(me, format):
661 def setcategory(me, cat):
662 if me._format.CATEGORY is not cat:
664 "Accept format `%s' has category `%s', not `%s'" % \
665 (me._format.__class__.__name__,
666 me._format.CATEGORY.name, cat.name)
669 def actions(me, master, targetdir, id, cohort):
670 if me._format.check(id) and \
671 all(me._format.check(cid) for f, cid in cohort):
672 return [CopyAction(master, targetdir)]
676 Accept = K('accept') - Format
677 Accept.setParseAction(lambda s, l, t: AcceptPolicy(t[0]))
679 class ConvertPolicy (FormatPolicy):
681 A FormatPolicy which copies files in a particular format or converts if
684 def __init__(me, format):
687 def setcategory(me, cat):
688 if me._format.CATEGORY is not cat:
690 "Accept format `%s' has category `%s', not `%s'" % \
691 (me._format.__class__.__name__,
692 me._format.CATEGORY.name, cat.name)
695 def actions(me, master, targetdir, id, cohort):
696 if me._format.check(id):
697 return [CopyAction(master, targetdir)]
699 return [ConvertAction(master, targetdir, id, me._format)]
701 Convert = K('convert') - Format
702 Convert.setParseAction(lambda s, l, t: ConvertPolicy(t[0]))
704 Policy << (And | Or | Accept | Convert)
706 ###--------------------------------------------------------------------------
707 ### Audio handling, based on GStreamer.
709 def make_element(factory, name = None, **props):
710 "Return a new element from the FACTORY with the given NAME and PROPS."
711 elt = GS.ElementFactory.make(factory, name)
712 if elt is None: raise ValueError, 'failed to make `%s\' element' % factory
713 elt.set_properties(**props)
716 def link_elements(elts):
717 "Link the elements ELTS together, in order."
720 if e0 is not None: e0.link(e1)
723 def bin_children(bin):
724 "Iterate over the (direct) children of a BIN."
725 iter = bin.iterate_elements()
727 rc, elt = iter.next()
728 if rc == GS.IteratorResult.DONE: break
729 elif rc != GS.IteratorResult.OK:
730 raise ValueError, 'iteration failed (%s)' % rc
733 class GStreamerProgressEyecandy (ProgressEyecandy):
735 Provide amusement while GStreamer is busy doing something.
737 The GStreamerProgressEyecandy object is a context manager. Wrap it round
738 your GStreamer loop to provide progress information for an operation.
741 def __init__(me, what, elt, **kw):
743 Initialize a progress meter.
745 WHAT is a prefix string to be written before the progress eyecandy
746 itself. ELT is a GStreamer element to interrogate to find the progress
750 ProgressEyecandy.__init__(me, what, **kw)
753 "Called by GLib main event loop to update the eyecandy."
759 Update the progress meter.
761 This is called periodically by the GLib main event-processing loop.
767 "Return the current progress as a pair (CURRENT, MAX)."
769 ## Fetch the current progress information. We get the duration each
770 ## time, because (particularly with VBR-encoded MP3 inputs) the estimated
771 ## duration can change as we progress. Hopefully it settles down fairly
773 ok, t = me._elt.query_position(GS.Format.TIME)
774 if ok: ok, end = me._elt.query_duration(GS.Format.TIME)
776 else: return None, None
779 "Enter context: attach progress meter display."
781 ## If we're not showing pointless frippery, don't bother at all.
782 if not STATUS.eyecandyp:
785 ## Update regularly. The pipeline runs asynchronously.
786 me._id = G.timeout_add(100, me._update)
788 def __exit__(me, ty, val, tb):
789 "Leave context: remove display and report completion or failure."
791 ## If we're not showing pointless frippery, there's nothing to remove.
793 G.source_remove(me._id)
795 ## Report completion anyway.
801 class AudioIdentifier (object):
803 Analyses and identifies an audio file.
805 Important properties are:
807 cap A capabilities structure describing the audio file data. The most
808 interesting thing in here is probably its name, which is a MIME
809 type describing the data.
