mirror of
https://github.com/rsms/inter.git
synced 2024-11-25 07:45:18 +03:00
651 lines
18 KiB
Python
Executable File
651 lines
18 KiB
Python
Executable File
#!/usr/bin/env python
|
|
# encoding: utf8
|
|
from __future__ import print_function
|
|
import os
|
|
import sys
|
|
import argparse
|
|
import json
|
|
import plistlib
|
|
import re
|
|
from collections import OrderedDict
|
|
from textwrap import TextWrapper
|
|
from StringIO import StringIO
|
|
from ConfigParser import RawConfigParser
|
|
from fontTools import ttLib
|
|
from robofab.objects.objectsRF import RFont, OpenFont
|
|
|
|
# from feaTools import parser as feaParser
|
|
# from feaTools.parser import parseFeatures
|
|
# from feaTools import FDKSyntaxFeatureWriter
|
|
# from fontbuild.features import updateFeature, compileFeatureRE
|
|
|
|
# Regex matching "default" glyph names, like "uni2043" and "u01C5"
|
|
uniNameRe = re.compile(r'^u(?:ni)[0-9A-F]{4,8}$')
|
|
|
|
|
|
def defaultGlyphName(uc):
|
|
return 'uni%04X' % uc
|
|
|
|
def defaultGlyphName2(uc):
|
|
return 'u%04X' % uc
|
|
|
|
|
|
def isDefaultGlyphName(name):
|
|
return True if uniNameRe.match(name) else False
|
|
|
|
|
|
def isDefaultGlyphNameForUnicode(name, uc):
|
|
return name == defaultGlyphName(uc) or name == defaultGlyphName2(uc)
|
|
|
|
|
|
def getFirstNonDefaultGlyphName(uc, names):
|
|
for name in names:
|
|
if not isDefaultGlyphNameForUnicode(name, uc):
|
|
return name
|
|
return None
|
|
|
|
|
|
def getTTGlyphList(font): # -> { 'Omega': [2126, ...], ... }
|
|
if isinstance(font, str):
|
|
font = ttLib.TTFont(font)
|
|
|
|
if not 'cmap' in font:
|
|
raise Exception('missing cmap table')
|
|
|
|
gl = {}
|
|
bestCodeSubTable = None
|
|
bestCodeSubTableFormat = 0
|
|
|
|
for st in font['cmap'].tables:
|
|
if st.platformID == 0: # 0=unicode, 1=mac, 2=(reserved), 3=microsoft
|
|
if st.format > bestCodeSubTableFormat:
|
|
bestCodeSubTable = st
|
|
bestCodeSubTableFormat = st.format
|
|
|
|
if bestCodeSubTable is not None:
|
|
for cp, glyphname in bestCodeSubTable.cmap.items():
|
|
if glyphname in gl:
|
|
gl[glyphname].append(cp)
|
|
else:
|
|
gl[glyphname] = [cp]
|
|
|
|
return gl, font
|
|
|
|
|
|
def getUFOGlyphList(font): # -> { 'Omega': [2126, ...], ... }
|
|
# Note: font.getCharacterMapping() returns {2126:['Omega', ...], ...}
|
|
gl = {}
|
|
for g in font:
|
|
ucv = g.unicodes
|
|
if len(ucv) > 0:
|
|
gl[g.name] = ucv
|
|
return gl
|
|
|
|
|
|
def appendNames(uc2names, extraUc2names, uc, name, isDestination):
|
|
if uc in uc2names:
|
|
names = uc2names[uc]
|
|
if name not in names:
|
|
names.append(name)
|
|
elif isDestination:
|
|
uc2names[uc] = [name]
|
|
else:
|
|
if uc in extraUc2names:
|
|
names = extraUc2names[uc]
|
|
if name not in names:
|
|
names.append(name)
|
|
else:
|
|
extraUc2names[uc] = [name]
|
|
|
|
|
|
def buildGlyphNames(dstFonts, srcFonts, glyphOrder, fallbackGlyphNames):
|
|
