Browse Source

added subset.py script from Just van Rossum's FontTools project

Chris Simpkins 9 years ago
parent
commit
8812aba66c
1 changed files with 2748 additions and 0 deletions
  1. 2748 0
      tools/subset.py

+ 2748 - 0
tools/subset.py

@@ -0,0 +1,2748 @@
+# Copyright 2013 Google, Inc. All Rights Reserved.
+#
+# Google Author(s): Behdad Esfahbod
+
+# subset.py file from Just van Rossom's fonttools project
+#
+# Copyright 1999-2004
+# by Just van Rossum, Letterror, The Netherlands.
+# The full text of the license is available at https://github.com/behdad/fonttools/blob/master/LICENSE.txt
+
+from __future__ import print_function, division, absolute_import
+from fontTools.misc.py23 import *
+from fontTools import ttLib
+from fontTools.ttLib.tables import otTables
+from fontTools.misc import psCharStrings
+import sys
+import struct
+import time
+import array
+
+__usage__ = "pyftsubset font-file [glyph...] [--option=value]..."
+
+__doc__="""\
+pyftsubset -- OpenType font subsetter and optimizer
+
+  pyftsubset is an OpenType font subsetter and optimizer, based on fontTools.
+  It accepts any TT- or CFF-flavored OpenType (.otf or .ttf) or WOFF (.woff)
+  font file. The subsetted glyph set is based on the specified glyphs
+  or characters, and specified OpenType layout features.
+
+  The tool also performs some size-reducing optimizations, aimed for using
+  subset fonts as webfonts.  Individual optimizations can be enabled or
+  disabled, and are enabled by default when they are safe.
+
+Usage:
+  """+__usage__+"""
+
+  At least one glyph or one of --gids, --gids-file, --glyphs, --glyphs-file,
+  --text, --text-file, --unicodes, or --unicodes-file, must be specified.
+
+Arguments:
+  font-file
+    The input font file.
+  glyph
+    Specify one or more glyph identifiers to include in the subset. Must be
+    PS glyph names, or the special string '*' to keep the entire glyph set.
+
+Initial glyph set specification:
+  These options populate the initial glyph set. Same option can appear
+  multiple times, and the results are accummulated.
+  --gids=<NNN>[,<NNN>...]
+      Specify comma/whitespace-separated list of glyph IDs or ranges as
+      decimal numbers.  For example, --gids=10-12,14 adds glyphs with
+      numbers 10, 11, 12, and 14.
+  --gids-file=<path>
+      Like --gids but reads from a file. Anything after a '#' on any line
+      is ignored as comments.
+  --glyphs=<glyphname>[,<glyphname>...]
+      Specify comma/whitespace-separated PS glyph names to add to the subset.
+      Note that only PS glyph names are accepted, not gidNNN, U+XXXX, etc
+      that are accepted on the command line.  The special string '*' wil keep
+      the entire glyph set.
+  --glyphs-file=<path>
+      Like --glyphs but reads from a file. Anything after a '#' on any line
+      is ignored as comments.
+  --text=<text>
+      Specify characters to include in the subset, as UTF-8 string.
+  --text-file=<path>
+      Like --text but reads from a file. Newline character are not added to
+      the subset.
+  --unicodes=<XXXX>[,<XXXX>...]
+      Specify comma/whitespace-separated list of Unicode codepoints or
+      ranges as hex numbers, optionally prefixed with 'U+', 'u', etc.
+      For example, --unicodes=41-5a,61-7a adds ASCII letters, so does
+      the more verbose --unicodes=U+0041-005A,U+0061-007A.
+      The special strings '*' will choose all Unicode characters mapped
+      by the font.
+  --unicodes-file=<path>
+      Like --unicodes, but reads from a file. Anything after a '#' on any
+      line in the file is ignored as comments.
+  --ignore-missing-glyphs
+      Do not fail if some requested glyphs or gids are not available in
+      the font.
+  --no-ignore-missing-glyphs
+      Stop and fail if some requested glyphs or gids are not available
+      in the font. [default]
+  --ignore-missing-unicodes [default]
+      Do not fail if some requested Unicode characters (including those
+      indirectly specified using --text or --text-file) are not available
+      in the font.
+  --no-ignore-missing-unicodes
+      Stop and fail if some requested Unicode characters are not available
+      in the font.
+      Note the default discrepancy between ignoring missing glyphs versus
+      unicodes.  This is for historical reasons and in the future
+      --no-ignore-missing-unicodes might become default.
+
+Other options:
+  For the other options listed below, to see the current value of the option,
+  pass a value of '?' to it, with or without a '='.
+  Examples:
+    $ pyftsubset --glyph-names?
+    Current setting for 'glyph-names' is: False
+    $ ./pyftsubset --name-IDs=?
+    Current setting for 'name-IDs' is: [1, 2]
+    $ ./pyftsubset --hinting? --no-hinting --hinting?
+    Current setting for 'hinting' is: True
+    Current setting for 'hinting' is: False
+
+Output options:
+  --output-file=<path>
+      The output font file. If not specified, the subsetted font
+      will be saved in as font-file.subset.
+  --flavor=<type>
+      Specify flavor of output font file. May be 'woff' or 'woff2'.
+      Note that WOFF2 requires the Brotli Python extension, available
+      at https://github.com/google/brotli
+
+Glyph set expansion:
+  These options control how additional glyphs are added to the subset.
+  --notdef-glyph
+      Add the '.notdef' glyph to the subset (ie, keep it). [default]
+  --no-notdef-glyph
+      Drop the '.notdef' glyph unless specified in the glyph set. This
+      saves a few bytes, but is not possible for Postscript-flavored
+      fonts, as those require '.notdef'. For TrueType-flavored fonts,
+      this works fine as long as no unsupported glyphs are requested
+      from the font.
+  --notdef-outline
+      Keep the outline of '.notdef' glyph. The '.notdef' glyph outline is
+      used when glyphs not supported by the font are to be shown. It is not
+      needed otherwise.
+  --no-notdef-outline
+      When including a '.notdef' glyph, remove its outline. This saves
+      a few bytes. [default]
+  --recommended-glyphs
+      Add glyphs 0, 1, 2, and 3 to the subset, as recommended for
+      TrueType-flavored fonts: '.notdef', 'NULL' or '.null', 'CR', 'space'.
+      Some legacy software might require this, but no modern system does.
+  --no-recommended-glyphs
+      Do not add glyphs 0, 1, 2, and 3 to the subset, unless specified in
+      glyph set. [default]
+  --layout-features[+|-]=<feature>[,<feature>...]
+      Specify (=), add to (+=) or exclude from (-=) the comma-separated
+      set of OpenType layout feature tags that will be preserved.
+      Glyph variants used by the preserved features are added to the
+      specified subset glyph set. By default, 'calt', 'ccmp', 'clig', 'curs',
+      'kern', 'liga', 'locl', 'mark', 'mkmk', 'rclt', 'rlig' and all features
+      required for script shaping are preserved. To see the full list, try
+      '--layout-features=?'. Use '*' to keep all features.
+      Multiple --layout-features options can be provided if necessary.
+      Examples:
+        --layout-features+=onum,pnum,ss01
+            * Keep the default set of features and 'onum', 'pnum', 'ss01'.
+        --layout-features-='mark','mkmk'
+            * Keep the default set of features but drop 'mark' and 'mkmk'.
+        --layout-features='kern'
+            * Only keep the 'kern' feature, drop all others.
+        --layout-features=''
+            * Drop all features.
+        --layout-features='*'
+            * Keep all features.
+        --layout-features+=aalt --layout-features-=vrt2
+            * Keep default set of features plus 'aalt', but drop 'vrt2'.
+
+Hinting options:
+  --hinting
+      Keep hinting [default]
+  --no-hinting
+      Drop glyph-specific hinting and font-wide hinting tables, as well
+      as remove hinting-related bits and pieces from other tables (eg. GPOS).
+      See --hinting-tables for list of tables that are dropped by default.
+      Instructions and hints are stripped from 'glyf' and 'CFF ' tables
+      respectively. This produces (sometimes up to 30%) smaller fonts that
+      are suitable for extremely high-resolution systems, like high-end
+      mobile devices and retina displays.
+      XXX Note: Currently there is a known bug in 'CFF ' hint stripping that
+      might make the font unusable as a webfont as they will be rejected by
+      OpenType Sanitizer used in common browsers. For more information see:
+      https://github.com/behdad/fonttools/issues/144
+      The --desubroutinize options works around that bug.
+
+Optimization options:
+  --desubroutinize
+      Remove CFF use of subroutinizes.  Subroutinization is a way to make CFF
+      fonts smaller.  For small subsets however, desubroutinizing might make
+      the font smaller.  It has even been reported that desubroutinized CFF
+      fonts compress better (produce smaller output) WOFF and WOFF2 fonts.
+      Also see note under --no-hinting.
+  --no-desubroutinize [default]
+      Leave CFF subroutinizes as is, only throw away unused subroutinizes.
+
+Font table options:
+  --drop-tables[+|-]=<table>[,<table>...]
+      Specify (=), add to (+=) or exclude from (-=) the comma-separated
+      set of tables that will be be dropped.
+      By default, the following tables are dropped:
+      'BASE', 'JSTF', 'DSIG', 'EBDT', 'EBLC', 'EBSC', 'SVG ', 'PCLT', 'LTSH'
+      and Graphite tables: 'Feat', 'Glat', 'Gloc', 'Silf', 'Sill'
+      and color tables: 'CBLC', 'CBDT', 'sbix', 'COLR', 'CPAL'.
+      The tool will attempt to subset the remaining tables.
+      Examples:
+        --drop-tables-='SVG '
+            * Drop the default set of tables but keep 'SVG '.
+        --drop-tables+=GSUB
+            * Drop the default set of tables and 'GSUB'.
+        --drop-tables=DSIG
+            * Only drop the 'DSIG' table, keep all others.
+        --drop-tables=
+            * Keep all tables.
+  --no-subset-tables+=<table>[,<table>...]
+      Add to the set of tables that will not be subsetted.
+      By default, the following tables are included in this list, as
+      they do not need subsetting (ignore the fact that 'loca' is listed
+      here): 'gasp', 'head', 'hhea', 'maxp', 'vhea', 'OS/2', 'loca',
+      'name', 'cvt ', 'fpgm', 'prep', 'VMDX', and 'DSIG'. Tables that the tool
+      does not know how to subset and are not specified here will be dropped
+      from the font.
+      Example:
+         --no-subset-tables+=FFTM
+            * Keep 'FFTM' table in the font by preventing subsetting.
+  --hinting-tables[-]=<table>[,<table>...]
+      Specify (=), add to (+=) or exclude from (-=) the list of font-wide
+      hinting tables that will be dropped if --no-hinting is specified,
+      Examples:
+        --hinting-tables-='VDMX'
+            * Drop font-wide hinting tables except 'VDMX'.
+        --hinting-tables=''
+            * Keep all font-wide hinting tables (but strip hints from glyphs).
+  --legacy-kern
+      Keep TrueType 'kern' table even when OpenType 'GPOS' is available.
+  --no-legacy-kern
+      Drop TrueType 'kern' table if OpenType 'GPOS' is available. [default]
+
+Font naming options:
+  These options control what is retained in the 'name' table. For numerical
+  codes, see: http://www.microsoft.com/typography/otspec/name.htm
+  --name-IDs[+|-]=<nameID>[,<nameID>...]
+      Specify (=), add to (+=) or exclude from (-=) the set of 'name' table
+      entry nameIDs that will be preserved. By default only nameID 1 (Family)
+      and nameID 2 (Style) are preserved. Use '*' to keep all entries.
+      Examples:
+        --name-IDs+=0,4,6
+            * Also keep Copyright, Full name and PostScript name entry.
+        --name-IDs=''
+            * Drop all 'name' table entries.
+        --name-IDs='*'
+            * keep all 'name' table entries
+  --name-legacy
+      Keep legacy (non-Unicode) 'name' table entries (0.x, 1.x etc.).
+      XXX Note: This might be needed for some fonts that have no Unicode name
+      entires for English. See: https://github.com/behdad/fonttools/issues/146
+  --no-name-legacy
+      Drop legacy (non-Unicode) 'name' table entries [default]
+  --name-languages[+|-]=<langID>[,<langID>]
+      Specify (=), add to (+=) or exclude from (-=) the set of 'name' table
+      langIDs that will be preserved. By default only records with langID
+      0x0409 (English) are preserved. Use '*' to keep all langIDs.
+  --obfuscate-names
+      Make the font unusable as a system font by replacing name IDs 1, 2, 3, 4,
+      and 6 with dummy strings (it is still fully functional as webfont).
+
+Glyph naming and encoding options:
+  --glyph-names
+      Keep PS glyph names in TT-flavored fonts. In general glyph names are
+      not needed for correct use of the font. However, some PDF generators
+      and PDF viewers might rely on glyph names to extract Unicode text
+      from PDF documents.
+  --no-glyph-names
+      Drop PS glyph names in TT-flavored fonts, by using 'post' table
+      version 3.0. [default]
+  --legacy-cmap
+      Keep the legacy 'cmap' subtables (0.x, 1.x, 4.x etc.).
+  --no-legacy-cmap
+      Drop the legacy 'cmap' subtables. [default]
+  --symbol-cmap
+      Keep the 3.0 symbol 'cmap'.
+  --no-symbol-cmap
+      Drop the 3.0 symbol 'cmap'. [default]
+
+Other font-specific options:
+  --recalc-bounds
+      Recalculate font bounding boxes.
+  --no-recalc-bounds
+      Keep original font bounding boxes. This is faster and still safe
+      for all practical purposes. [default]
+  --recalc-timestamp
+      Set font 'modified' timestamp to current time.
+  --no-recalc-timestamp
+      Do not modify font 'modified' timestamp. [default]
+  --canonical-order
+      Order tables as recommended in the OpenType standard. This is not
+      required by the standard, nor by any known implementation.