811 dcap A capabilities structure describing the decoded audio data. This
812 is of interest during conversion.
814 tags A dictionary containing metadata tags from the file. These are in
815 GStreamer's encoding-independent format.
817 bitrate An approximation to the stream's bitrate, in kilobits per second.
818 This might be slow to work out for some files so it's computed on
822 def _prepare_pipeline(me):
826 ## The basic recognition kit is based around `decodebin'. We must keep
827 ## it happy by giving it sinks for the streams it's found, which it
828 ## announces asynchronously.
829 source = make_element('filesrc', 'file', location = me._file)
830 decoder = make_element('decodebin', 'decode')
831 sink = make_element('fakesink')
832 def decoder_pad_arrived(elt, pad):
833 if pad.get_current_caps()[0].get_name().startswith('audio/'):
834 elt.link_pads(pad.get_name(), sink, 'sink')
835 decoder.connect('pad-added', decoder_pad_arrived)
836 for i in [source, decoder, sink]: pipe.add(i)
837 link_elements([source, decoder])
840 return pipe, bus, decoder, sink
842 def __init__(me, file, mime):
843 "Initialize the object suitably for identifying FILE."
846 pipe, bus, decoder, sink = me._prepare_pipeline()
848 ## Make some initial GStreamer objects. We'll want the pipeline later if
849 ## we need to analyse a poorly tagged MP3 stream, so save it away.
852 ## Arrange to collect tags from the pipeline's bus as they're reported.
855 def bus_message(bus, msg):
856 ty, s = msg.type, msg.get_structure()
857 if ty == GS.MessageType.ERROR:
858 fail[:] = (ValueError, s['debug'], None)
860 elif ty == GS.MessageType.STATE_CHANGED:
861 if s['new-state'] == GS.State.PAUSED and \
864 elif ty == GS.MessageType.TAG:
866 for i in xrange(tt.n_tags()):
867 t = tt.nth_tag_name(i)
868 if tt.get_tag_size(t) != 1: continue
869 v = tt.get_value_index(t, 0)
871 bmid = bus.connect('message', bus_message)
873 ## We want to identify the kind of stream this is. (Hmm. The MIME type
874 ## recognizer has already done this work, but GStreamer is probably more
875 ## reliable.) The `decodebin' has a `typefind' element inside which will
876 ## announce the identified media type. All we need to do is find it and
877 ## attach a signal handler. (Note that the handler might be run in the
878 ## thread context of the pipeline element, but Python's GIL will keep
879 ## things from being too awful.)
882 for e in bin_children(decoder):
883 if e.get_factory().get_name() == 'typefind':
887 assert False, 'failed to find typefind element'
889 ## Crank up most of the heavy machinery. The message handler will stop
890 ## the loop when things seem to be sufficiently well underway.
891 bus.add_signal_watch()
892 pipe.set_state(GS.State.PAUSED)
895 bus.remove_signal_watch()
897 pipe.set_state(GS.State.NULL)
898 raise fail[0], fail[1], fail[2]
900 ## Store the collected tags.
903 ## Gather the capabilities. The `typefind' element knows the input data
904 ## type. The 'decodebin' knows the raw data type.
905 me.cap = tfelt.get_static_pad('src').get_allowed_caps()[0]
906 me.mime = set([mime, me.cap.get_name()])
907 me.dcap = sink.get_static_pad('sink').get_allowed_caps()[0]
909 ## If we found a plausible bitrate then stash it. Otherwise note that we
910 ## failed. If anybody asks then we'll work it out then.
911 if 'nominal-bitrate' in tags:
912 me._bitrate = tags['nominal-bitrate']/1000
913 elif 'bitrate' in tags and tags['bitrate'] >= 80000:
914 me._bitrate = tags['bitrate']/1000
917 pipe.set_state(GS.State.NULL)
922 Return the approximate bit-rate of the input file.
924 This might take a while if we have to work it out the hard way.
927 ## If we already know the answer then just return it.
