"""ttLib/sfnt.py -- low-level module to deal with the sfnt file format. Defines two public classes: SFNTReader SFNTWriter (Normally you don't have to use these classes explicitly; they are used automatically by ttLib.TTFont.) The reading and writing of sfnt files is separated in two distinct classes, since whenever to number of tables changes or whenever a table's length chages you need to rewrite the whole file anyway. """ import sys import struct, sstruct import numpy import os class SFNTReader: def __init__(self, file, checkChecksums=1, fontNumber=-1): self.file = file self.checkChecksums = checkChecksums data = self.file.read(sfntDirectorySize) if len(data) <> sfntDirectorySize: from fontTools import ttLib raise ttLib.TTLibError, "Not a TrueType or OpenType font (not enough data)" sstruct.unpack(sfntDirectoryFormat, data, self) if self.sfntVersion == "ttcf": assert ttcHeaderSize == sfntDirectorySize sstruct.unpack(ttcHeaderFormat, data, self) assert self.Version == 0x00010000 or self.Version == 0x00020000, "unrecognized TTC version 0x%08x" % self.Version if not 0 <= fontNumber < self.numFonts: from fontTools import ttLib raise ttLib.TTLibError, "specify a font number between 0 and %d (inclusive)" % (self.numFonts - 1) offsetTable = struct.unpack(">%dL" % self.numFonts, self.file.read(self.numFonts * 4)) if self.Version == 0x00020000: pass # ignoring version 2.0 signatures self.file.seek(offsetTable[fontNumber]) data = self.file.read(sfntDirectorySize) sstruct.unpack(sfntDirectoryFormat, data, self) if self.sfntVersion not in ("\000\001\000\000", "OTTO", "true"): from fontTools import ttLib raise ttLib.TTLibError, "Not a TrueType or OpenType font (bad sfntVersion)" self.tables = {} for i in range(self.numTables): entry = SFNTDirectoryEntry() entry.fromFile(self.file) if entry.length > 0: self.tables[entry.tag] = entry else: # Ignore zero-length tables. This doesn't seem to be documented, # yet it's apparently how the Windows TT rasterizer behaves. # Besides, at least one font has been sighted which actually # *has* a zero-length table. pass def has_key(self, tag): return self.tables.has_key(tag) def keys(self): return self.tables.keys() def __getitem__(self, tag): """Fetch the raw table data.""" entry = self.tables[tag] self.file.seek(entry.offset) data = self.file.read(entry.length) if self.checkChecksums: if tag == 'head': # Beh: we have to special-case the 'head' table. checksum = calcChecksum(data[:8] + '\0\0\0\0' + data[12:]) else: checksum = calcChecksum(data) if self.checkChecksums > 1: # Be obnoxious, and barf when it's wrong assert checksum == entry.checksum, "bad checksum for '%s' table" % tag elif checksum <> entry.checkSum: # Be friendly, and just print a warning. print "bad checksum for '%s' table" % tag return data def __delitem__(self, tag): del self.tables[tag] def close(self): self.file.close() class SFNTWriter: def __init__(self, file, numTables, sfntVersion="\000\001\000\000"): self.file = file self.numTables = numTables self.sfntVersion = sfntVersion self.searchRange, self.entrySelector, self.rangeShift = getSearchRange(numTables) self.nextTableOffset = sfntDirectorySize + numTables * sfntDirectoryEntrySize # clear out directory area self.file.seek(self.nextTableOffset) # make sure we're actually where we want to be. (XXX old cStringIO bug) self.file.write('\0' * (self.nextTableOffset - self.file.tell())) self.tables = {} def __setitem__(self, tag, data): """Write raw table data to disk.""" if self.tables.has_key(tag): # We've written this table to file before. If the length # of the data is still the same, we allow overwriting it. entry = self.tables[tag] if len(data) <> entry.length: from fontTools import ttLib raise ttLib.TTLibError, "cannot rewrite '%s' table: length does not match directory entry" % tag else: entry = SFNTDirectoryEntry() entry.tag = tag entry.offset = self.nextTableOffset entry.length = len(data) self.nextTableOffset = self.nextTableOffset + ((len(data) + 3) & ~3) self.file.seek(entry.offset) self.file.write(data) # Add