+  --no-canonical-order
+      Keep original order of font tables. This is faster. [default]
+
+Application options:
+  --verbose
+      Display verbose information of the subsetting process.
+  --timing
+      Display detailed timing information of the subsetting process.
+  --xml
+      Display the TTX XML representation of subsetted font.
+
+Example:
+  Produce a subset containing the characters ' !"#$%' without performing
+  size-reducing optimizations:
+
+  $ pyftsubset font.ttf --unicodes="U+0020-0025" \\
+    --layout-features='*' --glyph-names --symbol-cmap --legacy-cmap \\
+    --notdef-glyph --notdef-outline --recommended-glyphs \\
+    --name-IDs='*' --name-legacy --name-languages='*'
+"""
+
+
+def _add_method(*clazzes):
+    """Returns a decorator function that adds a new method to one or
+    more classes."""
+    def wrapper(method):
+        for clazz in clazzes:
+            assert clazz.__name__ != 'DefaultTable', \
+                    'Oops, table class not found.'
+            assert not hasattr(clazz, method.__name__), \
+                    "Oops, class '%s' has method '%s'." % (clazz.__name__,
+                                                           method.__name__)
+            setattr(clazz, method.__name__, method)
+        return None
+    return wrapper
+
+def _uniq_sort(l):
+    return sorted(set(l))
+
+def _set_update(s, *others):
+    # Jython's set.update only takes one other argument.
+    # Emulate real set.update...
+    for other in others:
+        s.update(other)
+
+def _dict_subset(d, glyphs):
+	return {g:d[g] for g in glyphs}
+
+
+@_add_method(otTables.Coverage)
+def intersect(self, glyphs):
+    """Returns ascending list of matching coverage values."""
+    return [i for i,g in enumerate(self.glyphs) if g in glyphs]
+
+@_add_method(otTables.Coverage)
+def intersect_glyphs(self, glyphs):
+    """Returns set of intersecting glyphs."""
+    return set(g for g in self.glyphs if g in glyphs)
+
+@_add_method(otTables.Coverage)
+def subset(self, glyphs):
+    """Returns ascending list of remaining coverage values."""
+    indices = self.intersect(glyphs)
+    self.glyphs = [g for g in self.glyphs if g in glyphs]
+    return indices
+
+@_add_method(otTables.Coverage)
+def remap(self, coverage_map):
+    """Remaps coverage."""
+    self.glyphs = [self.glyphs[i] for i in coverage_map]
+
+@_add_method(otTables.ClassDef)
+def intersect(self, glyphs):
+    """Returns ascending list of matching class values."""
+    return _uniq_sort(
+         ([0] if any(g not in self.classDefs for g in glyphs) else []) +
+            [v for g,v in self.classDefs.items() if g in glyphs])
+
+@_add_method(otTables.ClassDef)
+def intersect_class(self, glyphs, klass):
+    """Returns set of glyphs matching class."""
+    if klass == 0:
+        return set(g for g in glyphs if g not in self.classDefs)
+    return set(g for g,v in self.classDefs.items()
+                            if v == klass and g in glyphs)
+
+@_add_method(otTables.ClassDef)
+def subset(self, glyphs, remap=False):
+    """Returns ascending list of remaining classes."""
+    self.classDefs = {g:v for g,v in self.classDefs.items() if g in glyphs}
+    # Note: while class 0 has the special meaning of "not matched",
+    # if no glyph will ever /not match/, we can optimize class 0 out too.
+    indices = _uniq_sort(
+         ([0] if any(g not in self.classDefs for g in glyphs) else []) +
+            list(self.classDefs.values()))
+    if remap:
+        self.remap(indices)
+    return indices
+
+@_add_method(otTables.ClassDef)
+def remap(self, class_map):
+    """Remaps classes."""
+    self.classDefs = {g:class_map.index(v) for g,v in self.classDefs.items()}
+
+@_add_method(otTables.SingleSubst)
+def closure_glyphs(self, s, cur_glyphs):
+    s.glyphs.update(v for g,v in self.mapping.items() if g in cur_glyphs)
+
+@_add_method(otTables.SingleSubst)
+def subset_glyphs(self, s):
+    self.mapping = {g:v for g,v in self.mapping.items()
+                    if g in s.glyphs and v in s.glyphs}
+    return bool(self.mapping)
+
+@_add_method(otTables.MultipleSubst)
+def closure_glyphs(self, s, cur_glyphs):
+    indices = self.Coverage.intersect(cur_glyphs)
+    _set_update(s.glyphs, *(self.Sequence[i].Substitute for i in indices))
+
+@_add_method(otTables.MultipleSubst)
+def subset_glyphs(self, s):
+    indices = self.Coverage.subset(s.glyphs)
+    self.Sequence = [self.Sequence[i] for i in indices]
+    # Now drop rules generating glyphs we don't want
+    indices = [i for i,seq in enumerate(self.Sequence)
+               if all(sub in s.glyphs for sub in seq.Substitute)]
+    self.Sequence = [self.Sequence[i] for i in indices]
+    self.Coverage.remap(indices)
+    self.SequenceCount = len(self.Sequence)
+    return bool(self.SequenceCount)
+
+@_add_method(otTables.AlternateSubst)
+def closure_glyphs(self, s, cur_glyphs):
+    _set_update(s.glyphs, *(vlist for g,vlist in self.alternates.items()
+                            if g in cur_glyphs))
+
+@_add_method(otTables.AlternateSubst)
+def subset_glyphs(self, s):
+    self.alternates = {g:vlist
+                       for g,vlist in self.alternates.items()
+                       if g in s.glyphs and
+                       all(v in s.glyphs for v in vlist)}
+    return bool(self.alternates)
+
+@_add_method(otTables.LigatureSubst)
+def closure_glyphs(self, s, cur_glyphs):
+    _set_update(s.glyphs, *([seq.LigGlyph for seq in seqs
+                             if all(c in s.glyphs for c in seq.Component)]
+                            for g,seqs in self.ligatures.items()
+                            if g in cur_glyphs))
+
+@_add_method(otTables.LigatureSubst)
+def subset_glyphs(self, s):
+    self.ligatures = {g:v for g,v in self.ligatures.items()
+                      if g in s.glyphs}
+    self.ligatures = {g:[seq for seq in seqs
+                             if seq.LigGlyph in s.glyphs and
+                                all(c in s.glyphs for c in seq.Component)]
+                      for g,seqs in self.ligatures.items()}
+    self.ligatures = {g:v for g,v in self.ligatures.items() if v}
+    return bool(self.ligatures)
+
+@_add_method(otTables.ReverseChainSingleSubst)
+def closure_glyphs(self, s, cur_glyphs):
+    if self.Format == 1:
+        indices = self.Coverage.intersect(cur_glyphs)
+        if(not indices or
+           not all(c.intersect(s.glyphs)
+                   for c in self.LookAheadCoverage + self.BacktrackCoverage)):
+            return
+        s.glyphs.update(self.Substitute[i] for i in indices)
+    else:
+        assert 0, "unknown format: %s" % self.Format
+
+@_add_method(otTables.ReverseChainSingleSubst)
+def subset_glyphs(self, s):
+    if self.Format == 1:
+        indices = self.Coverage.subset(s.glyphs)
+        self.Substitute = [self.Substitute[i] for i in indices]
+        # Now drop rules generating glyphs we don't want
+        indices = [i for i,sub in enumerate(self.Substitute)
+                 if sub in s.glyphs]
+        self.Substitute = [self.Substitute[i] for i in indices]
+        self.Coverage.remap(indices)
+        self.GlyphCount = len(self.Substitute)
+        return bool(self.GlyphCount and
+                    all(c.subset(s.glyphs)
+                        for c in self.LookAheadCoverage+self.BacktrackCoverage))
+    else:
+        assert 0, "unknown format: %s" % self.Format
+
+@_add_method(otTables.SinglePos)
+def subset_glyphs(self, s):
+    if self.Format == 1:
+        return len(self.Coverage.subset(s.glyphs))
+    elif self.Format == 2:
+        indices = self.Coverage.subset(s.glyphs)
+        self.Value = [self.Value[i] for i in indices]
+        self.ValueCount = len(self.Value)
+        return bool(self.ValueCount)
+    else:
+        assert 0, "unknown format: %s" % self.Format
+
+@_add_method(otTables.SinglePos)
+def prune_post_subset(self, options):
+    if not options.hinting:
+        # Drop device tables
+        self.ValueFormat &= ~0x00F0
+    return True
+
+@_add_method(otTables.PairPos)
+def subset_glyphs(self, s):
+    if self.Format == 1:
+        indices = self.Coverage.subset(s.glyphs)
+        self.PairSet = [self.PairSet[i] for i in indices]
+        for p in self.PairSet:
+            p.PairValueRecord = [r for r in p.PairValueRecord
+                                 if r.SecondGlyph in s.glyphs]
+            p.PairValueCount = len(p.PairValueRecord)
+        # Remove empty pairsets
+        indices = [i for i,p in enumerate(self.PairSet) if p.PairValueCount]
+        self.Coverage.remap(indices)
+        self.PairSet = [self.PairSet[i] for i in indices]
+        self.PairSetCount = len(self.PairSet)
+        return bool(self.PairSetCount)
+    elif self.Format == 2:
+        class1_map = self.ClassDef1.subset(s.glyphs, remap=True)
+        class2_map = self.ClassDef2.subset(s.glyphs, remap=True)
+        self.Class1Record = [self.Class1Record[i] for i in class1_map]
+        for c in self.Class1Record:
+            c.Class2Record = [c.Class2Record[i] for i in class2_map]
+        self.Class1Count = len(class1_map)
+        self.Class2Count = len(class2_map)
+        return bool(self.Class1Count and
+                    self.Class2Count and
+                    self.Coverage.subset(s.glyphs))
+    else:
+        assert 0, "unknown format: %s" % self.Format
+
+@_add_method(otTables.PairPos)
+def prune_post_subset(self, options):
+    if not options.hinting:
+        # Drop device tables
+        self.ValueFormat1 &= ~0x00F0
+        self.ValueFormat2 &= ~0x00F0
+    return True
+
+@_add_method(otTables.CursivePos)
+def subset_glyphs(self, s):
+    if self.Format == 1:
+        indices = self.Coverage.subset(s.glyphs)
+        self.EntryExitRecord = [self.EntryExitRecord[i] for i in indices]
+        self.EntryExitCount = len(self.EntryExitRecord)
+        return bool(self.EntryExitCount)
+    else:
+        assert 0, "unknown format: %s" % self.Format
+
+@_add_method(otTables.Anchor)
+def prune_hints(self):
+    # Drop device tables / contour anchor point
+    self.ensureDecompiled()
+    self.Format = 1
+
+@_add_method(otTables.CursivePos)
+def prune_post_subset(self, options):
+    if not options.hinting:
+        for rec in self.EntryExitRecord:
+            if rec.EntryAnchor: rec.EntryAnchor.prune_hints()
+            if rec.ExitAnchor: rec.ExitAnchor.prune_hints()
+    return True
+
+@_add_method(otTables.MarkBasePos)
+def subset_glyphs(self, s):
+    if self.Format == 1:
+        mark_indices = self.MarkCoverage.subset(s.glyphs)
+        self.MarkArray.MarkRecord = [self.MarkArray.MarkRecord[i]
+                                     for i in mark_indices]
+        self.MarkArray.MarkCount = len(self.MarkArray.MarkRecord)
+        base_indices = self.BaseCoverage.subset(s.glyphs)
+        self.BaseArray.BaseRecord = [self.BaseArray.BaseRecord[i]
+                                     for i in base_indices]
+        self.BaseArray.BaseCount = len(self.BaseArray.BaseRecord)
+        # Prune empty classes
+        class_indices = _uniq_sort(v.Class for v in self.MarkArray.MarkRecord)
+        self.ClassCount = len(class_indices)
+        for m in self.MarkArray.MarkRecord:
+            m.Class = class_indices.index(m.Class)
+        for b in self.BaseArray.BaseRecord:
+            b.BaseAnchor = [b.BaseAnchor[i] for i in class_indices]
+        return bool(self.ClassCount and
+                    self.MarkArray.MarkCount and
+                    self.BaseArray.BaseCount)
+    else:
+        assert 0, "unknown format: %s" % self.Format
+
+@_add_method(otTables.MarkBasePos)
+def prune_post_subset(self, options):
+        if not options.hinting:
+            for m in self.MarkArray.MarkRecord:
+                if m.MarkAnchor:
+                    m.MarkAnchor.prune_hints()
+            for b in self.BaseArray.BaseRecord:
+                for a in b.BaseAnchor:
+                    if a:
+                        a.prune_hints()
+        return True
+
+@_add_method(otTables.MarkLigPos)
+def subset_glyphs(self, s):
+    if self.Format == 1:
+        mark_indices = self.MarkCoverage.subset(s.glyphs)
+        self.MarkArray.MarkRecord = [self.MarkArray.MarkRecord[i]
+                                     for i in mark_indices]
+        self.MarkArray.MarkCount = len(self.MarkArray.MarkRecord)
+        ligature_indices = self.LigatureCoverage.subset(s.glyphs)
+        self.LigatureArray.LigatureAttach = [self.LigatureArray.LigatureAttach[i]
+                                             for i in ligature_indices]
+        self.LigatureArray.LigatureCount = len(self.LigatureArray.LigatureAttach)
+        # Prune empty classes
+        class_indices = _uniq_sort(v.Class for v in self.MarkArray.MarkRecord)
+        self.ClassCount = len(class_indices)
+        for m in self.MarkArray.MarkRecord:
+            m.Class = class_indices.index(m.Class)
+        for l in self.LigatureArray.LigatureAttach:
+            for c in l.ComponentRecord:
+                c.LigatureAnchor = [c.LigatureAnchor[i] for i in class_indices]
+        return bool(self.ClassCount and
+                    self.MarkArray.MarkCount and
+                    self.LigatureArray.LigatureCount)
+    else:
+        assert 0, "unknown format: %s" % self.Format
+
+@_add_method(otTables.MarkLigPos)
+def prune_post_subset(self, options):
+        if not options.hinting:
+            for m in self.MarkArray.MarkRecord:
+                if m.MarkAnchor:
+                    m.MarkAnchor.prune_hints()
+            for l in self.LigatureArray.LigatureAttach:
+                for c in l.ComponentRecord:
+                    for a in c.LigatureAnchor:
+                        if a:
+                            a.prune_hints()
+        return True
+
+@_add_method(otTables.MarkMarkPos)
+def subset_glyphs(self, s):
+    if self.Format == 1:
+        mark1_indices = self.Mark1Coverage.subset(s.glyphs)
+        self.Mark1Array.MarkRecord = [self.Mark1Array.MarkRecord[i]
+                                      for i in mark1_indices]
+        self.Mark1Array.MarkCount = len(self.Mark1Array.MarkRecord)
+        mark2_indices = self.Mark2Coverage.subset(s.glyphs)
+        self.Mark2Array.Mark2Record = [self.Mark2Array.Mark2Record[i]
+                                       for i in mark2_indices]
+        self.Mark2Array.MarkCount = len(self.Mark2Array.Mark2Record)
+        # Prune empty classes
+        class_indices = _uniq_sort(v.Class for v in self.Mark1Array.MarkRecord)
+        self.ClassCount = len(class_indices)
+        for m in self.Mark1Array.MarkRecord:
+            m.Class = class_indices.index(m.Class)
+        for b in self.Mark2Array.Mark2Record:
+            b.Mark2Anchor = [b.Mark2Anchor[i] for i in class_indices]
+        return bool(self.ClassCount and
+                    self.Mark1Array.MarkCount and
+                    self.Mark2Array.MarkCount)
+    else:
+        assert 0, "unknown format: %s" % self.Format
+
+@_add_method(otTables.MarkMarkPos)
+def prune_post_subset(self, options):
+        if not options.hinting:
+            # Drop device tables or contour anchor point
+            for m in self.Mark1Array.MarkRecord:
+                if m.MarkAnchor:
+                    m.MarkAnchor.prune_hints()
+            for b in self.Mark2Array.Mark2Record:
+                for m in b.Mark2Anchor:
+                    if m:
+                        m.prune_hints()
+        return True
+
+@_add_method(otTables.SingleSubst,
+             otTables.MultipleSubst,
+             otTables.AlternateSubst,
+             otTables.LigatureSubst,
+             otTables.ReverseChainSingleSubst,
+             otTables.SinglePos,
+             otTables.PairPos,
+             otTables.CursivePos,
+             otTables.MarkBasePos,
+             otTables.MarkLigPos,
+             otTables.MarkMarkPos)
+def subset_lookups(self, lookup_indices):
+    pass
+
+@_add_method(otTables.SingleSubst,
+             otTables.MultipleSubst,
+             otTables.AlternateSubst,
+             otTables.LigatureSubst,
+             otTables.ReverseChainSingleSubst,
+             otTables.SinglePos,
+             otTables.PairPos,
+             otTables.CursivePos,
+             otTables.MarkBasePos,
+             otTables.MarkLigPos,
+             otTables.MarkMarkPos)
+def collect_lookups(self):
+    return []
+
+@_add_method(otTables.SingleSubst,
+             otTables.MultipleSubst,
+             otTables.AlternateSubst,
+             otTables.LigatureSubst,
+             otTables.ReverseChainSingleSubst,
+             otTables.ContextSubst,
+             otTables.ChainContextSubst,
+             otTables.ContextPos,
+             otTables.ChainContextPos)
+def prune_post_subset(self, options):
+    return True
+
+@_add_method(otTables.SingleSubst,
+             otTables.AlternateSubst,
+             otTables.ReverseChainSingleSubst)
+def may_have_non_1to1(self):
+    return False
+
+@_add_method(otTables.MultipleSubst,
+             otTables.LigatureSubst,
+             otTables.ContextSubst,
+             otTables.ChainContextSubst)
+def may_have_non_1to1(self):
+    return True
+
+@_add_method(otTables.ContextSubst,
+             otTables.ChainContextSubst,
+             otTables.ContextPos,
+             otTables.ChainContextPos)
+def __subset_classify_context(self):
+
+    class ContextHelper(object):
+        def __init__(self, klass, Format):
+            if klass.__name__.endswith('Subst'):
+                Typ = 'Sub'
+                Type = 'Subst'
+            else:
+                Typ = 'Pos'
+                Type = 'Pos'
+            if klass.__name__.startswith('Chain'):
+                Chain = 'Chain'
+            else:
+                Chain = ''
+            ChainTyp = Chain+Typ
+
+            self.Typ = Typ
+            self.Type = Type
+            self.Chain = Chain
+            self.ChainTyp = ChainTyp
+
+            self.LookupRecord = Type+'LookupRecord'
+
+            if Format == 1:
+                Coverage = lambda r: r.Coverage
+                ChainCoverage = lambda r: r.Coverage
+                ContextData = lambda r:(None,)
+                ChainContextData = lambda r:(None, None, None)
+                RuleData = lambda r:(r.Input,)
+                ChainRuleData = lambda r:(r.Backtrack, r.Input, r.LookAhead)
+                SetRuleData = None
+                ChainSetRuleData = None
+            elif Format == 2:
+                Coverage = lambda r: r.Coverage
+                ChainCoverage = lambda r: r.Coverage
+                ContextData = lambda r:(r.ClassDef,)
+                ChainContextData = lambda r:(r.BacktrackClassDef,
+                                             r.InputClassDef,
+                                             r.LookAheadClassDef)
+                RuleData = lambda r:(r.Class,)
+                ChainRuleData = lambda r:(r.Backtrack, r.Input, r.LookAhead)
+                def SetRuleData(r, d):(r.Class,) = d
+                def ChainSetRuleData(r, d):(r.Backtrack, r.Input, r.LookAhead) = d
+            elif Format == 3:
+                Coverage = lambda r: r.Coverage[0]
+                ChainCoverage = lambda r: r.InputCoverage[0]
+                ContextData = None
+                ChainContextData = None
+                RuleData = lambda r: r.Coverage
+                ChainRuleData = lambda r:(r.BacktrackCoverage +
+                                          r.InputCoverage +
+                                          r.LookAheadCoverage)
+                SetRuleData = None
+                ChainSetRuleData = None
+            else:
+                assert 0, "unknown format: %s" % Format
+
+            if Chain:
+                self.Coverage = ChainCoverage
+                self.ContextData = ChainContextData
+                self.RuleData = ChainRuleData
+                self.SetRuleData = ChainSetRuleData
+            else:
+                self.Coverage = Coverage
+                self.ContextData = ContextData
+                self.RuleData = RuleData
+                self.SetRuleData = SetRuleData
+
+            if Format == 1:
+                self.Rule = ChainTyp+'Rule'
+                self.RuleCount = ChainTyp+'RuleCount'
+                self.RuleSet = ChainTyp+'RuleSet'
+                self.RuleSetCount = ChainTyp+'RuleSetCount'
+                self.Intersect = lambda glyphs, c, r: [r] if r in glyphs else []
+            elif Format == 2:
+                self.Rule = ChainTyp+'ClassRule'
+                self.RuleCount = ChainTyp+'ClassRuleCount'
+                self.RuleSet = ChainTyp+'ClassSet'
+                self.RuleSetCount = ChainTyp+'ClassSetCount'
+                self.Intersect = lambda glyphs, c, r: (c.intersect_class(glyphs, r) if c
+                                                                                             else (set(glyphs) if r == 0 else set()))
+
+                self.ClassDef = 'InputClassDef' if Chain else 'ClassDef'
+                self.ClassDefIndex = 1 if Chain else 0
+                self.Input = 'Input' if Chain else 'Class'
+
+    if self.Format not in [1, 2, 3]:
+        return None    # Don't shoot the messenger; let it go
+    if not hasattr(self.__class__, "__ContextHelpers"):
+        self.__class__.__ContextHelpers = {}
+    if self.Format not in self.__class__.__ContextHelpers:
+        helper = ContextHelper(self.__class__, self.Format)
+        self.__class__.__ContextHelpers[self.Format] = helper
+    return self.__class__.__ContextHelpers[self.Format]
+
+@_add_method(otTables.ContextSubst,
+             otTables.ChainContextSubst)
+def closure_glyphs(self, s, cur_glyphs):
+    c = self.__subset_classify_context()
+
+    indices = c.Coverage(self).intersect(cur_glyphs)
+    if not indices:
+        return []
+    cur_glyphs = c.Coverage(self).intersect_glyphs(cur_glyphs)
+
+    if self.Format == 1:
+        ContextData = c.ContextData(self)
+        rss = getattr(self, c.RuleSet)
+        rssCount = getattr(self, c.RuleSetCount)
+        for i in indices:
+            if i >= rssCount or not rss[i]: continue
+            for r in getattr(rss[i], c.Rule):
+                if not r: continue
+                if not all(all(c.Intersect(s.glyphs, cd, k) for k in klist)
+                           for cd,klist in zip(ContextData, c.RuleData(r))):
+                    continue
+                chaos = set()
+                for ll in getattr(r, c.LookupRecord):
+                    if not ll: continue
+                    seqi = ll.SequenceIndex
+                    if seqi in chaos:
+                        # TODO Can we improve this?
+                        pos_glyphs = None
+                    else:
+                        if seqi == 0:
+                            pos_glyphs = frozenset([c.Coverage(self).glyphs[i]])
+                        else:
+                            pos_glyphs = frozenset([r.Input[seqi - 1]])
+                    lookup = s.table.LookupList.Lookup[ll.LookupListIndex]
+                    chaos.add(seqi)
+                    if lookup.may_have_non_1to1():
+                        chaos.update(range(seqi, len(r.Input)+2))
+                    lookup.closure_glyphs(s, cur_glyphs=pos_glyphs)
+    elif self.Format == 2:
+        ClassDef = getattr(self, c.ClassDef)
+        indices = ClassDef.intersect(cur_glyphs)
+        ContextData = c.ContextData(self)
+        rss = getattr(self, c.RuleSet)
+        rssCount = getattr(self, c.RuleSetCount)
+        for i in indices:
+            if i >= rssCount or not rss[i]: continue
+            for r in getattr(rss[i], c.Rule):
+                if not r: continue
+                if not all(all(c.Intersect(s.glyphs, cd, k) for k in klist)
+                           for cd,klist in zip(ContextData, c.RuleData(r))):
+                    continue
+                chaos = set()
+                for ll in getattr(r, c.LookupRecord):
+                    if not ll: continue
+                    seqi = ll.SequenceIndex
+                    if seqi in chaos:
+                        # TODO Can we improve this?
+                        pos_glyphs = None
+                    else:
+                        if seqi == 0:
+                            pos_glyphs = frozenset(ClassDef.intersect_class(cur_glyphs, i))
+                        else:
+                            pos_glyphs = frozenset(ClassDef.intersect_class(s.glyphs, getattr(r, c.Input)[seqi - 1]))
+                    lookup = s.table.LookupList.Lookup[ll.LookupListIndex]
+                    chaos.add(seqi)
+                    if lookup.may_have_non_1to1():
+                        chaos.update(range(seqi, len(getattr(r, c.Input))+2))
+                    lookup.closure_glyphs(s, cur_glyphs=pos_glyphs)
+    elif self.Format == 3:
+        if not all(x.intersect(s.glyphs) for x in c.RuleData(self)):
+            return []
+        r = self
+        chaos = set()
+        for ll in getattr(r, c.LookupRecord):
+            if not ll: continue
+            seqi = ll.SequenceIndex
+            if seqi in chaos:
+                # TODO Can we improve this?
+                pos_glyphs = None
+            else:
+                if seqi == 0:
+                    pos_glyphs = frozenset(cur_glyphs)
+                else:
+                    pos_glyphs = frozenset(r.InputCoverage[seqi].intersect_glyphs(s.glyphs))
+            lookup = s.table.LookupList.Lookup[ll.LookupListIndex]
+            chaos.add(seqi)
+            if lookup.may_have_non_1to1():
+                chaos.update(range(seqi, len(r.InputCoverage)+1))
+            lookup.closure_glyphs(s, cur_glyphs=pos_glyphs)
+    else:
+        assert 0, "unknown format: %s" % self.Format
+
+@_add_method(otTables.ContextSubst,
+             otTables.ContextPos,
+             otTables.ChainContextSubst,
+             otTables.ChainContextPos)
+def subset_glyphs(self, s):
+    c = self.__subset_classify_context()
+
+    if self.Format == 1:
+        indices = self.Coverage.subset(s.glyphs)
+        rss = getattr(self, c.RuleSet)
+        rssCount = getattr(self, c.RuleSetCount)
+        rss = [rss[i] for i in indices if i < rssCount]
+        for rs in rss:
+            if not rs: continue
+            ss = getattr(rs, c.Rule)
+            ss = [r for r in ss
+                  if r and all(all(g in s.glyphs for g in glist)
+                               for glist in c.RuleData(r))]
+            setattr(rs, c.Rule, ss)
+            setattr(rs, c.RuleCount, len(ss))
+        # Prune empty rulesets
+        indices = [i for i,rs in enumerate(rss) if rs and getattr(rs, c.Rule)]
+        self.Coverage.remap(indices)
+        rss = [rss[i] for i in indices]
+        setattr(self, c.RuleSet, rss)
+        setattr(self, c.RuleSetCount, len(rss))
+        return bool(rss)
+    elif self.Format == 2:
+        if not self.Coverage.subset(s.glyphs):
+            return False
+        ContextData = c.ContextData(self)
+        klass_maps = [x.subset(s.glyphs, remap=True) if x else None for x in ContextData]
+
+        # Keep rulesets for class numbers that survived.