# fallbackGlyphNames: { 2126: 'Omega', ...}
|
|
uc2names = {} # { 2126: ['Omega', 'Omegagreek', ...], ...}
|
|
extraUc2names = {} # { 2126: ['Omega', 'Omegagreek', ...], ...}
|
|
# -- codepoints in Nth fonts, not found in first font
|
|
name2ucsv = [] # [ { 'Omega': [2126, ...] }, ... ] -- same order as fonts
|
|
|
|
fontIndex = 0
|
|
for font in dstFonts + srcFonts:
|
|
gl = None
|
|
if isinstance(font, RFont):
|
|
print('Inspecting', font.info.familyName, font.info.styleName)
|
|
gl = getUFOGlyphList(font)
|
|
else:
|
|
print('Inspecting', font)
|
|
gl, font = getTTGlyphList(font)
|
|
|
|
name2ucsv.append(gl)
|
|
|
|
isDestination = fontIndex < len(dstFonts)
|
|
|
|
for name, unicodes in gl.iteritems():
|
|
# if len(uc2names) > 100: break
|
|
for uc in unicodes:
|
|
appendNames(uc2names, extraUc2names, uc, name, isDestination)
|
|
if isDestination:
|
|
fallbackName = fallbackGlyphNames.get(uc)
|
|
if fallbackName is not None:
|
|
appendNames(uc2names, extraUc2names, uc, fallbackName, isDestination)
|
|
|
|
fontIndex += 1
|
|
|
|
# for name in glyphOrder:
|
|
# if len(name) > 7 and name.startswith('uni') and name.find('.') == -1 and name.find('_') == -1:
|
|
# try:
|
|
# print('name: %r, %r' % (name, name[3:]))
|
|
# uc = int(name[3:], 16)
|
|
# appendNames(uc2names, extraUc2names, uc, name, isDestination=True)
|
|
# except:
|
|
# print()
|
|
# pass
|
|
|
|
return uc2names, extraUc2names, name2ucsv
|
|
|
|
|
|
def renameStrings(listofstrs, newNames):
|
|
v = []
|
|
for s in listofstrs:
|
|
s2 = newNames.get(s)
|
|
if s2 is not None:
|
|
s = s2
|
|
v.append(s)
|
|
return v
|
|
|
|
|
|
def renameUFOLib(ufoPath, newNames, dryRun=False, print=print):
|
|
filename = os.path.join(ufoPath, 'lib.plist')
|
|
plist = plistlib.readPlist(filename)
|
|
|
|
glyphOrder = plist.get('public.glyphOrder')
|
|
if glyphOrder is not None:
|
|
plist['public.glyphOrder'] = renameStrings(glyphOrder, newNames)
|
|
|
|
roboSort = plist.get('com.typemytype.robofont.sort')
|
|
if roboSort is not None:
|
|
for entry in roboSort:
|
|
if isinstance(entry, dict) and entry.get('type') == 'glyphList':
|
|
asc = entry.get('ascending')
|
|
desc = entry.get('descending')
|
|
if asc is not None:
|
|
entry['ascending'] = renameStrings(asc, newNames)
|
|
if desc is not None:
|
|
entry['descending'] = renameStrings(desc, newNames)
|
|
|
|
print('Writing', filename)
|
|
if not dryRun:
|
|
plistlib.writePlist(plist, filename)
|
|
|
|
|
|
def renameUFOGroups(ufoPath, newNames, dryRun=False, print=print):
|
|
filename = os.path.join(ufoPath, 'groups.plist')
|
|
|
|
plist = None
|
|
try:
|
|
plist = plistlib.readPlist(filename)
|
|
except:
|
|
return
|
|
|
|
didChange = False
|
|
|
|
for groupName, glyphNames in plist.items():
|
|
for i in range(len(glyphNames)):
|
|
name = glyphNames[i]
|
|
if name in newNames:
|
|
didChange = True
|
|
glyphNames[i] = newNames[name]
|
|
|
|
if didChange:
|
|
print('Writing', filename)