928 if me._bitrate is not None:
931 ## Make up a new pipeline and main loop.
932 pipe, bus, _, _ = me._prepare_pipeline()
935 ## Watch for bus messages. We'll stop when we reach the end of the
936 ## stream: then we'll have a clear idea of how long the track was.
938 def bus_message(bus, msg):
939 ty, s = msg.type, msg.get_structure()
940 if ty == GS.MessageType.ERROR:
941 fail[:] = (ValueError, s['debug'], None)
943 elif ty == GS.MessageType.EOS:
946 bmid = bus.connect('message', bus_message)
948 ## Get everything moving, and keep the user amused while we work.
949 bus.add_signal_watch()
950 pipe.set_state(GS.State.PLAYING)
951 with GStreamerProgressEyecandy(filestatus(me._file, 'measure bitrate'),
952 pipe, silentp = True):
954 bus.remove_signal_watch()
957 pipe.set_state(GS.State.NULL)
958 raise fail[0], fail[1], fail[2]
960 ## The bitrate computation wants the file size. Ideally we'd want the
961 ## total size of the frames' contents, but that seems hard to dredge
962 ## out. If the framing overhead is small, this should be close enough
964 bytes = OS.stat(me._file).st_size
966 ## Now we should be able to find out our position accurately and work out
967 ## a bitrate. Cache it in case anybody asks again.
968 ok, t = pipe.query_position(GS.Format.TIME)
969 assert ok, 'failed to discover bitrate'
970 me._bitrate = int(8*bytes*1e6/t)
971 pipe.set_state(GS.State.NULL)
976 class AudioFormat (BaseFormat):
978 An AudioFormat is a kind of Format specialized for audio files.
980 Format checks are done on an AudioIdentifier object.
983 PROPS = prop('bitrate', Num)
985 ## libmagic reports `application/ogg' for Ogg Vorbis files. We've switched
986 ## to GIO now, which reports either `audio/ogg' or `audio/x-vorbis+ogg'
987 ## depending on how thorough it's trying to be. Still, it doesn't do any
988 ## harm here; the main risk is picking up Ogg Theora files by accident, and
989 ## we'll probably be able to extract the audio from them anyway.
990 CATEGORY = FileCategory('audio', ['audio/*', 'application/ogg'],
993 def __init__(me, bitrate = None):
994 "Construct an object, requiring an approximate bitrate."
999 Return whether the AudioIdentifier ID is suitable for our purposes.
1001 Subclasses can either override this method or provide a property
1002 `MIMETYPES', which is a list (other thing that implements `__contains__')
1003 of GStreamer MIME types matching this format.
1005 return id.mime & me.MIMETYPES and \
1006 (me.bitrate is None or id.bitrate <= me.bitrate * sqrt(2))
1010 Constructs a GStreamer element to encode audio input.
1012 Subclasses can either override this method (or replace `encode'
1013 entirely), or provide a method `encoder_chain' which returns a list of
1014 elements to be linked together in sequence. The first element in the
1015 chain must have a pad named `sink' and the last must have a pad named
1018 elts = me.encoder_chain()
1020 for i in elts: bin.add(i)
1022 bin.add_pad(GS.GhostPad('sink', elts[0].get_static_pad('sink')))
1023 bin.add_pad(GS.GhostPad('src', elts[-1].get_static_pad('src')))
1026 def convert(me, master, id, target):
1028 Encode audio from MASTER, already identified as ID, writing it to TARGET.
1030 See `encoder' for subclasses' responsibilities.
1033 ## Construct the necessary equipment.
1034 pipe = GS.Pipeline()
1035 bus = pipe.get_bus()
1038 ## Make sure that there isn't anything in the way of our output. We're
1039 ## going to write to a scratch file so that we don't get confused by
1040 ## half-written rubbish left by a crashed program.