+        indices = klass_maps[c.ClassDefIndex]
+        rss = getattr(self, c.RuleSet)
+        rssCount = getattr(self, c.RuleSetCount)
+        rss = [rss[i] for i in indices if i < rssCount]
+        del rssCount
+        # Delete, but not renumber, unreachable rulesets.
+        indices = getattr(self, c.ClassDef).intersect(self.Coverage.glyphs)
+        rss = [rss if i in indices else None for i,rss in enumerate(rss)]
+
+        for rs in rss:
+            if not rs: continue
+            ss = getattr(rs, c.Rule)
+            ss = [r for r in ss
+                  if r and all(all(k in klass_map for k in klist)
+                               for klass_map,klist in zip(klass_maps, c.RuleData(r)))]
+            setattr(rs, c.Rule, ss)
+            setattr(rs, c.RuleCount, len(ss))
+
+            # Remap rule classes
+            for r in ss:
+                c.SetRuleData(r, [[klass_map.index(k) for k in klist]
+                                  for klass_map,klist in zip(klass_maps, c.RuleData(r))])
+
+        # Prune empty rulesets
+        rss = [rs if rs and getattr(rs, c.Rule) else None for rs in rss]
+        while rss and rss[-1] is None:
+            del rss[-1]
+        setattr(self, c.RuleSet, rss)
+        setattr(self, c.RuleSetCount, len(rss))
+
+        # TODO: We can do a second round of remapping class values based
+        # on classes that are actually used in at least one rule.    Right
+        # now we subset classes to c.glyphs only.    Or better, rewrite
+        # the above to do that.
+
+        return bool(rss)
+    elif self.Format == 3:
+        return all(x.subset(s.glyphs) for x in c.RuleData(self))
+    else:
+        assert 0, "unknown format: %s" % self.Format
+
+@_add_method(otTables.ContextSubst,
+             otTables.ChainContextSubst,
+             otTables.ContextPos,
+             otTables.ChainContextPos)
+def subset_lookups(self, lookup_indices):
+    c = self.__subset_classify_context()
+
+    if self.Format in [1, 2]:
+        for rs in getattr(self, c.RuleSet):
+            if not rs: continue
+            for r in getattr(rs, c.Rule):
+                if not r: continue
+                setattr(r, c.LookupRecord,
+                        [ll for ll in getattr(r, c.LookupRecord)
+                         if ll and ll.LookupListIndex in lookup_indices])
+                for ll in getattr(r, c.LookupRecord):
+                    if not ll: continue
+                    ll.LookupListIndex = lookup_indices.index(ll.LookupListIndex)
+    elif self.Format == 3:
+        setattr(self, c.LookupRecord,
+                [ll for ll in getattr(self, c.LookupRecord)
+                 if ll and ll.LookupListIndex in lookup_indices])
+        for ll in getattr(self, c.LookupRecord):
+            if not ll: continue
+            ll.LookupListIndex = lookup_indices.index(ll.LookupListIndex)
+    else:
+        assert 0, "unknown format: %s" % self.Format
+
+@_add_method(otTables.ContextSubst,
+             otTables.ChainContextSubst,
+             otTables.ContextPos,
+             otTables.ChainContextPos)
+def collect_lookups(self):
+    c = self.__subset_classify_context()
+
+    if self.Format in [1, 2]:
+        return [ll.LookupListIndex
+            for rs in getattr(self, c.RuleSet) if rs
+            for r in getattr(rs, c.Rule) if r
+            for ll in getattr(r, c.LookupRecord) if ll]
+    elif self.Format == 3:
+        return [ll.LookupListIndex
+            for ll in getattr(self, c.LookupRecord) if ll]
+    else:
+        assert 0, "unknown format: %s" % self.Format
+
+@_add_method(otTables.ExtensionSubst)
+def closure_glyphs(self, s, cur_glyphs):
+    if self.Format == 1:
+        self.ExtSubTable.closure_glyphs(s, cur_glyphs)
+    else:
+        assert 0, "unknown format: %s" % self.Format
+
+@_add_method(otTables.ExtensionSubst)
+def may_have_non_1to1(self):
+    if self.Format == 1:
+        return self.ExtSubTable.may_have_non_1to1()
+    else:
+        assert 0, "unknown format: %s" % self.Format
+
+@_add_method(otTables.ExtensionSubst,
+             otTables.ExtensionPos)
+def subset_glyphs(self, s):
+    if self.Format == 1:
+        return self.ExtSubTable.subset_glyphs(s)
+    else:
+        assert 0, "unknown format: %s" % self.Format
+
+@_add_method(otTables.ExtensionSubst,
+             otTables.ExtensionPos)
+def prune_post_subset(self, options):
+    if self.Format == 1:
+        return self.ExtSubTable.prune_post_subset(options)
+    else:
+        assert 0, "unknown format: %s" % self.Format
+
+@_add_method(otTables.ExtensionSubst,
+             otTables.ExtensionPos)
+def subset_lookups(self, lookup_indices):
+    if self.Format == 1:
+        return self.ExtSubTable.subset_lookups(lookup_indices)
+    else:
+        assert 0, "unknown format: %s" % self.Format
+
+@_add_method(otTables.ExtensionSubst,
+             otTables.ExtensionPos)
+def collect_lookups(self):
+    if self.Format == 1:
+        return self.ExtSubTable.collect_lookups()
+    else:
+        assert 0, "unknown format: %s" % self.Format
+
+@_add_method(otTables.Lookup)
+def closure_glyphs(self, s, cur_glyphs=None):
+    if cur_glyphs is None:
+        cur_glyphs = frozenset(s.glyphs)
+
+    # Memoize
+    if (id(self), cur_glyphs) in s._doneLookups:
+        return
+    s._doneLookups.add((id(self), cur_glyphs))
+
+    if self in s._activeLookups:
+        raise Exception("Circular loop in lookup recursion")
+    s._activeLookups.append(self)
+    for st in self.SubTable:
+        if not st: continue
+        st.closure_glyphs(s, cur_glyphs)
+    assert(s._activeLookups[-1] == self)
+    del s._activeLookups[-1]
+
+@_add_method(otTables.Lookup)
+def subset_glyphs(self, s):
+    self.SubTable = [st for st in self.SubTable if st and st.subset_glyphs(s)]
+    self.SubTableCount = len(self.SubTable)
+    return bool(self.SubTableCount)
+
+@_add_method(otTables.Lookup)
+def prune_post_subset(self, options):
+    ret = False
+    for st in self.SubTable:
+        if not st: continue
+        if st.prune_post_subset(options): ret = True
+    return ret
+
+@_add_method(otTables.Lookup)
+def subset_lookups(self, lookup_indices):
+    for s in self.SubTable:
+        s.subset_lookups(lookup_indices)
+
+@_add_method(otTables.Lookup)
+def collect_lookups(self):
+    return _uniq_sort(sum((st.collect_lookups() for st in self.SubTable
+                           if st), []))
+
+@_add_method(otTables.Lookup)
+def may_have_non_1to1(self):
+    return any(st.may_have_non_1to1() for st in self.SubTable if st)
+
+@_add_method(otTables.LookupList)
+def subset_glyphs(self, s):
+    """Returns the indices of nonempty lookups."""
+    return [i for i,l in enumerate(self.Lookup) if l and l.subset_glyphs(s)]
+
+@_add_method(otTables.LookupList)
+def prune_post_subset(self, options):
+    ret = False
+    for l in self.Lookup:
+        if not l: continue
+        if l.prune_post_subset(options): ret = True
+    return ret
+
+@_add_method(otTables.LookupList)
+def subset_lookups(self, lookup_indices):
+    self.ensureDecompiled()
+    self.Lookup = [self.Lookup[i] for i in lookup_indices
+                   if i < self.LookupCount]
+    self.LookupCount = len(self.Lookup)
+    for l in self.Lookup:
+        l.subset_lookups(lookup_indices)
+
+@_add_method(otTables.LookupList)
+def neuter_lookups(self, lookup_indices):
+    """Sets lookups not in lookup_indices to None."""
+    self.ensureDecompiled()
+    self.Lookup = [l if i in lookup_indices else None for i,l in enumerate(self.Lookup)]
+
+@_add_method(otTables.LookupList)
+def closure_lookups(self, lookup_indices):
+    lookup_indices = _uniq_sort(lookup_indices)
+    recurse = lookup_indices
+    while True:
+        recurse_lookups = sum((self.Lookup[i].collect_lookups()
+                               for i in recurse if i < self.LookupCount), [])
+        recurse_lookups = [l for l in recurse_lookups
+                           if l not in lookup_indices and l < self.LookupCount]
+        if not recurse_lookups:
+            return _uniq_sort(lookup_indices)
+        recurse_lookups = _uniq_sort(recurse_lookups)
+        lookup_indices.extend(recurse_lookups)
+        recurse = recurse_lookups
+
+@_add_method(otTables.Feature)
+def subset_lookups(self, lookup_indices):
+    self.LookupListIndex = [l for l in self.LookupListIndex
+                            if l in lookup_indices]
+    # Now map them.
+    self.LookupListIndex = [lookup_indices.index(l)
+                            for l in self.LookupListIndex]
+    self.LookupCount = len(self.LookupListIndex)
+    return self.LookupCount or self.FeatureParams
+
+@_add_method(otTables.Feature)
+def collect_lookups(self):
+    return self.LookupListIndex[:]
+
+@_add_method(otTables.FeatureList)
+def subset_lookups(self, lookup_indices):
+    """Returns the indices of nonempty features."""
+    # Note: Never ever drop feature 'pref', even if it's empty.
+    # HarfBuzz chooses shaper for Khmer based on presence of this
+    # feature.    See thread at:
+    # http://lists.freedesktop.org/archives/harfbuzz/2012-November/002660.html
+    feature_indices = [i for i,f in enumerate(self.FeatureRecord)
+                       if (f.Feature.subset_lookups(lookup_indices) or
+                           f.FeatureTag == 'pref')]
+    self.subset_features(feature_indices)
+    return feature_indices
+
+@_add_method(otTables.FeatureList)
+def collect_lookups(self, feature_indices):
+    return _uniq_sort(sum((self.FeatureRecord[i].Feature.collect_lookups()
+                           for i in feature_indices
+                           if i < self.FeatureCount), []))
+
+@_add_method(otTables.FeatureList)
+def subset_features(self, feature_indices):
+    self.ensureDecompiled()
+    self.FeatureRecord = [self.FeatureRecord[i] for i in feature_indices]
+    self.FeatureCount = len(self.FeatureRecord)
+    return bool(self.FeatureCount)
+
+@_add_method(otTables.DefaultLangSys,
+             otTables.LangSys)
+def subset_features(self, feature_indices):
+    if self.ReqFeatureIndex in feature_indices:
+        self.ReqFeatureIndex = feature_indices.index(self.ReqFeatureIndex)
+    else:
+        self.ReqFeatureIndex = 65535
+    self.FeatureIndex = [f for f in self.FeatureIndex if f in feature_indices]
+    # Now map them.
+    self.FeatureIndex = [feature_indices.index(f) for f in self.FeatureIndex
+                         if f in feature_indices]
+    self.FeatureCount = len(self.FeatureIndex)
+    return bool(self.FeatureCount or self.ReqFeatureIndex != 65535)
+
+@_add_method(otTables.DefaultLangSys,
+             otTables.LangSys)
+def collect_features(self):
+    feature_indices = self.FeatureIndex[:]
+    if self.ReqFeatureIndex != 65535:
+        feature_indices.append(self.ReqFeatureIndex)
+    return _uniq_sort(feature_indices)
+
+@_add_method(otTables.Script)
+def subset_features(self, feature_indices):
+    if(self.DefaultLangSys and
+       not self.DefaultLangSys.subset_features(feature_indices)):
+        self.DefaultLangSys = None
+    self.LangSysRecord = [l for l in self.LangSysRecord
+                          if l.LangSys.subset_features(feature_indices)]
+    self.LangSysCount = len(self.LangSysRecord)
+    return bool(self.LangSysCount or self.DefaultLangSys)
+
+@_add_method(otTables.Script)
+def collect_features(self):
+    feature_indices = [l.LangSys.collect_features() for l in self.LangSysRecord]
+    if self.DefaultLangSys:
+        feature_indices.append(self.DefaultLangSys.collect_features())
+    return _uniq_sort(sum(feature_indices, []))
+
+@_add_method(otTables.ScriptList)
+def subset_features(self, feature_indices):
+    self.ScriptRecord = [s for s in self.ScriptRecord
+                         if s.Script.subset_features(feature_indices)]
+    self.ScriptCount = len(self.ScriptRecord)
+    return bool(self.ScriptCount)
+
+@_add_method(otTables.ScriptList)
+def collect_features(self):
+    return _uniq_sort(sum((s.Script.collect_features()
+                           for s in self.ScriptRecord), []))
+
+@_add_method(ttLib.getTableClass('GSUB'))
+def closure_glyphs(self, s):
+    s.table = self.table
+    if self.table.ScriptList:
+        feature_indices = self.table.ScriptList.collect_features()
+    else:
+        feature_indices = []
+    if self.table.FeatureList:
+        lookup_indices = self.table.FeatureList.collect_lookups(feature_indices)
+    else:
+        lookup_indices = []
+    if self.table.LookupList:
+        while True:
+            orig_glyphs = frozenset(s.glyphs)
+            s._activeLookups = []
+            s._doneLookups = set()
+            for i in lookup_indices:
+                if i >= self.table.LookupList.LookupCount: continue
+                if not self.table.LookupList.Lookup[i]: continue
+                self.table.LookupList.Lookup[i].closure_glyphs(s)
+            del s._activeLookups, s._doneLookups
+            if orig_glyphs == s.glyphs:
+                break
+    del s.table
+
+@_add_method(ttLib.getTableClass('GSUB'),
+             ttLib.getTableClass('GPOS'))
+def subset_glyphs(self, s):
+    s.glyphs = s.glyphs_gsubed
+    if self.table.LookupList:
+        lookup_indices = self.table.LookupList.subset_glyphs(s)
+    else:
+        lookup_indices = []
+    self.subset_lookups(lookup_indices)
+    return True
+
+@_add_method(ttLib.getTableClass('GSUB'),
+             ttLib.getTableClass('GPOS'))
+def subset_lookups(self, lookup_indices):
+    """Retains specified lookups, then removes empty features, language
+    systems, and scripts."""