|
|
if not dryRun:
|
|
plistlib.writePlist(plist, filename)
|
|
|
|
|
|
def renameUFOKerning(ufoPath, newNames, dryRun=False, print=print):
|
|
filename = os.path.join(ufoPath, 'kerning.plist')
|
|
|
|
plist = None
|
|
try:
|
|
plist = plistlib.readPlist(filename)
|
|
except:
|
|
return
|
|
|
|
didChange = False
|
|
|
|
newPlist = {}
|
|
for leftName, right in plist.items():
|
|
if leftName in newNames:
|
|
didChange = True
|
|
leftName = newNames[leftName]
|
|
newRight = {}
|
|
for rightName, kernValue in plist.items():
|
|
if rightName in newNames:
|
|
didChange = True
|
|
rightName = newNames[rightName]
|
|
newRight[rightName] = kernValue
|
|
newPlist[leftName] = right
|
|
|
|
if didChange:
|
|
print('Writing', filename)
|
|
if not dryRun:
|
|
plistlib.writePlist(newPlist, filename)
|
|
|
|
|
|
def subFeaName(m, newNames, state):
|
|
try:
|
|
int(m[3], 16)
|
|
except:
|
|
return m[0]
|
|
|
|
name = m[2]
|
|
|
|
if name in newNames:
|
|
# print('sub %r => %r' % (m[0], m[1] + newNames[name] + m[4]))
|
|
if name == 'uni0402':
|
|
print('sub %r => %r' % (m[0], m[1] + newNames[name] + m[4]))
|
|
state['didChange'] = True
|
|
return m[1] + newNames[name] + m[4]
|
|
|
|
return m[0]
|
|
|
|
|
|
FEA_TOK = 'tok'
|
|
FEA_SEP = 'sep'
|
|
FEA_END = 'end'
|
|
|
|
def feaTokenizer(feaText):
|
|
separators = set('; \t\r\n,[]\'"')
|
|
tokStartIndex = -1
|
|
sepStartIndex = -1
|
|
|
|
for i in xrange(len(feaText)):
|
|
ch = feaText[i]
|
|
if ch in separators:
|
|
if tokStartIndex != -1:
|
|
yield (FEA_TOK, feaText[tokStartIndex:i])
|
|
tokStartIndex = -1
|
|
if sepStartIndex == -1:
|
|
sepStartIndex = i
|
|
else:
|
|
if sepStartIndex != -1:
|
|
yield (FEA_SEP, feaText[sepStartIndex:i])
|
|
sepStartIndex = -1
|
|
if tokStartIndex == -1:
|
|
tokStartIndex = i
|
|
|
|
if sepStartIndex != -1 and tokStartIndex != -1:
|
|
yield (FEA_END, feaText[min(sepStartIndex, tokStartIndex):])
|
|
elif sepStartIndex != -1:
|
|
yield (FEA_END, feaText[sepStartIndex:])
|
|
elif tokStartIndex != -1:
|
|
yield (FEA_END, feaText[tokStartIndex:])
|
|
else:
|
|
yield (FEA_END, '')
|
|
|
|
|
|
def renameUFOFeatures(font, ufoPath, newNames, dryRun=False, print=print):
|
|
filename = os.path.join(ufoPath, 'features.fea')
|
|
|
|
feaText = ''
|
|
try:
|
|
with open(filename, 'r') as f:
|
|
feaText = f.read()
|
|
except:
|
|
return
|
|
|
|
didChange = False
|
|
feaText2 = ''
|
|
|
|
for t, v in feaTokenizer(feaText):
|
|
if t is FEA_TOK and len(v) > 6 and v.startswith('uni'):
|
|
if v in newNames:
|
|
# print('sub', v, newNames[v])
|
|
didChange = True
|
|
v = newNames[v]
|
|
feaText2 += v
|
|
|
|
feaText = feaText2
|
|
|
|
if didChange:
|
|
print('Writing', filename)
|
|
if not dryRun:
|
|
with open(filename, 'w') as f:
|
|
f.write(feaText)
|
|
print(
|
|
'Important: you need to manually verify that', filename, 'looks okay.',
|
|
'We did an optimistic update which is not perfect.'