1041 new = target + '.new'
1044 except OSError, err:
1045 if err.errno != E.ENOENT:
1048 ## Piece together our pipeline. The annoying part is that the
1049 ## `decodebin' doesn't have any source pads yet, so our chain is in two
1051 source = make_element('filesrc', 'source', location = master)
1052 decoder = make_element('decodebin', 'decode')
1053 convert = make_element('audioconvert', 'convert')
1054 encoder = me.encoder()
1055 sink = make_element('filesink', 'sink', location = new)
1056 for i in [source, decoder, convert, encoder, sink]: pipe.add(i)
1057 link_elements([source, decoder])
1058 link_elements([convert, encoder, sink])
1060 ## Some decoders (e.g., the AC3 decoder) include channel-position
1061 ## indicators in their output caps. The Vorbis encoder interferes with
1062 ## this, and you end up with a beautifully encoded mono signal from a
1063 ## stereo source. From a quick butchers at the `vorbisenc' source, I
1064 ## /think/ that this is only a problem with stereo signals: mono signals
1065 ## are mono already, and `vorbisenc' accepts channel positions if there
1066 ## are more than two channels.
1068 ## So we have this bodge. We already collected the decoded audio caps
1069 ## during identification. So if we see 2-channel audio with channel
1070 ## positions, we strip the positions off forcibly by adding a filter.
1071 if id.dcap.get_name().startswith('audio/x-raw-') and \
1072 id.dcap.has_field('channels') and \
1073 id.dcap['channels'] == 2 and \
1074 id.dcap.has_field('channel-positions'):
1077 c.remove_field('channel-positions')
1082 ## Hook onto the `decodebin' so we can link together the two halves of
1083 ## our encoding chain. For now, we'll hope that there's only one audio
1084 ## stream in there, and just throw everything else away.
1085 def decoder_pad_arrived(elt, pad):
1086 if pad.get_current_caps()[0].get_name().startswith('audio/'):
1088 elt.link_pads_filtered(pad.get_name(), convert, 'sink', dcap)
1090 elt.link_pads(pad.get_name(), convert, 'sink')
1091 decoder.connect('pad-added', decoder_pad_arrived)
1093 ## Watch the bus for completion messages.
1095 def bus_message(bus, msg):
1096 if msg.type == GS.MessageType.ERROR:
1097 fail[:] = (ValueError, msg.get_structure()['debug'], None)
1099 elif msg.type == GS.MessageType.EOS:
1101 bmid = bus.connect('message', bus_message)
1103 ## Get everything ready and let it go.
1104 bus.add_signal_watch()
1105 pipe.set_state(GS.State.PLAYING)
1106 with GStreamerProgressEyecandy(filestatus(master,
1107 'convert to %s' % me.NAME),
1110 pipe.set_state(GS.State.NULL)
1111 bus.remove_signal_watch()
1112 bus.disconnect(bmid)
1114 raise fail[0], fail[1], fail[2]
1116 ## Fix up the output file if we have to.
1120 OS.rename(new, target)
1122 class OggVorbisFormat (AudioFormat):
1123 "AudioFormat object for Ogg Vorbis."
1125 ## From https://en.wikipedia.org/wiki/Vorbis
1126 QMAP = [(-1, 45), ( 0, 64), ( 1, 80), ( 2, 96),
1127 ( 3, 112), ( 4, 128), ( 5, 160), ( 6, 192),
1128 ( 7, 224), ( 8, 256), ( 9, 320), (10, 500)]
1131 MIMETYPES = set(['application/ogg', 'audio/x-vorbis', 'audio/ogg',
1132 'audio/x-vorbis+ogg'])
1135 def encoder_chain(me):
1137 if me.bitrate is not None:
1138 for q, br in me.QMAP:
1139 if br >= me.bitrate:
1142 raise ValueError, 'no suitable quality setting found'
1143 encprops['quality'] = q/10.0
1144 return [make_element('vorbisenc', **encprops),
1145 make_element('oggmux')]
1147 defformat('ogg-vorbis', OggVorbisFormat)
1149 class MP3Format (AudioFormat):
1150 "AudioFormat object for MP3."