+    if self.table.LookupList:
+        self.table.LookupList.subset_lookups(lookup_indices)
+    if self.table.FeatureList:
+        feature_indices = self.table.FeatureList.subset_lookups(lookup_indices)
+    else:
+        feature_indices = []
+    if self.table.ScriptList:
+        self.table.ScriptList.subset_features(feature_indices)
+
+@_add_method(ttLib.getTableClass('GSUB'),
+             ttLib.getTableClass('GPOS'))
+def neuter_lookups(self, lookup_indices):
+    """Sets lookups not in lookup_indices to None."""
+    if self.table.LookupList:
+        self.table.LookupList.neuter_lookups(lookup_indices)
+
+@_add_method(ttLib.getTableClass('GSUB'),
+             ttLib.getTableClass('GPOS'))
+def prune_lookups(self, remap=True):
+    """Remove (default) or neuter unreferenced lookups"""
+    if self.table.ScriptList:
+        feature_indices = self.table.ScriptList.collect_features()
+    else:
+        feature_indices = []
+    if self.table.FeatureList:
+        lookup_indices = self.table.FeatureList.collect_lookups(feature_indices)
+    else:
+        lookup_indices = []
+    if self.table.LookupList:
+        lookup_indices = self.table.LookupList.closure_lookups(lookup_indices)
+    else:
+        lookup_indices = []
+    if remap:
+        self.subset_lookups(lookup_indices)
+    else:
+        self.neuter_lookups(lookup_indices)
+
+@_add_method(ttLib.getTableClass('GSUB'),
+                         ttLib.getTableClass('GPOS'))
+def subset_feature_tags(self, feature_tags):
+    if self.table.FeatureList:
+        feature_indices = \
+            [i for i,f in enumerate(self.table.FeatureList.FeatureRecord)
+             if f.FeatureTag in feature_tags]
+        self.table.FeatureList.subset_features(feature_indices)
+    else:
+        feature_indices = []
+    if self.table.ScriptList:
+        self.table.ScriptList.subset_features(feature_indices)
+
+@_add_method(ttLib.getTableClass('GSUB'),
+             ttLib.getTableClass('GPOS'))
+def prune_features(self):
+    """Remove unreferenced features"""
+    if self.table.ScriptList:
+        feature_indices = self.table.ScriptList.collect_features()
+    else:
+        feature_indices = []
+    if self.table.FeatureList:
+        self.table.FeatureList.subset_features(feature_indices)
+    if self.table.ScriptList:
+        self.table.ScriptList.subset_features(feature_indices)
+
+@_add_method(ttLib.getTableClass('GSUB'),
+             ttLib.getTableClass('GPOS'))
+def prune_pre_subset(self, options):
+    # Drop undesired features
+    if '*' not in options.layout_features:
+        self.subset_feature_tags(options.layout_features)
+    # Neuter unreferenced lookups
+    self.prune_lookups(remap=False)
+    return True
+
+@_add_method(ttLib.getTableClass('GSUB'),
+             ttLib.getTableClass('GPOS'))
+def remove_redundant_langsys(self):
+    table = self.table
+    if not table.ScriptList or not table.FeatureList:
+        return
+
+    features = table.FeatureList.FeatureRecord
+
+    for s in table.ScriptList.ScriptRecord:
+        d = s.Script.DefaultLangSys
+        if not d:
+            continue
+        for lr in s.Script.LangSysRecord[:]:
+            l = lr.LangSys
+            # Compare d and l
+            if len(d.FeatureIndex) != len(l.FeatureIndex):
+                continue
+            if (d.ReqFeatureIndex == 65535) != (l.ReqFeatureIndex == 65535):
+                continue
+
+            if d.ReqFeatureIndex != 65535:
+                if features[d.ReqFeatureIndex] != features[l.ReqFeatureIndex]:
+                    continue
+
+            for i in range(len(d.FeatureIndex)):
+                if features[d.FeatureIndex[i]] != features[l.FeatureIndex[i]]:
+                    break
+            else:
+                # LangSys and default are equal; delete LangSys
+                s.Script.LangSysRecord.remove(lr)
+
+@_add_method(ttLib.getTableClass('GSUB'),
+             ttLib.getTableClass('GPOS'))
+def prune_post_subset(self, options):
+    table = self.table
+
+    self.prune_lookups() # XXX Is this actually needed?!
+
+    if table.LookupList:
+        table.LookupList.prune_post_subset(options)
+        # XXX Next two lines disabled because OTS is stupid and
+        # doesn't like NULL offsets here.
+        #if not table.LookupList.Lookup:
+        #    table.LookupList = None
+
+    if not table.LookupList:
+        table.FeatureList = None
+
+    if table.FeatureList:
+        self.remove_redundant_langsys()
+        # Remove unreferenced features
+        self.prune_features()
+
+    # XXX Next two lines disabled because OTS is stupid and
+    # doesn't like NULL offsets here.
+    #if table.FeatureList and not table.FeatureList.FeatureRecord:
+    #    table.FeatureList = None
+
+    # Never drop scripts themselves as them just being available
+    # holds semantic significance.
+    # XXX Next two lines disabled because OTS is stupid and
+    # doesn't like NULL offsets here.
+    #if table.ScriptList and not table.ScriptList.ScriptRecord:
+    #    table.ScriptList = None
+
+    return True
+
+@_add_method(ttLib.getTableClass('GDEF'))
+def subset_glyphs(self, s):
+    glyphs = s.glyphs_gsubed
+    table = self.table
+    if table.LigCaretList:
+        indices = table.LigCaretList.Coverage.subset(glyphs)
+        table.LigCaretList.LigGlyph = [table.LigCaretList.LigGlyph[i]
+                                       for i in indices]
+        table.LigCaretList.LigGlyphCount = len(table.LigCaretList.LigGlyph)
+    if table.MarkAttachClassDef:
+        table.MarkAttachClassDef.classDefs = \
+            {g:v for g,v in table.MarkAttachClassDef.classDefs.items()
+             if g in glyphs}
+    if table.GlyphClassDef:
+        table.GlyphClassDef.classDefs = \
+            {g:v for g,v in table.GlyphClassDef.classDefs.items()
+             if g in glyphs}
+    if table.AttachList:
+        indices = table.AttachList.Coverage.subset(glyphs)
+        GlyphCount = table.AttachList.GlyphCount
+        table.AttachList.AttachPoint = [table.AttachList.AttachPoint[i]
+                                        for i in indices
+                                        if i < GlyphCount]
+        table.AttachList.GlyphCount = len(table.AttachList.AttachPoint)
+    if hasattr(table, "MarkGlyphSetsDef") and table.MarkGlyphSetsDef:
+        for coverage in table.MarkGlyphSetsDef.Coverage:
+            coverage.subset(glyphs)
+        # TODO: The following is disabled. If enabling, we need to go fixup all
+        # lookups that use MarkFilteringSet and map their set.
+        # indices = table.MarkGlyphSetsDef.Coverage = \
+        #   [c for c in table.MarkGlyphSetsDef.Coverage if c.glyphs]
+    return True
+
+@_add_method(ttLib.getTableClass('GDEF'))
+def prune_post_subset(self, options):
+    table = self.table
+    # XXX check these against OTS
+    if table.LigCaretList and not table.LigCaretList.LigGlyphCount:
+        table.LigCaretList = None
+    if table.MarkAttachClassDef and not table.MarkAttachClassDef.classDefs:
+        table.MarkAttachClassDef = None
+    if table.GlyphClassDef and not table.GlyphClassDef.classDefs:
+        table.GlyphClassDef = None
+    if table.AttachList and not table.AttachList.GlyphCount:
+        table.AttachList = None
+    if (hasattr(table, "MarkGlyphSetsDef") and
+        table.MarkGlyphSetsDef and
+        not table.MarkGlyphSetsDef.Coverage):
+        table.MarkGlyphSetsDef = None
+        if table.Version == 0x00010002/0x10000:
+            table.Version = 1.0
+    return bool(table.LigCaretList or
+                table.MarkAttachClassDef or
+                table.GlyphClassDef or
+                table.AttachList or
+                (table.Version >= 0x00010002/0x10000 and table.MarkGlyphSetsDef))
+
+@_add_method(ttLib.getTableClass('kern'))
+def prune_pre_subset(self, options):
+    # Prune unknown kern table types
+    self.kernTables = [t for t in self.kernTables if hasattr(t, 'kernTable')]
+    return bool(self.kernTables)
+
+@_add_method(ttLib.getTableClass('kern'))
+def subset_glyphs(self, s):
+    glyphs = s.glyphs_gsubed
+    for t in self.kernTables:
+        t.kernTable = {(a,b):v for (a,b),v in t.kernTable.items()
+                                           if a in glyphs and b in glyphs}
+    self.kernTables = [t for t in self.kernTables if t.kernTable]
+    return bool(self.kernTables)
+
+@_add_method(ttLib.getTableClass('vmtx'))
+def subset_glyphs(self, s):
+    self.metrics = _dict_subset(self.metrics, s.glyphs)
+    return bool(self.metrics)
+
+@_add_method(ttLib.getTableClass('hmtx'))
+def subset_glyphs(self, s):
+    self.metrics = _dict_subset(self.metrics, s.glyphs)
+    return True # Required table
+
+@_add_method(ttLib.getTableClass('hdmx'))
+def subset_glyphs(self, s):
+    self.hdmx = {sz:_dict_subset(l, s.glyphs) for sz,l in self.hdmx.items()}
+    return bool(self.hdmx)
+
+@_add_method(ttLib.getTableClass('VORG'))
+def subset_glyphs(self, s):
+    self.VOriginRecords = {g:v for g,v in self.VOriginRecords.items()
+                               if g in s.glyphs}
+    self.numVertOriginYMetrics = len(self.VOriginRecords)
+    return True    # Never drop; has default metrics
+
+@_add_method(ttLib.getTableClass('post'))
+def prune_pre_subset(self, options):
+    if not options.glyph_names:
+        self.formatType = 3.0
+    return True # Required table
+
+@_add_method(ttLib.getTableClass('post'))
+def subset_glyphs(self, s):
+    self.extraNames = []    # This seems to do it
+    return True # Required table
+
+@_add_method(ttLib.getTableModule('glyf').Glyph)
+def remapComponentsFast(self, indices):
+    if not self.data or struct.unpack(">h", self.data[:2])[0] >= 0:
+        return    # Not composite
+    data = array.array("B", self.data)
+    i = 10
+    more = 1
+    while more:
+        flags =(data[i] << 8) | data[i+1]
+        glyphID =(data[i+2] << 8) | data[i+3]
+        # Remap
+        glyphID = indices.index(glyphID)
+        data[i+2] = glyphID >> 8
+        data[i+3] = glyphID & 0xFF
+        i += 4
+        flags = int(flags)
+
+        if flags & 0x0001: i += 4    # ARG_1_AND_2_ARE_WORDS
+        else: i += 2
+        if flags & 0x0008: i += 2    # WE_HAVE_A_SCALE
+        elif flags & 0x0040: i += 4    # WE_HAVE_AN_X_AND_Y_SCALE
+        elif flags & 0x0080: i += 8    # WE_HAVE_A_TWO_BY_TWO
+        more = flags & 0x0020    # MORE_COMPONENTS
+
+    self.data = data.tostring()
+
+@_add_method(ttLib.getTableClass('glyf'))
+def closure_glyphs(self, s):
+    decompose = s.glyphs
+    while True:
+        components = set()
+        for g in decompose:
+            if g not in self.glyphs:
+                continue
+            gl = self.glyphs[g]
+            for c in gl.getComponentNames(self):
+                if c not in s.glyphs:
+                    components.add(c)
+        components = set(c for c in components if c not in s.glyphs)
+        if not components:
+            break
+        decompose = components
+        s.glyphs.update(components)
+
+@_add_method(ttLib.getTableClass('glyf'))
+def prune_pre_subset(self, options):
+    if options.notdef_glyph and not options.notdef_outline:
+        g = self[self.glyphOrder[0]]
+        # Yay, easy!
+        g.__dict__.clear()
+        g.data = ""
+    return True
+
+@_add_method(ttLib.getTableClass('glyf'))
+def subset_glyphs(self, s):
+    self.glyphs = _dict_subset(self.glyphs, s.glyphs)
+    indices = [i for i,g in enumerate(self.glyphOrder) if g in s.glyphs]
+    for v in self.glyphs.values():
+        if hasattr(v, "data"):
+            v.remapComponentsFast(indices)
+        else:
+            pass    # No need
+    self.glyphOrder = [g for g in self.glyphOrder if g in s.glyphs]
+    # Don't drop empty 'glyf' tables, otherwise 'loca' doesn't get subset.