|
|
)
|
|
|
|
# classes = feaParser.classDefinitionRE.findall(feaText)
|
|
# for precedingMark, className, classContent in classes:
|
|
# content = feaParser.classContentRE.findall(classContent)
|
|
# print('class', className, content)
|
|
|
|
# didChange = False
|
|
# content2 = []
|
|
# for name in content:
|
|
# if name in newNames:
|
|
# didChange = True
|
|
# content2.append(newNames[name])
|
|
# if didChange:
|
|
# print('content2', content2)
|
|
# feaText = feaParser.classDefinitionRE.sub('', feaText)
|
|
|
|
# featureTags = feaParser.feature_findAll_RE.findall(feaText)
|
|
# for precedingMark, featureTag in featureTags:
|
|
# print('feat', featureTag)
|
|
|
|
|
|
def renameUFODetails(font, ufoPath, newNames, dryRun=False, print=print):
|
|
renameUFOLib(ufoPath, newNames, dryRun, print)
|
|
renameUFOGroups(ufoPath, newNames, dryRun, print)
|
|
renameUFOKerning(ufoPath, newNames, dryRun, print)
|
|
renameUFOFeatures(font, ufoPath, newNames, dryRun, print)
|
|
|
|
|
|
def readLines(filename):
|
|
with open(filename, 'r') as f:
|
|
return f.read().strip().splitlines()
|
|
|
|
|
|
def readGlyphOrderFile(filename):
|
|
names = []
|
|
for line in readLines(filename):
|
|
line = line.lstrip()
|
|
if len(line) > 0 and line[0] != '#':
|
|
names.append(line)
|
|
return names
|
|
|
|
|
|
def renameGlyphOrderFile(filename, newNames, dryRun=False, print=print):
|
|
lines = []
|
|
didRename = False
|
|
for line in readLines(filename):
|
|
line = line.lstrip()
|
|
if len(line) > 0 and line[0] != '#':
|
|
newName = newNames.get(line)
|
|
if newName is not None:
|
|
didRename = True
|
|
line = newName
|
|
lines.append(line)
|
|
if didRename:
|
|
print('Writing', filename)
|
|
if not dryRun:
|
|
with open(filename, 'w') as f:
|
|
f.write('\n'.join(lines))
|
|
|
|
|
|
def parseGlyphComposition(composite):
|
|
c = composite.split("=")
|
|
d = c[1].split("/")
|
|
glyphName = d[0]
|
|
if len(d) == 1:
|
|
offset = [0, 0]
|
|
else:
|
|
offset = [int(i) for i in d[1].split(",")]
|
|
accentString = c[0]
|
|
accents = accentString.split("+")
|
|
baseName = accents.pop(0)
|
|
accentNames = [i.split(":") for i in accents]
|
|
return (glyphName, baseName, accentNames, offset)
|
|
|
|
|
|
def fmtGlyphComposition(glyphName, baseName, accentNames, offset):
|
|
# glyphName = 'uni03D3'
|
|
# baseName = 'uni03D2'
|
|
# accentNames = [['tonos', 'top'], ['acute', 'top']]
|
|
# offset = [100, 0]
|
|
# => "uni03D2+tonos:top+acute:top=uni03D3/100,0"
|
|
s = baseName
|
|
for accentNameTuple in accentNames:
|
|
s += '+' + accentNameTuple[0]
|
|
if len(accentNameTuple) > 1:
|
|
s += ':' + accentNameTuple[1]
|
|
s += '=' + glyphName
|
|
if offset[0] != 0 or offset[1] != 0:
|
|
s += '/%d,%d' % tuple(offset)
|
|
return s
|
|
|
|
|
|
def renameDiacriticsFile(filename, newNames, dryRun=False, print=print):
|
|
lines = []
|
|
didRename = False
|
|
for line in readLines(filename):
|
|
line = line.strip()
|
|
if len(line) > 0 and line[0] != '#':
|
|
glyphName, baseName, accentNames, offset = parseGlyphComposition(line)
|
|
|
|
# rename
|
|
glyphName = newNames.get(glyphName, glyphName)
|
|
baseName = newNames.get(baseName, baseName)
|
|
for accentTuple in accentNames:
|
|
accentTuple[0] = newNames.get(accentTuple[0], accentTuple[0])
|
|
|
|