1153 MIMETYPES = set(['audio/mpeg'])
1156 def encoder_chain(me):
1158 if me.bitrate is not None:
1159 encprops['bitrate'] = me.bitrate
1160 encprops['target'] = 'bitrate'
1162 encprops['quality'] = 4
1163 encprops['target'] = 'quality'
1164 return [make_element('lamemp3enc', quality = 4, **encprops),
1165 make_element('xingmux'),
1166 make_element('id3v2mux')]
1168 def fixup(me, path):
1172 GStreamer produces ID3v2 tags, but not ID3v1. This seems unnecessarily
1173 unkind to stupid players.
1176 if f is None: return
1178 if t is None: return
1179 for v in [E3.id3.ID3_V2_3, E3.id3.ID3_V1]:
1180 try: f.tag.save(version = v)
1181 except (UnicodeEncodeError,
1182 E3.id3.GenreException,
1183 E3.id3.TagException):
1186 defformat('mp3', MP3Format)
1188 ###--------------------------------------------------------------------------
1189 ### Image handling, based on the Python Imaging Library.
1191 class ImageIdentifier (object):
1193 Analyses and identifies an image file.
1195 Simply leaves an Image object in the `img' property which can be inspected.
1198 def __init__(me, file, mime):
1200 ## Get PIL to open the file. It will magically work out what kind of
1203 me.img = I.open(file)
1204 except IOError, exc:
1206 ## Unhelpful thing to raise on identification failure. We can
1207 ## distinguish this from an actual I/O error because it doesn't have an
1209 if exc.errno is None:
1210 raise IdentificationFailure
1213 me.mime = set([mime])
1215 class ImageFormat (BaseFormat):
1217 An ImageFormat is a kind of Format specialized for image files.
1219 Subclasses don't need to provide anything other than the properties
1220 required by all concrete Format subclasses. However, there is a
1221 requirement that the `NAME' property match PIL's `format' name for the
1225 PROPS = prop('size', Num)
1226 CATEGORY = FileCategory('image', ['image/*'], ImageIdentifier)
1228 def __init__(me, size = None, **kw):
1230 Initialize an ImageFormat object.
1232 Additional keywords are used when encoding, and may be recognized by
1233 enhanced `check' methods in subclasses.
1239 "Check whether the ImageIdentifier ID matches our requirements."
1240 return id.img.format == me.NAME and \
1241 (me._size is None or
1242 (id.img.size[0] <= me._size and
1243 id.img.size[1] <= me._size))
1245 def convert(me, master, id, target):
1246 "Encode the file MASTER, identified as ID, writing the result to TARGET."
1248 ## Write to a scratch file.
1249 new = target + '.new'
1251 ## The ImageIdentifier already contains a copy of the open file. It
1252 ## would be wasteful not to use it.
1254 STATUS.set(filestatus(master, 'convert to %s' % me.NAME))
1256 ## If there's a stated maximum size then scale the image down to match.
1257 ## But thumbnailing clobbers the original, so take a copy.
1258 if me._size is not None and \
1259 (img.size[0] > me._size or img.size[1] > me._size):
1261 img.thumbnail((me._size, me._size), I.ANTIALIAS)
1263 ## Write the output image.
1264 img.save(new, me.NAME, **me._props)
1266 ## Fix it up if necessary.
1270 OS.rename(new, target)
1273 class JPEGFormat (ImageFormat):
1275 Image format for JPEG (actually JFIF) files.
1277 Interesting properties to set:
1280 If present, take a second pass to select optimal encoder settings.
1283 If present, make a progressive file.
1285 quality Integer from 1--100 (worst to best); default is 75.
1289 PROPS = prop('optimize', None) \
1290 | prop('progressive', None, 'progression') \
1291 | prop('quality', Num)
1293 defformat('jpeg', JPEGFormat)
1295 class PNGFormat (ImageFormat):
1297 Image format for PNG files.
1299 Interesting properties:
1302 If present, make a special effort to minimize the output file.
1306 PROPS = prop('optimize', None)
1308 defformat('png', PNGFormat)
1310 class BMPFormat (ImageFormat):
1312 Image format for Windows BMP files, as used by RockBox.
1314 No additional properties.
1319 defformat('bmp', BMPFormat)
1321 ###--------------------------------------------------------------------------
1322 ### Remaining parsing machinery.