+    return True
+
+@_add_method(ttLib.getTableClass('glyf'))
+def prune_post_subset(self, options):
+    remove_hinting = not options.hinting
+    for v in self.glyphs.values():
+        v.trim(remove_hinting=remove_hinting)
+    return True
+
+@_add_method(ttLib.getTableClass('CFF '))
+def prune_pre_subset(self, options):
+    cff = self.cff
+    # CFF table must have one font only
+    cff.fontNames = cff.fontNames[:1]
+
+    if options.notdef_glyph and not options.notdef_outline:
+        for fontname in cff.keys():
+            font = cff[fontname]
+            c,_ = font.CharStrings.getItemAndSelector('.notdef')
+            # XXX we should preserve the glyph width
+            c.bytecode = '\x0e' # endchar
+            c.program = None
+
+    return True # bool(cff.fontNames)
+
+@_add_method(ttLib.getTableClass('CFF '))
+def subset_glyphs(self, s):
+    cff = self.cff
+    for fontname in cff.keys():
+        font = cff[fontname]
+        cs = font.CharStrings
+
+        # Load all glyphs
+        for g in font.charset:
+            if g not in s.glyphs: continue
+            c,sel = cs.getItemAndSelector(g)
+
+        if cs.charStringsAreIndexed:
+            indices = [i for i,g in enumerate(font.charset) if g in s.glyphs]
+            csi = cs.charStringsIndex
+            csi.items = [csi.items[i] for i in indices]
+            del csi.file, csi.offsets
+            if hasattr(font, "FDSelect"):
+                sel = font.FDSelect
+                # XXX We want to set sel.format to None, such that the
+                # most compact format is selected. However, OTS was
+                # broken and couldn't parse a FDSelect format 0 that
+                # happened before CharStrings. As such, always force
+                # format 3 until we fix cffLib to always generate
+                # FDSelect after CharStrings.
+                # https://github.com/khaledhosny/ots/pull/31
+                #sel.format = None
+                sel.format = 3
+                sel.gidArray = [sel.gidArray[i] for i in indices]
+            cs.charStrings = {g:indices.index(v)
+                              for g,v in cs.charStrings.items()
+                              if g in s.glyphs}
+        else:
+            cs.charStrings = {g:v
+                              for g,v in cs.charStrings.items()
+                              if g in s.glyphs}
+        font.charset = [g for g in font.charset if g in s.glyphs]
+        font.numGlyphs = len(font.charset)
+
+    return True # any(cff[fontname].numGlyphs for fontname in cff.keys())
+
+@_add_method(psCharStrings.T2CharString)
+def subset_subroutines(self, subrs, gsubrs):
+    p = self.program
+    assert len(p)
+    for i in range(1, len(p)):
+        if p[i] == 'callsubr':
+            assert isinstance(p[i-1], int)
+            p[i-1] = subrs._used.index(p[i-1] + subrs._old_bias) - subrs._new_bias
+        elif p[i] == 'callgsubr':
+            assert isinstance(p[i-1], int)
+            p[i-1] = gsubrs._used.index(p[i-1] + gsubrs._old_bias) - gsubrs._new_bias
+
+@_add_method(psCharStrings.T2CharString)
+def drop_hints(self):
+    hints = self._hints
+
+    if hints.has_hint:
+        self.program = self.program[hints.last_hint:]
+        if hasattr(self, 'width'):
+            # Insert width back if needed
+            if self.width != self.private.defaultWidthX:
+                self.program.insert(0, self.width - self.private.nominalWidthX)
+
+    if hints.has_hintmask:
+        i = 0
+        p = self.program
+        while i < len(p):
+            if p[i] in ['hintmask', 'cntrmask']:
+                assert i + 1 <= len(p)
+                del p[i:i+2]
+                continue
+            i += 1
+
+    # TODO: we currently don't drop calls to "empty" subroutines.
+
+    assert len(self.program)
+
+    del self._hints
+
+class _MarkingT2Decompiler(psCharStrings.SimpleT2Decompiler):
+
+    def __init__(self, localSubrs, globalSubrs):
+        psCharStrings.SimpleT2Decompiler.__init__(self,
+                                                  localSubrs,
+                                                  globalSubrs)
+        for subrs in [localSubrs, globalSubrs]:
+            if subrs and not hasattr(subrs, "_used"):
+                subrs._used = set()
+
+    def op_callsubr(self, index):
+        self.localSubrs._used.add(self.operandStack[-1]+self.localBias)
+        psCharStrings.SimpleT2Decompiler.op_callsubr(self, index)
+
+    def op_callgsubr(self, index):
+        self.globalSubrs._used.add(self.operandStack[-1]+self.globalBias)
+        psCharStrings.SimpleT2Decompiler.op_callgsubr(self, index)
+
+class _DehintingT2Decompiler(psCharStrings.SimpleT2Decompiler):
+
+    class Hints(object):
+        def __init__(self):
+            # Whether calling this charstring produces any hint stems
+            self.has_hint = False
+            # Index to start at to drop all hints
+            self.last_hint = 0
+            # Index up to which we know more hints are possible.
+            # Only relevant if status is 0 or 1.
+            self.last_checked = 0
+            # The status means:
+            # 0: after dropping hints, this charstring is empty
+            # 1: after dropping hints, there may be more hints
+            #    continuing after this
+            # 2: no more hints possible after this charstring
+            self.status = 0
+            # Has hintmask instructions; not recursive
+            self.has_hintmask = False
+        pass
+
+    def __init__(self, css, localSubrs, globalSubrs):
+        self._css = css
+        psCharStrings.SimpleT2Decompiler.__init__(self,
+                                                  localSubrs,
+                                                  globalSubrs)
+
+    def execute(self, charString):
+        old_hints = charString._hints if hasattr(charString, '_hints') else None
+        charString._hints = self.Hints()
+
+        psCharStrings.SimpleT2Decompiler.execute(self, charString)
+
+        hints = charString._hints
+
+        if hints.has_hint or hints.has_hintmask:
+            self._css.add(charString)
+
+        if hints.status != 2:
+            # Check from last_check, make sure we didn't have any operators.
+            for i in range(hints.last_checked, len(charString.program) - 1):
+                if isinstance(charString.program[i], str):
+                    hints.status = 2
+                    break
+                else:
+                    hints.status = 1 # There's *something* here
+            hints.last_checked = len(charString.program)
+
+        if old_hints:
+            assert hints.__dict__ == old_hints.__dict__
+
+    def op_callsubr(self, index):
+        subr = self.localSubrs[self.operandStack[-1]+self.localBias]
+        psCharStrings.SimpleT2Decompiler.op_callsubr(self, index)
+        self.processSubr(index, subr)
+
+    def op_callgsubr(self, index):
+        subr = self.globalSubrs[self.operandStack[-1]+self.globalBias]
+        psCharStrings.SimpleT2Decompiler.op_callgsubr(self, index)
+        self.processSubr(index, subr)
+
+    def op_hstem(self, index):
+        psCharStrings.SimpleT2Decompiler.op_hstem(self, index)
+        self.processHint(index)
+    def op_vstem(self, index):
+        psCharStrings.SimpleT2Decompiler.op_vstem(self, index)
+        self.processHint(index)
+    def op_hstemhm(self, index):
+        psCharStrings.SimpleT2Decompiler.op_hstemhm(self, index)
+        self.processHint(index)
+    def op_vstemhm(self, index):
+        psCharStrings.SimpleT2Decompiler.op_vstemhm(self, index)
+        self.processHint(index)
+    def op_hintmask(self, index):
+        psCharStrings.SimpleT2Decompiler.op_hintmask(self, index)
+        self.processHintmask(index)
+    def op_cntrmask(self, index):
+        psCharStrings.SimpleT2Decompiler.op_cntrmask(self, index)
+        self.processHintmask(index)
+
+    def processHintmask(self, index):
+        cs = self.callingStack[-1]
+        hints = cs._hints
+        hints.has_hintmask = True
+        if hints.status != 2 and hints.has_hint:
+            # Check from last_check, see if we may be an implicit vstem
+            for i in range(hints.last_checked, index - 1):
+                if isinstance(cs.program[i], str):
+                    hints.status = 2
+                    break
+            if hints.status != 2:
+                # We are an implicit vstem
+                hints.last_hint = index + 1
+                hints.status = 0
+        hints.last_checked = index + 1
+
+    def processHint(self, index):
+        cs = self.callingStack[-1]
+        hints = cs._hints
+        hints.has_hint = True
+        hints.last_hint = index
+        hints.last_checked = index
+
+    def processSubr(self, index, subr):
+        cs = self.callingStack[-1]
+        hints = cs._hints
+        subr_hints = subr._hints
+
+        if subr_hints.has_hint:
+            if hints.status != 2:
+                hints.has_hint = True
+                hints.last_checked = index
+                hints.status = subr_hints.status
+                # Decide where to chop off from
+                if subr_hints.status == 0:
+                    hints.last_hint = index
+                else:
+                    hints.last_hint = index - 2 # Leave the subr call in
+            else:
+                # In my understanding, this is a font bug.
+                # I.e., it has hint stems *after* path construction.
+                # I've seen this in widespread fonts.
+                # Best to ignore the hints I suppose...
+                pass
+                #assert 0
+        else:
+            hints.status = max(hints.status, subr_hints.status)
+            if hints.status != 2:
+                # Check from last_check, make sure we didn't have
+                # any operators.
+                for i in range(hints.last_checked, index - 1):
+                    if isinstance(cs.program[i], str):
+                        hints.status = 2
+                        break
+                hints.last_checked = index
+            if hints.status != 2:
+                # Decide where to chop off from
+                if subr_hints.status == 0:
+                    hints.last_hint = index
+                else:
+                    hints.last_hint = index - 2 # Leave the subr call in
+
+class _DesubroutinizingT2Decompiler(psCharStrings.SimpleT2Decompiler):
+
+    def __init__(self, localSubrs, globalSubrs):
+        psCharStrings.SimpleT2Decompiler.__init__(self,
+                                                  localSubrs,
+                                                  globalSubrs)
+
+    def execute(self, charString):
+        # Note: Currently we recompute _desubroutinized each time.
+        # This is more robust in some cases, but in other places we assume
+        # that each subroutine always expands to the same code, so
+        # maybe it doesn't matter. To speed up we can just not
+        # recompute _desubroutinized if it's there. For now I just
+        # double-check that it desubroutinized to the same thing.
+        old_desubroutinized = charString._desubroutinized if hasattr(charString, '_desubroutinized') else None
+
+        charString._patches = []
+        psCharStrings.SimpleT2Decompiler.execute(self, charString)
+        desubroutinized = charString.program[:]
+        for idx,expansion in reversed (charString._patches):
+            assert idx >= 2
+            assert desubroutinized[idx - 1] in ['callsubr', 'callgsubr'], desubroutinized[idx - 1]
+            assert type(desubroutinized[idx - 2]) == int
+            if expansion[-1] == 'return':
+                expansion = expansion[:-1]
+            desubroutinized[idx-2:idx] = expansion
+        if 'endchar' in desubroutinized:
+            # Cut off after first endchar
+            desubroutinized = desubroutinized[:desubroutinized.index('endchar') + 1]
+        else:
+            if not len(desubroutinized) or desubroutinized[-1] != 'return':
+                desubroutinized.append('return')
+
+        charString._desubroutinized = desubroutinized
+        del charString._patches
+
+        if old_desubroutinized:
+            assert desubroutinized == old_desubroutinized
+
+    def op_callsubr(self, index):
+        subr = self.localSubrs[self.operandStack[-1]+self.localBias]
+        psCharStrings.SimpleT2Decompiler.op_callsubr(self, index)
+        self.processSubr(index, subr)
+
+    def op_callgsubr(self, index):
+        subr = self.globalSubrs[self.operandStack[-1]+self.globalBias]
+        psCharStrings.SimpleT2Decompiler.op_callgsubr(self, index)
+        self.processSubr(index, subr)
+
+    def processSubr(self, index, subr):
+        cs = self.callingStack[-1]
+        cs._patches.append((index, subr._desubroutinized))
+
+
+@_add_method(ttLib.getTableClass('CFF '))
+def prune_post_subset(self, options):
+    cff = self.cff
+    for fontname in cff.keys():
+        font = cff[fontname]
+        cs = font.CharStrings
+
+        # Drop unused FontDictionaries
+        if hasattr(font, "FDSelect"):
+            sel = font.FDSelect
+            indices = _uniq_sort(sel.gidArray)
+            sel.gidArray = [indices.index (ss) for ss in sel.gidArray]
+            arr = font.FDArray
+            arr.items = [arr[i] for i in indices]
+            del arr.file, arr.offsets
+
+        # Desubroutinize if asked for
+        if options.desubroutinize:
+            for g in font.charset:
+                c,sel = cs.getItemAndSelector(g)
+                c.decompile()
+                subrs = getattr(c.private, "Subrs", [])
+                decompiler = _DesubroutinizingT2Decompiler(subrs, c.globalSubrs)
+                decompiler.execute(c)
+                c.program = c._desubroutinized
+
+        # Drop hints if not needed
+        if not options.hinting:
+
+            # This can be tricky, but doesn't have to. What we do is:
+            #
+            # - Run all used glyph charstrings and recurse into subroutines,
+            # - For each charstring (including subroutines), if it has any
+            #   of the hint stem operators, we mark it as such.
+            #   Upon returning, for each charstring we note all the
+            #   subroutine calls it makes that (recursively) contain a stem,
+            # - Dropping hinting then consists of the following two ops:
+            #     * Drop the piece of the program in each charstring before the
+            #         last call to a stem op or a stem-calling subroutine,
+            #     * Drop all hintmask operations.