NUL bytes to pad the table data to a 4-byte boundary. # Don't depend on f.seek() as we need to add the padding even if no # subsequent write follows (seek is lazy), ie. after the final table # in the font. self.file.write('\0' * (self.nextTableOffset - self.file.tell())) assert self.nextTableOffset == self.file.tell() if tag == 'head': entry.checkSum = calcChecksum(data[:8] + '\0\0\0\0' + data[12:]) else: entry.checkSum = calcChecksum(data) self.tables[tag] = entry def close(self): """All tables must have been written to disk. Now write the directory. """ tables = self.tables.items() tables.sort() if len(tables) <> self.numTables: from fontTools import ttLib raise ttLib.TTLibError, "wrong number of tables; expected %d, found %d" % (self.numTables, len(tables)) directory = sstruct.pack(sfntDirectoryFormat, self) self.file.seek(sfntDirectorySize) seenHead = 0 for tag, entry in tables: if tag == "head": seenHead = 1 directory = directory + entry.toString() if seenHead: self.calcMasterChecksum(directory) self.file.seek(0) self.file.write(directory) def calcMasterChecksum(self, directory): # calculate checkSumAdjustment tags = self.tables.keys() checksums = numpy.zeros(len(tags)+1, numpy.int32) for i in range(len(tags)): checksums[i] = self.tables[tags[i]].checkSum directory_end = sfntDirectorySize + len(self.tables) * sfntDirectoryEntrySize assert directory_end == len(directory) checksums[-1] = calcChecksum(directory) checksum = numpy.add.reduce(checksums) # BiboAfba! checksumadjustment = numpy.array(0xb1b0afbaL - 0x100000000L, numpy.int32) - checksum # write the checksum to the file self.file.seek(self.tables['head'].offset + 8) self.file.write(struct.pack(">l", checksumadjustment)) # -- sfnt directory helpers and cruft ttcHeaderFormat = """ > # big endian TTCTag: 4s # "ttcf" Version: L # 0x00010000 or 0x00020000 numFonts: L # number of fonts # OffsetTable[numFonts]: L # array with offsets from beginning of file # ulDsigTag: L # version 2.0 only # ulDsigLength: L # version 2.0 only # ulDsigOffset: L # version 2.0 only """ ttcHeaderSize = sstruct.calcsize(ttcHeaderFormat) sfntDirectoryFormat = """ > # big endian sfntVersion: 4s numTables: H # number of tables searchRange: H # (max2 <= numTables)*16 entrySelector: H # log2(max2 <= numTables) rangeShift: H # numTables*16-searchRange """ sfntDirectorySize = sstruct.calcsize(sfntDirectoryFormat) sfntDirectoryEntryFormat = """ > # big endian tag: 4s checkSum: l offset: l length: l """ sfntDirectoryEntrySize = sstruct.calcsize(sfntDirectoryEntryFormat) class SFNTDirectoryEntry: def fromFile(self, file): sstruct.unpack(sfntDirectoryEntryFormat, file.read(sfntDirectoryEntrySize), self) def fromString(self, str): sstruct.unpack(sfntDirectoryEntryFormat, str, self) def toString(self): return sstruct.pack(sfntDirectoryEntryFormat, self) def __repr__(self): if hasattr(self, "tag"): return "" % (self.tag, id(self)) else: return "" % id(self) def calcChecksum(data, start=0): """Calculate the checksum for an arbitrary block of data. Optionally takes a 'start' argument, which allows you to calculate a checksum in chunks by feeding it a previous result. If the data length is not a multiple of four, it assumes it is to be padded with null byte. """ from fontTools import ttLib remainder = len(data) % 4 if remainder: data = data + '\0' * (4-remainder) a = numpy.fromstring(struct.pack(">l", start) + data, numpy.int32) if sys.byteorder <> "big": a = a.byteswap() return numpy.add.reduce(a) def maxPowerOfTwo(x): """Return the highest exponent of two, so that (2 ** exponent) <= x """ exponent = 0 while x: x = x >> 1 exponent = exponent + 1 return max(exponent - 1, 0) def getSearchRange(n): """Calculate searchRange, entrySelector, rangeShift for the sfnt directory. 'n' is the number of tables. """ # This stuff needs to be stored in the file, because? import math exponent = maxPowerOfTwo(n) searchRange = (2 ** exponent) * 16 entrySelector = exponent rangeShift = n * 16 - searchRange return searchRange, entrySelector, rangeShift