1324 Type = K('type') - Name - D('{') - R(Policy) - D('}')
1325 def build_type(s, l, t):
1327 cat = CATEGORYMAP[t[0]]
1329 raise P.ParseException(s, loc, "Unknown category `%s'" % t[0])
1331 if len(pols) == 1: pol = pols[0]
1332 else: pol = AndPolicy(pols)
1333 pol.setcategory(cat)
1335 Type.setParseAction(build_type)
1338 class TargetJob (object):
1339 def __init__(me, targetdir, policies):
1340 me.targetdir = targetdir
1341 me.policies = policies
1345 Target = K('target') - String - D('{') - R(Type) - D('}')
1346 def build_target(s, l, t):
1347 return TargetJob(t[0], t[1])
1348 Target.setParseAction(build_target)
1350 VARS = { 'master': None }
1351 class VarsJob (object):
1352 def __init__(me, vars):
1355 for k, v in me.vars:
1358 Var = prop('master', String)
1359 Vars = K('vars') - D('{') - R(Var) - D('}')
1360 def build_vars(s, l, t):
1361 return VarsJob(t[0])
1362 Vars.setParseAction(build_vars)
1364 TopLevel = Vars | Target
1365 Config = R(TopLevel)
1366 Config.ignore(P.pythonStyleComment)
1368 ###--------------------------------------------------------------------------
1369 ### The directory grobbler.
1371 def grobble(master, targets, noact = False):
1373 Work through the MASTER directory, writing converted files to TARGETS.
1375 The TARGETS are a list of `TargetJob' objects, each describing a target
1376 directory and a policy to apply to it.
1378 If NOACT is true, then don't actually do anything permanent to the
1382 ## Transform the targets into a more convenient data structure.
1386 tpolmap.append(pmap)
1387 for p in t.policies: pmap.setdefault(p.cat, []).append(p)
1389 ## Keep track of the current position in the master tree.
1392 ## And the files which haven't worked.
1395 def grobble_file(master, pmap, targetdir, cohorts):
1396 ## Convert MASTER, writing the result to TARGETDIR.
1398 ## The COHORTS are actually (CAT, ID, COHORT) triples, where a COHORT is
1399 ## a list of (FILENAME, ID) pairs.
1401 ## Since this function might convert the MASTER file, the caller doesn't
1402 ## know the name of the output files, so we return then as a list.
1405 st_m = OS.stat(master)
1407 ## Work through each category listed and apply its policy.
1408 for cat, id, cohort in cohorts:
1410 ## Go through the category's policies and see if any match. If we fail
1411 ## here, see if there are more categories to try.
1412 for pol in pmap[cat]:
1413 acts = pol.actions(master, targetdir, id, cohort)
1418 ## Work through the targets one by one.
1422 ## Find out whether the target file already exists and is up-to-date
1423 ## with respect to the master. (Caution here with low-resolution
1424 ## timestamps.) If it's OK, then just move on.
1426 st_t = OS.stat(a.target)
1427 if st_m.st_mtime < st_t.st_mtime or \
1428 (st_m.st_ino, st_m.st_dev) == (st_t.st_ino, st_t.st_dev):
1430 except OSError, err:
1431 if err.errno not in (E.ENOENT, E.ENOTDIR):
1434 ## We have real work to do. If there's a current status message,
1435 ## it's the containing directory so flush it so that people know
1439 ## Remove the target. (A hardlink will fail if the target already
1444 except OSError, err:
1445 if err.errno not in (E.ENOENT, E.ENOTDIR):
1448 ## Do whatever it is we decided to do.
1450 STATUS.commit(filestatus(master, a))
1454 ## We're done. Return the names of the targets.