line2 = fmtGlyphComposition(glyphName, baseName, accentNames, offset)
|
|
|
|
if line != line2:
|
|
line = line2
|
|
didRename = True
|
|
# print(line, '=>', line2)
|
|
|
|
lines.append(line)
|
|
|
|
if didRename:
|
|
print('Writing', filename)
|
|
if not dryRun:
|
|
with open(filename, 'w') as f:
|
|
f.write('\n'.join(lines))
|
|
|
|
|
|
def configFindResFile(config, basedir, name):
|
|
fn = os.path.join(basedir, config.get("res", name))
|
|
if not os.path.isfile(fn):
|
|
basedir = os.path.dirname(basedir)
|
|
fn = os.path.join(basedir, config.get("res", name))
|
|
if not os.path.isfile(fn):
|
|
fn = None
|
|
return fn
|
|
|
|
|
|
def renameConfigFile(config, filename, newNames, dryRun=False, print=print):
|
|
wrapper = TextWrapper()
|
|
wrapper.width = 80
|
|
wrapper.break_long_words = False
|
|
wrapper.break_on_hyphens = False
|
|
|
|
wrap = lambda names: '\n'.join(wrapper.wrap(' '.join(names)))
|
|
|
|
didRename = False
|
|
for propertyName, values in config.items('glyphs'):
|
|
glyphNames = values.split()
|
|
# print(propertyName, glyphNames)
|
|
propChanged = False
|
|
for name in glyphNames:
|
|
if name in newNames:
|
|
sectionChanged = True
|
|
if sectionChanged:
|
|
config.set('glyphs', propertyName, wrap(glyphNames)+'\n')
|
|
didRename = True
|
|
|
|
# config.set(section, option, value)
|
|
if didRename:
|
|
s = StringIO()
|
|
config.write(s)
|
|
s = s.getvalue()
|
|
s = re.sub(r'\n(\w+)\s+=\s*', '\n\\1: ', s, flags=re.M)
|
|
s = re.sub(r'((?:^|\n)\[[^\]]*\])', '\\1\n', s, flags=re.M)
|
|
s = re.sub(r'\n\t\n', '\n\n', s, flags=re.M)
|
|
s = s.strip() + '\n'
|
|
print('Writing', filename)
|
|
if not dryRun:
|
|
with open(filename, 'w') as f:
|
|
f.write(s)
|
|
|
|
|
|
def parseAGL(filename): # -> { 2126: 'Omega', ... }
|
|
m = {}
|
|
for line in readLines(filename):
|
|
# Omega;2126
|
|
# dalethatafpatah;05D3 05B2 # higher-level combinations; ignored
|
|
line = line.strip()
|
|
if len(line) > 0 and line[0] != '#':
|
|
name, uc = tuple([c.strip() for c in line.split(';')])
|
|
if uc.find(' ') == -1:
|
|
# it's a 1:1 mapping
|
|
m[int(uc, 16)] = name
|
|
return m
|
|
|
|
|
|
def main():
|
|
argparser = argparse.ArgumentParser(description='Enrich UFO glyphnames')
|
|
|
|
argparser.add_argument(
|
|
'-dry', dest='dryRun', action='store_const', const=True, default=False,
|
|
help='Do not modify anything, but instead just print what would happen.')
|
|
|
|
argparser.add_argument(
|
|
'-list-missing', dest='listMissing', action='store_const', const=True, default=False,
|
|
help='List glyphs with unicodes found in source files but missing in any of the target UFOs.')
|
|
|
|
argparser.add_argument(
|
|
'-list-unnamed', dest='listUnnamed', action='store_const', const=True, default=False,
|
|
help="List glyphs with unicodes in target UFOs that don't have symbolic names.")
|
|
|
|
argparser.add_argument(
|
|
'-backfill-agl', dest='backfillWithAgl', action='store_const', const=True, default=False,
|
|
help="Use glyphnames from Adobe Glyph List for any glyphs that no names in any of"+
|
|
" the input font files")
|
|
|
|
argparser.add_argument(
|
|
'-src', dest='srcFonts', metavar='<fontfile>', type=str, nargs='*',
|
|
help='TrueType, OpenType or UFO fonts to gather glyph info from. '+
|
|
'Names found in earlier-listed fonts are prioritized over later listings.')