+            # - It's trickier... A hintmask right after hints and a few numbers
+            #     will act as an implicit vstemhm. As such, we track whether
+            #     we have seen any non-hint operators so far and do the right
+            #     thing, recursively... Good luck understanding that :(
+            css = set()
+            for g in font.charset:
+                c,sel = cs.getItemAndSelector(g)
+                c.decompile()
+                subrs = getattr(c.private, "Subrs", [])
+                decompiler = _DehintingT2Decompiler(css, subrs, c.globalSubrs)
+                decompiler.execute(c)
+            for charstring in css:
+                charstring.drop_hints()
+            del css
+
+            # Drop font-wide hinting values
+            all_privs = []
+            if hasattr(font, 'FDSelect'):
+                all_privs.extend(fd.Private for fd in font.FDArray)
+            else:
+                all_privs.append(font.Private)
+            for priv in all_privs:
+                for k in ['BlueValues', 'OtherBlues',
+                          'FamilyBlues', 'FamilyOtherBlues',
+                          'BlueScale', 'BlueShift', 'BlueFuzz',
+                          'StemSnapH', 'StemSnapV', 'StdHW', 'StdVW']:
+                    if hasattr(priv, k):
+                        setattr(priv, k, None)
+
+        # Renumber subroutines to remove unused ones
+
+        # Mark all used subroutines
+        for g in font.charset:
+            c,sel = cs.getItemAndSelector(g)
+            subrs = getattr(c.private, "Subrs", [])
+            decompiler = _MarkingT2Decompiler(subrs, c.globalSubrs)
+            decompiler.execute(c)
+
+        all_subrs = [font.GlobalSubrs]
+        if hasattr(font, 'FDSelect'):
+            all_subrs.extend(fd.Private.Subrs for fd in font.FDArray if hasattr(fd.Private, 'Subrs') and fd.Private.Subrs)
+        elif hasattr(font.Private, 'Subrs') and font.Private.Subrs:
+            all_subrs.append(font.Private.Subrs)
+
+        subrs = set(subrs) # Remove duplicates
+
+        # Prepare
+        for subrs in all_subrs:
+            if not hasattr(subrs, '_used'):
+                subrs._used = set()
+            subrs._used = _uniq_sort(subrs._used)
+            subrs._old_bias = psCharStrings.calcSubrBias(subrs)
+            subrs._new_bias = psCharStrings.calcSubrBias(subrs._used)
+
+        # Renumber glyph charstrings
+        for g in font.charset:
+            c,sel = cs.getItemAndSelector(g)
+            subrs = getattr(c.private, "Subrs", [])
+            c.subset_subroutines (subrs, font.GlobalSubrs)
+
+        # Renumber subroutines themselves
+        for subrs in all_subrs:
+            if subrs == font.GlobalSubrs:
+                if not hasattr(font, 'FDSelect') and hasattr(font.Private, 'Subrs'):
+                    local_subrs = font.Private.Subrs
+                else:
+                    local_subrs = []
+            else:
+                local_subrs = subrs
+
+            subrs.items = [subrs.items[i] for i in subrs._used]
+            del subrs.file
+            if hasattr(subrs, 'offsets'):
+                del subrs.offsets
+
+            for subr in subrs.items:
+                subr.subset_subroutines (local_subrs, font.GlobalSubrs)
+
+        # Cleanup
+        for subrs in all_subrs:
+            del subrs._used, subrs._old_bias, subrs._new_bias
+
+    return True
+
+@_add_method(ttLib.getTableClass('cmap'))
+def closure_glyphs(self, s):
+    tables = [t for t in self.tables if t.isUnicode()]
+
+    # Close glyphs
+    for table in tables:
+        if table.format == 14:
+            for cmap in table.uvsDict.values():
+                glyphs = {g for u,g in cmap if u in s.unicodes_requested}
+                if None in glyphs:
+                    glyphs.remove(None)
+                s.glyphs.update(glyphs)
+        else:
+            cmap = table.cmap
+            intersection = s.unicodes_requested.intersection(cmap.keys())
+            s.glyphs.update(cmap[u] for u in intersection)
+
+    # Calculate unicodes_missing
+    s.unicodes_missing = s.unicodes_requested.copy()
+    for table in tables:
+        s.unicodes_missing.difference_update(table.cmap)
+
+@_add_method(ttLib.getTableClass('cmap'))
+def prune_pre_subset(self, options):
+    if not options.legacy_cmap:
+        # Drop non-Unicode / non-Symbol cmaps
+        self.tables = [t for t in self.tables if t.isUnicode() or t.isSymbol()]
+    if not options.symbol_cmap:
+        self.tables = [t for t in self.tables if not t.isSymbol()]
+    # TODO(behdad) Only keep one subtable?
+    # For now, drop format=0 which can't be subset_glyphs easily?
+    self.tables = [t for t in self.tables if t.format != 0]
+    self.numSubTables = len(self.tables)
+    return True # Required table
+
+@_add_method(ttLib.getTableClass('cmap'))
+def subset_glyphs(self, s):
+    s.glyphs = None # We use s.glyphs_requested and s.unicodes_requested only
+    for t in self.tables:
+        if t.format == 14:
+            # TODO(behdad) We drop all the default-UVS mappings
+            # for glyphs_requested.  So it's the caller's responsibility to make
+            # sure those are included.
+            t.uvsDict = {v:[(u,g) for u,g in l
+                                  if g in s.glyphs_requested or u in s.unicodes_requested]
+                         for v,l in t.uvsDict.items()}
+            t.uvsDict = {v:l for v,l in t.uvsDict.items() if l}
+        elif t.isUnicode():
+            t.cmap = {u:g for u,g in t.cmap.items()
+                          if g in s.glyphs_requested or u in s.unicodes_requested}
+        else:
+            t.cmap = {u:g for u,g in t.cmap.items()
+                          if g in s.glyphs_requested}
+    self.tables = [t for t in self.tables
+                   if (t.cmap if t.format != 14 else t.uvsDict)]
+    self.numSubTables = len(self.tables)
+    # TODO(behdad) Convert formats when needed.
+    # In particular, if we have a format=12 without non-BMP
+    # characters, either drop format=12 one or convert it
+    # to format=4 if there's not one.
+    return True # Required table
+
+@_add_method(ttLib.getTableClass('DSIG'))
+def prune_pre_subset(self, options):
+    # Drop all signatures since they will be invalid
+    self.usNumSigs = 0
+    self.signatureRecords = []
+    return True
+
+@_add_method(ttLib.getTableClass('maxp'))
+def prune_pre_subset(self, options):
+    if not options.hinting:
+        if self.tableVersion == 0x00010000:
+            self.maxZones = 1
+            self.maxTwilightPoints = 0
+            self.maxFunctionDefs = 0
+            self.maxInstructionDefs = 0
+            self.maxStackElements = 0
+            self.maxSizeOfInstructions = 0
+    return True
+
+@_add_method(ttLib.getTableClass('name'))
+def prune_pre_subset(self, options):
+    if '*' not in options.name_IDs:
+        self.names = [n for n in self.names if n.nameID in options.name_IDs]
+    if not options.name_legacy:
+        # TODO(behdad) Sometimes (eg Apple Color Emoji) there's only a macroman
+        # entry for Latin and no Unicode names.
+        self.names = [n for n in self.names if n.isUnicode()]
+    # TODO(behdad) Option to keep only one platform's
+    if '*' not in options.name_languages:
+        # TODO(behdad) This is Windows-platform specific!
+        self.names = [n for n in self.names
+                      if n.langID in options.name_languages]
+    if options.obfuscate_names:
+        namerecs = []
+        for n in self.names:
+            if n.nameID in [1, 4]:
+                n.string = ".\x7f".encode('utf_16_be') if n.isUnicode() else ".\x7f"
+            elif n.nameID in [2, 6]:
+                n.string = "\x7f".encode('utf_16_be') if n.isUnicode() else "\x7f"
+            elif n.nameID == 3:
+                n.string = ""
+            elif n.nameID in [16, 17, 18]:
+                continue
+            namerecs.append(n)
+        self.names = namerecs
+    return True    # Required table
+
+
+# TODO(behdad) OS/2 ulUnicodeRange / ulCodePageRange?
+# TODO(behdad) Drop AAT tables.
+# TODO(behdad) Drop unneeded GSUB/GPOS Script/LangSys entries.
+# TODO(behdad) Drop empty GSUB/GPOS, and GDEF if no GSUB/GPOS left
+# TODO(behdad) Drop GDEF subitems if unused by lookups
+# TODO(behdad) Avoid recursing too much (in GSUB/GPOS and in CFF)
+# TODO(behdad) Text direction considerations.
+# TODO(behdad) Text script / language considerations.
+# TODO(behdad) Optionally drop 'kern' table if GPOS available
+# TODO(behdad) Implement --unicode='*' to choose all cmap'ed
+# TODO(behdad) Drop old-spec Indic scripts
+
+
+class Options(object):
+
+    class OptionError(Exception): pass
+    class UnknownOptionError(OptionError): pass
+
+    _drop_tables_default = ['BASE', 'JSTF', 'DSIG', 'EBDT', 'EBLC',
+                            'EBSC', 'SVG ', 'PCLT', 'LTSH']
+    _drop_tables_default += ['Feat', 'Glat', 'Gloc', 'Silf', 'Sill']  # Graphite
+    _drop_tables_default += ['CBLC', 'CBDT', 'sbix', 'COLR', 'CPAL']  # Color
+    _no_subset_tables_default = ['gasp', 'head', 'hhea', 'maxp',
+                                 'vhea', 'OS/2', 'loca', 'name', 'cvt ',
+                                 'fpgm', 'prep', 'VDMX', 'DSIG']
+    _hinting_tables_default = ['cvt ', 'fpgm', 'prep', 'hdmx', 'VDMX']
+
+    # Based on HarfBuzz shapers
+    _layout_features_groups = {
+        # Default shaper
+        'common': ['ccmp', 'liga', 'locl', 'mark', 'mkmk', 'rlig'],
+        'horizontal': ['calt', 'clig', 'curs', 'kern', 'rclt'],
+        'vertical': ['valt', 'vert', 'vkrn', 'vpal', 'vrt2'],
+        'ltr': ['ltra', 'ltrm'],
+        'rtl': ['rtla', 'rtlm'],
+        # Complex shapers
+        'arabic': ['init', 'medi', 'fina', 'isol', 'med2', 'fin2', 'fin3',
+                   'cswh', 'mset'],
+        'hangul': ['ljmo', 'vjmo', 'tjmo'],
+        'tibetan': ['abvs', 'blws', 'abvm', 'blwm'],
+        'indic': ['nukt', 'akhn', 'rphf', 'rkrf', 'pref', 'blwf', 'half',
+                  'abvf', 'pstf', 'cfar', 'vatu', 'cjct', 'init', 'pres',
+                  'abvs', 'blws', 'psts', 'haln', 'dist', 'abvm', 'blwm'],
+    }
+    _layout_features_default = _uniq_sort(sum(
+            iter(_layout_features_groups.values()), []))
+
+    drop_tables = _drop_tables_default
+    no_subset_tables = _no_subset_tables_default
+    hinting_tables = _hinting_tables_default
+    legacy_kern = False    # drop 'kern' table if GPOS available
+    layout_features = _layout_features_default
+    ignore_missing_glyphs = False
+    ignore_missing_unicodes = True
+    hinting = True
+    glyph_names = False
+    legacy_cmap = False
+    symbol_cmap = False
+    name_IDs = [1, 2]    # Family and Style
+    name_legacy = False
+    name_languages = [0x0409]    # English
+    obfuscate_names = False    # to make webfont unusable as a system font
+    notdef_glyph = True # gid0 for TrueType / .notdef for CFF
+    notdef_outline = False # No need for notdef to have an outline really
+    recommended_glyphs = False    # gid1, gid2, gid3 for TrueType
+    recalc_bounds = False # Recalculate font bounding boxes
+    recalc_timestamp = False # Recalculate font modified timestamp
+    canonical_order = False # Order tables as recommended
+    flavor = None  # May be 'woff' or 'woff2'
+    desubroutinize = False # Desubroutinize CFF CharStrings
+
+    def __init__(self, **kwargs):
+        self.set(**kwargs)
+
+    def set(self, **kwargs):
+        for k,v in kwargs.items():
+            if not hasattr(self, k):
+                raise self.UnknownOptionError("Unknown option '%s'" % k)
+            setattr(self, k, v)
+
+    def parse_opts(self, argv, ignore_unknown=False):
+        ret = []
+        for a in argv:
+            orig_a = a
+            if not a.startswith('--'):
+                ret.append(a)
+                continue
+            a = a[2:]
+            i = a.find('=')
+            op = '='
+            if i == -1:
+                if a.startswith("no-"):
+                    k = a[3:]
+                    v = False
+                else:
+                    k = a
+                    v = True
+                if k.endswith("?"):
+                    k = k[:-1]
+                    v = '?'
+            else:
+                k = a[:i]
+                if k[-1] in "-+":
+                    op = k[-1]+'='    # Op is '-=' or '+=' now.