1458 def wrap(masterfile):
1459 ## Handle exceptions found while trying to convert a particular file or
1465 ## Something bad happened. Report the error, but continue. (This list
1466 ## of exceptions needs a lot of work.)
1467 except (IOError, OSError), exc:
1469 STATUS.commit(filestatus(masterfile, 'failed (%s)' % exc))
1470 broken.append((masterfile, exc))
1472 def grobble_dir(master, targets):
1473 ## Recursively convert files in MASTER, writing them to the TARGETS.
1475 ## Keep track of the subdirectories we encounter, because we'll need to
1476 ## do all of those in one go at the end.
1479 ## Work through each target directory in turn.
1480 for target, pmap in zip(targets, tpolmap):
1482 ## Make sure the TARGET exists and is a directory. It's a fundamental
1483 ## assumption of this program that the entire TARGET tree is
1484 ## disposable, so if something exists but isn't a directory, we should
1486 if OS.path.isdir(target):
1489 if OS.path.exists(target):
1490 STATUS.commit(filestatus(target, 'clear nondirectory'))
1493 STATUS.commit(filestatus(target, 'create directory'))
1497 ## Keep a list of things in the target. As we convert files, we'll
1498 ## check them off. Anything left over is rubbish and needs to be
1502 for i in OS.listdir(target):
1503 checklist[i] = False
1504 except OSError, err:
1505 if err.errno not in (E.ENOENT, E.ENOTDIR):
1508 ## Keep track of the files in each category.
1513 ## Work through the master files.
1514 for f in sorted(OS.listdir(master)):
1516 ## If the killswitch has been pulled then stop. The whole idea is
1517 ## that we want to cause a clean shutdown if possible, so we don't
1518 ## want to do it in the middle of encoding because the encoding
1519 ## effort will have been wasted. This is the only place we need to
1520 ## check. If we've exited the loop, then clearing old files will
1521 ## probably be fast, and we'll either end up here when the recursive
1522 ## call returns or we'll be in the same boat as before, clearing old
1523 ## files, only up a level. If worst comes to worst, we'll be killed
1524 ## forcibly somewhere inside `SH.rmtree', and that can continue where
1526 if KILLSWITCH.is_set():
1529 ## Do something with the file.
1530 with wrap(OS.path.join(master, f)) as masterfile:
1532 ## If it's a directory then prepare to grobble it recursively, but
1533 ## don't do that yet.
1534 if OS.path.isdir(masterfile):
1536 done.append(OS.path.join(target, f))
1538 ## Otherwise it's a file. Work out what kind, and stash it under
1539 ## the appropriate categories. Later, we'll apply policy to the
1540 ## files, by category, and work out what to do with them all.
1542 mime = GIO.file_new_for_path(masterfile) \
1543 .query_info('standard::content-type', 0) \
1546 for cat in pmap.iterkeys():
1547 id = cat.identify(masterfile, mime)
1548 if id is None: continue
1549 catmap.setdefault(cat, []).append((masterfile, id))
1550 cats.append((cat, id))
1552 catmap.setdefault(None, []).append((masterfile, id))
1553 todo.append((masterfile, cats))
1555 ## Work through the categorized files to see what actions to do for
1557 for masterfile, cats in todo:
1558 with wrap(masterfile):
1559 done += grobble_file(masterfile, pmap, target,
1560 [(cat, id, catmap[cat]) for cat, id in cats])
1562 ## Check the results off the list so that we don't clear it later.
1564 checklist[OS.path.basename(f)] = True
1566 ## Maybe there's stuff in the target which isn't accounted for. Delete
1567 ## it: either the master has changed, or the policy for this target has
1568 ## changed. Either way, the old files aren't wanted.
1570 if not checklist[f]:
1571 STATUS.commit(filestatus(f, 'clear bogus file'))
1573 bogus = OS.path.join(target, f)
1575 if OS.path.isdir(bogus):
1579 except OSError, err:
1580 if err.errno != E.ENOENT:
1583 ## If there are subdirectories which want processing then do those.