|
|
|
|
argparser.add_argument(
|
|
'dstFonts', metavar='<ufofile>', type=str, nargs='+', help='UFO fonts to update')
|
|
|
|
args = argparser.parse_args()
|
|
|
|
# Load UFO fonts
|
|
dstFonts = []
|
|
dstFontPaths = {} # keyed by RFont object
|
|
srcDir = None
|
|
for fn in args.dstFonts:
|
|
fn = fn.rstrip('/')
|
|
font = OpenFont(fn)
|
|
dstFonts.append(font)
|
|
dstFontPaths[font] = fn
|
|
srcDir2 = os.path.dirname(fn)
|
|
if srcDir is None:
|
|
srcDir = srcDir2
|
|
elif srcDir != srcDir2:
|
|
raise Exception('All <ufofile>s must be rooted in same directory')
|
|
|
|
# load fontbuild configuration
|
|
config = RawConfigParser(dict_type=OrderedDict)
|
|
configFilename = os.path.join(srcDir, 'fontbuild.cfg')
|
|
config.read(configFilename)
|
|
glyphOrderFile = configFindResFile(config, srcDir, 'glyphorder')
|
|
diacriticsFile = configFindResFile(config, srcDir, 'diacriticfile')
|
|
glyphOrder = readGlyphOrderFile(glyphOrderFile)
|
|
|
|
fallbackGlyphNames = {} # { 2126: 'Omega', ... }
|
|
if args.backfillWithAgl:
|
|
fallbackGlyphNames = parseAGL(configFindResFile(config, srcDir, 'agl_glyphlistfile'))
|
|
|
|
# find glyph names
|
|
uc2names, extraUc2names, name2ucsv = buildGlyphNames(
|
|
dstFonts,
|
|
args.srcFonts,
|
|
glyphOrder,
|
|
fallbackGlyphNames
|
|
)
|
|
# Note: name2ucsv has same order as parameters to buildGlyphNames
|
|
|
|
if args.listMissing:
|
|
print('# Missing glyphs: (found in -src but not in any <ufofile>)')
|
|
for uc, names in extraUc2names.iteritems():
|
|
print('U+%04X\t%s' % (uc, ', '.join(names)))
|
|
return
|
|
|
|
elif args.listUnnamed:
|
|
print('# Unnamed glyphs:')
|
|
unnamed = set()
|
|
for name in glyphOrder:
|
|
if len(name) > 7 and name.startswith('uni'):
|
|
unnamed.add(name)
|
|
for gl in name2ucsv[:len(dstFonts)]:
|
|
for name, ucs in gl.iteritems():
|
|
for uc in ucs:
|
|
if isDefaultGlyphNameForUnicode(name, uc):
|
|
unnamed.add(name)
|
|
break
|
|
for name in unnamed:
|
|
print(name)
|
|
return
|
|
|
|
printDry = lambda *args: print(*args)
|
|
if args.dryRun:
|
|
printDry = lambda *args: print('[dry-run]', *args)
|
|
|
|
newNames = {}
|
|
renameGlyphsQueue = {} # keyed by RFont object
|
|
|
|
for font in dstFonts:
|
|
renameGlyphsQueue[font] = {}
|
|
|
|
for uc, names in uc2names.iteritems():
|
|
if len(names) < 2:
|
|
continue
|
|
dstGlyphName = names[0]
|
|
if isDefaultGlyphNameForUnicode(dstGlyphName, uc):
|
|
newGlyphName = getFirstNonDefaultGlyphName(uc, names[1:])
|
|
# if newGlyphName is None:
|
|
# # if we found no symbolic name, check in fallback list
|
|
# newGlyphName = fallbackGlyphNames.get(uc)
|
|
# if newGlyphName is not None:
|
|
# printDry('Using fallback %s' % newGlyphName)
|
|
if newGlyphName is not None:
|
|
printDry('Rename %s -> %s' % (dstGlyphName, newGlyphName))
|
|
for font in dstFonts:
|
|
if dstGlyphName in font:
|
|
renameGlyphsQueue[font][dstGlyphName] = newGlyphName
|
|
newNames[dstGlyphName] = newGlyphName
|
|
|
|
if len(newNames) == 0:
|
|
printDry('No changes')
|
|
return
|
|
|
|
# rename component instances
|
|
for font in dstFonts:
|
|
componentMap = font.getReverseComponentMapping()
|
|
for currName, newName in renameGlyphsQueue[font].iteritems():
|
|
for depName in componentMap.get(currName, []):
|
|
depG = font[depName]
|
|
for c in depG.components:
|
|
if c.baseGlyph == currName:
|
|
c.baseGlyph = newName
|
|
c.setChanged()
|
|
|
|
# rename glyphs
|
|
for font in dstFonts:
|
|
for currName, newName in renameGlyphsQueue[font].iteritems():
|
|
font[currName].name = newName
|
|
|
|
# save fonts and update font data
|
|
for font in dstFonts:
|
|
fontPath = dstFontPaths[font]
|
|
printDry('Saving %d glyphs in %s' % (len(newNames), fontPath))
|
|
if not args.dryRun:
|
|
font.save()
|
|
renameUFODetails(font, fontPath, newNames, dryRun=args.dryRun, print=printDry)
|
|
|
|
# update resource files
|
|
renameGlyphOrderFile(glyphOrderFile, newNames, dryRun=args.dryRun, print=printDry)
|
|
renameDiacriticsFile(diacriticsFile, newNames, dryRun=args.dryRun, print=printDry)
|
|
renameConfigFile(config, configFilename, newNames, dryRun=args.dryRun, print=printDry)
|
|
|
|
|
|
if __name__ == '__main__':
|
|
main()
|