+                    k = k[:-1]
+                v = a[i+1:]
+            ok = k
+            k = k.replace('-', '_')
+            if not hasattr(self, k):
+                if ignore_unknown is True or ok in ignore_unknown:
+                    ret.append(orig_a)
+                    continue
+                else:
+                    raise self.UnknownOptionError("Unknown option '%s'" % a)
+
+            ov = getattr(self, k)
+            if v == '?':
+                    print("Current setting for '%s' is: %s" % (ok, ov))
+                    continue
+            if isinstance(ov, bool):
+                v = bool(v)
+            elif isinstance(ov, int):
+                v = int(v)
+            elif isinstance(ov, str):
+                v = str(v) # redundant
+            elif isinstance(ov, list):
+                if isinstance(v, bool):
+                    raise self.OptionError("Option '%s' requires values to be specified using '='" % a)
+                vv = v.replace(',', ' ').split()
+                if vv == ['']:
+                    vv = []
+                vv = [int(x, 0) if len(x) and x[0] in "0123456789" else x for x in vv]
+                if op == '=':
+                    v = vv
+                elif op == '+=':
+                    v = ov
+                    v.extend(vv)
+                elif op == '-=':
+                    v = ov
+                    for x in vv:
+                        if x in v:
+                            v.remove(x)
+                else:
+                    assert False
+
+            setattr(self, k, v)
+
+        return ret
+
+
+class Subsetter(object):
+
+    class SubsettingError(Exception): pass
+    class MissingGlyphsSubsettingError(SubsettingError): pass
+    class MissingUnicodesSubsettingError(SubsettingError): pass
+
+    def __init__(self, options=None, log=None):
+
+        if not log:
+            log = Logger()
+        if not options:
+            options = Options()
+
+        self.options = options
+        self.log = log
+        self.unicodes_requested = set()
+        self.glyph_names_requested = set()
+        self.glyph_ids_requested = set()
+
+    def populate(self, glyphs=[], gids=[], unicodes=[], text=""):
+        self.unicodes_requested.update(unicodes)
+        if isinstance(text, bytes):
+            text = text.decode("utf_8")
+        for u in text:
+            self.unicodes_requested.add(ord(u))
+        self.glyph_names_requested.update(glyphs)
+        self.glyph_ids_requested.update(gids)
+
+    def _prune_pre_subset(self, font):
+
+        for tag in font.keys():
+            if tag == 'GlyphOrder': continue
+
+            if(tag in self.options.drop_tables or
+                 (tag in self.options.hinting_tables and not self.options.hinting) or
+                 (tag == 'kern' and (not self.options.legacy_kern and 'GPOS' in font))):
+                self.log(tag, "dropped")
+                del font[tag]
+                continue
+
+            clazz = ttLib.getTableClass(tag)
+
+            if hasattr(clazz, 'prune_pre_subset'):
+                table = font[tag]
+                self.log.lapse("load '%s'" % tag)
+                retain = table.prune_pre_subset(self.options)
+                self.log.lapse("prune '%s'" % tag)
+                if not retain:
+                    self.log(tag, "pruned to empty; dropped")
+                    del font[tag]
+                    continue
+                else:
+                    self.log(tag, "pruned")
+
+    def _closure_glyphs(self, font):
+
+        realGlyphs = set(font.getGlyphOrder())
+        glyph_order = font.getGlyphOrder()
+
+        self.glyphs_requested = set()
+        self.glyphs_requested.update(self.glyph_names_requested)
+        self.glyphs_requested.update(glyph_order[i]
+                                     for i in self.glyph_ids_requested
+                                     if i < len(glyph_order))
+
+        self.glyphs_missing = set()
+        self.glyphs_missing.update(self.glyphs_requested.difference(realGlyphs))
+        self.glyphs_missing.update(i for i in self.glyph_ids_requested
+                                   if i >= len(glyph_order))
+        if self.glyphs_missing:
+            self.log("Missing requested glyphs: %s" % self.glyphs_missing)
+            if not self.options.ignore_missing_glyphs:
+                raise self.MissingGlyphsSubsettingError(self.glyphs_missing)
+
+        self.glyphs = self.glyphs_requested.copy()
+
+        self.unicodes_missing = set()
+        if 'cmap' in font:
+            font['cmap'].closure_glyphs(self)
+            self.glyphs.intersection_update(realGlyphs)
+            self.log.lapse("close glyph list over 'cmap'")
+        self.glyphs_cmaped = frozenset(self.glyphs)
+        if self.unicodes_missing:
+            missing = ["U+%04X" % u for u in self.unicodes_missing]
+            self.log("Missing glyphs for requested Unicodes: %s" % missing)
+            if not self.options.ignore_missing_unicodes:
+                raise self.MissingUnicodesSubsettingError(missing)
+            del missing
+
+        if self.options.notdef_glyph:
+            if 'glyf' in font:
+                self.glyphs.add(font.getGlyphName(0))
+                self.log("Added gid0 to subset")
+            else:
+                self.glyphs.add('.notdef')
+                self.log("Added .notdef to subset")
+        if self.options.recommended_glyphs:
+            if 'glyf' in font:
+                for i in range(min(4, len(font.getGlyphOrder()))):
+                    self.glyphs.add(font.getGlyphName(i))
+                self.log("Added first four glyphs to subset")
+
+        if 'GSUB' in font:
+            self.log("Closing glyph list over 'GSUB': %d glyphs before" %
+                     len(self.glyphs))
+            self.log.glyphs(self.glyphs, font=font)
+            font['GSUB'].closure_glyphs(self)
+            self.glyphs.intersection_update(realGlyphs)
+            self.log("Closed glyph list over 'GSUB': %d glyphs after" %
+                     len(self.glyphs))
+            self.log.glyphs(self.glyphs, font=font)
+            self.log.lapse("close glyph list over 'GSUB'")
+        self.glyphs_gsubed = frozenset(self.glyphs)
+
+        if 'glyf' in font:
+            self.log("Closing glyph list over 'glyf': %d glyphs before" %
+                     len(self.glyphs))
+            self.log.glyphs(self.glyphs, font=font)
+            font['glyf'].closure_glyphs(self)
+            self.glyphs.intersection_update(realGlyphs)
+            self.log("Closed glyph list over 'glyf': %d glyphs after" %
+                     len(self.glyphs))
+            self.log.glyphs(self.glyphs, font=font)
+            self.log.lapse("close glyph list over 'glyf'")
+        self.glyphs_glyfed = frozenset(self.glyphs)
+
+        self.glyphs_all = frozenset(self.glyphs)
+
+        self.log("Retaining %d glyphs: " % len(self.glyphs_all))
+
+        del self.glyphs
+
+    def _subset_glyphs(self, font):
+        for tag in font.keys():
+            if tag == 'GlyphOrder': continue
+            clazz = ttLib.getTableClass(tag)
+
+            if tag in self.options.no_subset_tables:
+                self.log(tag, "subsetting not needed")
+            elif hasattr(clazz, 'subset_glyphs'):
+                table = font[tag]
+                self.glyphs = self.glyphs_all
+                retain = table.subset_glyphs(self)
+                del self.glyphs
+                self.log.lapse("subset '%s'" % tag)
+                if not retain:
+                    self.log(tag, "subsetted to empty; dropped")
+                    del font[tag]
+                else:
+                    self.log(tag, "subsetted")
+            else:
+                self.log(tag, "NOT subset; don't know how to subset; dropped")
+                del font[tag]
+
+        glyphOrder = font.getGlyphOrder()
+        glyphOrder = [g for g in glyphOrder if g in self.glyphs_all]
+        font.setGlyphOrder(glyphOrder)
+        font._buildReverseGlyphOrderDict()
+        self.log.lapse("subset GlyphOrder")
+
+    def _prune_post_subset(self, font):
+        for tag in font.keys():
+            if tag == 'GlyphOrder': continue
+            clazz = ttLib.getTableClass(tag)
+            if hasattr(clazz, 'prune_post_subset'):
+                table = font[tag]
+                retain = table.prune_post_subset(self.options)
+                self.log.lapse("prune '%s'" % tag)
+                if not retain:
+                    self.log(tag, "pruned to empty; dropped")
+                    del font[tag]
+                else:
+                    self.log(tag, "pruned")
+
+    def subset(self, font):
+
+        self._prune_pre_subset(font)
+        self._closure_glyphs(font)
+        self._subset_glyphs(font)
+        self._prune_post_subset(font)
+
+
+class Logger(object):
+
+    def __init__(self, verbose=False, xml=False, timing=False):
+        self.verbose = verbose
+        self.xml = xml
+        self.timing = timing
+        self.last_time = self.start_time = time.time()
+
+    def parse_opts(self, argv):
+        argv = argv[:]
+        for v in ['verbose', 'xml', 'timing']:
+            if "--"+v in argv:
+                setattr(self, v, True)
+                argv.remove("--"+v)
+        return argv
+
+    def __call__(self, *things):
+        if not self.verbose:
+            return
+        print(' '.join(str(x) for x in things))
+
+    def lapse(self, *things):
+        if not self.timing:
+            return
+        new_time = time.time()
+        print("Took %0.3fs to %s" %(new_time - self.last_time,
+                                    ' '.join(str(x) for x in things)))
+        self.last_time = new_time
+
+    def glyphs(self, glyphs, font=None):
+        if not self.verbose:
+            return
+        self("Glyph names:", sorted(glyphs))
+        if font:
+            reverseGlyphMap = font.getReverseGlyphMap()
+            self("Glyph IDs:    ", sorted(reverseGlyphMap[g] for g in glyphs))
+
+    def font(self, font, file=sys.stdout):
+        if not self.xml:
+            return
+        from fontTools.misc import xmlWriter
+        writer = xmlWriter.XMLWriter(file)
+        for tag in font.keys():
+            writer.begintag(tag)
+            writer.newline()
+            font[tag].toXML(writer, font)
+            writer.endtag(tag)
+            writer.newline()
+
+
+def load_font(fontFile,
+              options,
+              allowVID=False,
+              checkChecksums=False,
+              dontLoadGlyphNames=False,
+              lazy=True):
+
+    font = ttLib.TTFont(fontFile,
+                        allowVID=allowVID,
+                        checkChecksums=checkChecksums,
+                        recalcBBoxes=options.recalc_bounds,
+                        recalcTimestamp=options.recalc_timestamp,
+                        lazy=lazy)
+
+    # Hack:
+    #
+    # If we don't need glyph names, change 'post' class to not try to
+    # load them.    It avoid lots of headache with broken fonts as well
+    # as loading time.
+    #
+    # Ideally ttLib should provide a way to ask it to skip loading
+    # glyph names.    But it currently doesn't provide such a thing.
+    #
+    if dontLoadGlyphNames:
+        post = ttLib.getTableClass('post')
+        saved = post.decode_format_2_0
+        post.decode_format_2_0 = post.decode_format_3_0
+        f = font['post']
+        if f.formatType == 2.0:
+            f.formatType = 3.0
+        post.decode_format_2_0 = saved
+
+    return font
+
+def save_font(font, outfile, options):
+    if options.flavor and not hasattr(font, 'flavor'):
+        raise Exception("fonttools version does not support flavors.")
+    font.flavor = options.flavor
+    font.save(outfile, reorderTables=options.canonical_order)
+
+def parse_unicodes(s):
+    import re
+    s = re.sub (r"0[xX]", " ", s)
+    s = re.sub (r"[<+>,;&#\\xXuU\n    ]", " ", s)
+    l = []
+    for item in s.split():
+        fields = item.split('-')
+        if len(fields) == 1:
+            l.append(int(item, 16))
+        else:
+            start,end = fields
+            l.extend(range(int(start, 16), int(end, 16)+1))
+    return l
+
+def parse_gids(s):
+    l = []
+    for item in s.replace(',', ' ').split():
+        fields = item.split('-')
+        if len(fields) == 1:
+            l.append(int(fields[0]))
+        else:
+            l.extend(range(int(fields[0]), int(fields[1])+1))
+    return l
+
+def parse_glyphs(s):
+    return s.replace(',', ' ').split()
+
+def main(args=None):
+
+    if args is None:
+        args = sys.argv[1:]
+
+    if '--help' in args:
+        print(__doc__)
+        sys.exit(0)
+
+    log = Logger()
+    args = log.parse_opts(args)
+
+    options = Options()
+    args = options.parse_opts(args,
+        ignore_unknown=['gids', 'gids-file',
+                        'glyphs', 'glyphs-file',
+                        'text', 'text-file',
+                        'unicodes', 'unicodes-file',
+                        'output-file'])
+
+    if len(args) < 2:
+        print("usage:", __usage__, file=sys.stderr)
+        print("Try pyftsubset --help for more information.", file=sys.stderr)
+        sys.exit(1)
+
+    fontfile = args[0]
+    args = args[1:]
+
+    subsetter = Subsetter(options=options, log=log)
+    outfile = fontfile + '.subset'
+    glyphs = []
+    gids = []
+    unicodes = []
+    wildcard_glyphs = False
+    wildcard_unicodes = False
+    text = ""
+    for g in args:
+        if g == '*':
+            wildcard_glyphs = True
+            continue
+        if g.startswith('--output-file='):
+            outfile = g[14:]
+            continue
+        if g.startswith('--text='):
+            text += g[7:]
+            continue
+        if g.startswith('--text-file='):
+            text += open(g[12:]).read().replace('\n', '')
+            continue
+        if g.startswith('--unicodes='):
+            if g[11:] == '*':
+                wildcard_unicodes = True
+            else:
+                unicodes.extend(parse_unicodes(g[11:]))
+            continue
+        if g.startswith('--unicodes-file='):
+            for line in open(g[16:]).readlines():
+                unicodes.extend(parse_unicodes(line.split('#')[0]))
+            continue
+        if g.startswith('--gids='):
+            gids.extend(parse_gids(g[7:]))
+            continue
+        if g.startswith('--gids-file='):
+            for line in open(g[12:]).readlines():
+                gids.extend(parse_gids(line.split('#')[0]))
+            continue
+        if g.startswith('--glyphs='):
+            if g[9:] == '*':
+                wildcard_glyphs = True
+            else:
+                glyphs.extend(parse_glyphs(g[9:]))
+            continue
+        if g.startswith('--glyphs-file='):
+            for line in open(g[14:]).readlines():
+                glyphs.extend(parse_glyphs(line.split('#')[0]))
+            continue
+        glyphs.append(g)
+
+    dontLoadGlyphNames = not options.glyph_names and not glyphs
+    font = load_font(fontfile, options, dontLoadGlyphNames=dontLoadGlyphNames)
+    log.lapse("load font")
+    if wildcard_glyphs:
+            glyphs.extend(font.getGlyphOrder())
+    if wildcard_unicodes:
+            for t in font['cmap'].tables:
+                if t.isUnicode():
+                    unicodes.extend(t.cmap.keys())
+    assert '' not in glyphs
+
+    log.lapse("compile glyph list")
+    log("Text: '%s'" % text)
+    log("Unicodes:", unicodes)
+    log("Glyphs:", glyphs)
+    log("Gids:", gids)
+
+    subsetter.populate(glyphs=glyphs, gids=gids, unicodes=unicodes, text=text)
+    subsetter.subset(font)
+
+    save_font (font, outfile, options)
+    log.lapse("compile and save font")
+
+    log.last_time = log.start_time
+    log.lapse("make one with everything(TOTAL TIME)")
+
+    if log.verbose:
+        import os
+        log("Input font:% 7d bytes: %s" % (os.path.getsize(fontfile), fontfile))
+        log("Subset font:% 7d bytes: %s" % (os.path.getsize(outfile), outfile))
+
+    log.font(font)
+
+    font.close()
+
+
+__all__ = [
+    'Options',
+    'Subsetter',
+    'Logger',
+    'load_font',
+    'save_font',
+    'parse_gids',
+    'parse_glyphs',
+    'parse_unicodes',
+    'main'
+]
+
+if __name__ == '__main__':
+    main()