1584 ## Keep the user amused by telling him where we are in the tree.
1585 for d in sorted(subdirs):
1587 STATUS.set('/'.join(dirs))
1588 with wrap(OS.path.join(master, d)) as masterdir:
1590 grobble_dir(masterdir,
1591 [OS.path.join(target, d) for target in targets])
1594 STATUS.set('/'.join(dirs))
1596 ## Right. We're ready to go.
1597 grobble_dir(master, [t.targetdir for t in targets])
1600 ###--------------------------------------------------------------------------
1601 ### Command-line interface.
1603 QUIS = OS.path.basename(SYS.argv[0])
1606 "Report a warning message to the user."
1607 SYS.stderr.write('%s: %s\n' % (QUIS, msg))
1610 "Report a fatal error message to the user."
1614 def parse_opts(args):
1616 Parse command-line arguments in ARGS.
1618 Returns a Grobbler object and the MASTER and TARGET directories to be
1622 ## Build the option parser object.
1623 op = OP.OptionParser(prog = QUIS, version = VERSION,
1624 usage = '%prog [-in] [-t TIMEOUT] [-T TIMEOUT] '
1627 Convert a directory tree of files according to the configuration file
1631 ## Timeout handling.
1632 def cb_time(opt, ostr, arg, op):
1633 m = RX.match(r'\s*(\d+)\s*([dhms]?)\s*', arg)
1635 raise OP.OptionValueerror, 'bad time value `%s\'' % arg
1637 t = int(t) * { '': 1, 's': 1, 'm': 60, 'h': 3600, 'd': 86400 }[u]
1638 setattr(op.values, opt.dest, t)
1639 op.add_option('-t', '--timeout', type = 'string', metavar = 'SECS',
1641 help = 'stop processing nicely after SECS',
1642 action = 'callback', callback = cb_time)
1643 op.add_option('-T', '--timeout-nasty', type = 'string', metavar = 'SECS',
1644 dest = 'timeout_nasty',
1645 help = 'stop processing unpleasantly after further SECS',
1646 action = 'callback', callback = cb_time)
1649 op.add_option('-i', '--interactive', action = 'store_true', dest = 'tty',
1650 help = 'provide progress information')
1651 op.add_option('-n', '--no-act', action = 'store_true', dest = 'noact',
1652 help = 'don\'t actually modify the filesystem')
1655 op.set_defaults(formats = [], noact = False,
1656 timeout = None, timeout_nasty = 300)
1657 opts, args = op.parse_args(args)
1659 ## Check that we got the non-option arguments that we want.
1661 op.error('wrong number of arguments')
1663 ## Act on the options.
1665 STATUS.eyecandyp = True
1666 if opts.timeout is not None:
1667 to = TH.Thread(target = timeout,
1668 args = (opts.timeout, opts.timeout_nasty))
1672 ## Parse the configuration file.
1673 with open(args[0]) as conf:
1674 jobs, = Config.parseFile(conf, True)
1680 if __name__ == '__main__':
1681 opts = parse_opts(SYS.argv[1:])
1682 if 'master' not in VARS:
1683 die("no master directory set")
1684 broken = grobble(VARS['master'], TARGETS, opts.noact)
1686 moan('failed to convert some files:')
1687 for file, exc in broken:
1688 moan('%s: %s' % (file, exc))
1691 ## This is basically a successful completion: we did what we were asked to
1692 ## do. It seems polite to report a message, though.
1694 ## Why don't we have a nonzero exit status? The idea would be that a
1695 ## calling script would be interested that we used up all of our time, and
1696 ## not attempt to convert some other directory as well. But that doesn't
1697 ## quite work. Such a script would need to account correctly for time we
1698 ## had spent even if we complete successfully. And if the script is having
1699 ## to watch the clock itself, it can do that without our help here.
1700 if KILLSWITCH.is_set():
1701 moan('killed by timeout')
1703 ###----- That's all, folks --------------------------------------------------