mirror of
https://github.com/noDRM/DeDRM_tools
synced 2024-12-28 22:24:27 +01:00
776 lines
25 KiB
Python
776 lines
25 KiB
Python
#! /usr/bin/python
|
|
# vim:ts=4:sw=4:softtabstop=4:smarttab:expandtab
|
|
# For use with Topaz Scripts Version 2.0
|
|
|
|
from __future__ import with_statement
|
|
import csv
|
|
import sys
|
|
import os
|
|
import getopt
|
|
from struct import pack
|
|
from struct import unpack
|
|
|
|
|
|
# Get a 7 bit encoded number from string. The most
|
|
# significant byte comes first and has the high bit (8th) set
|
|
|
|
def readEncodedNumber(file):
|
|
flag = False
|
|
c = file.read(1)
|
|
if (len(c) == 0):
|
|
return None
|
|
data = ord(c)
|
|
|
|
if data == 0xFF:
|
|
flag = True
|
|
c = file.read(1)
|
|
if (len(c) == 0):
|
|
return None
|
|
data = ord(c)
|
|
|
|
if data >= 0x80:
|
|
datax = (data & 0x7F)
|
|
while data >= 0x80 :
|
|
c = file.read(1)
|
|
if (len(c) == 0):
|
|
return None
|
|
data = ord(c)
|
|
datax = (datax <<7) + (data & 0x7F)
|
|
data = datax
|
|
|
|
if flag:
|
|
data = -data
|
|
return data
|
|
|
|
|
|
# returns a binary string that encodes a number into 7 bits
|
|
# most significant byte first which has the high bit set
|
|
|
|
def encodeNumber(number):
|
|
result = ""
|
|
negative = False
|
|
flag = 0
|
|
|
|
if number < 0 :
|
|
number = -number + 1
|
|
negative = True
|
|
|
|
while True:
|
|
byte = number & 0x7F
|
|
number = number >> 7
|
|
byte += flag
|
|
result += chr(byte)
|
|
flag = 0x80
|
|
if number == 0 : break
|
|
|
|
if negative:
|
|
result += chr(0xFF)
|
|
|
|
return result[::-1]
|
|
|
|
|
|
|
|
# create / read a length prefixed string from the file
|
|
|
|
def lengthPrefixString(data):
|
|
return encodeNumber(len(data))+data
|
|
|
|
def readString(file):
|
|
stringLength = readEncodedNumber(file)
|
|
if (stringLength == None):
|
|
return ""
|
|
sv = file.read(stringLength)
|
|
if (len(sv) != stringLength):
|
|
return ""
|
|
return unpack(str(stringLength)+"s",sv)[0]
|
|
|
|
|
|
# convert a binary string generated by encodeNumber (7 bit encoded number)
|
|
# to the value you would find inside the page*.dat files to be processed
|
|
|
|
def convert(i):
|
|
result = ''
|
|
val = encodeNumber(i)
|
|
for j in xrange(len(val)):
|
|
c = ord(val[j:j+1])
|
|
result += '%02x' % c
|
|
return result
|
|
|
|
|
|
|
|
# the complete string table used to store all book text content
|
|
# as well as the xml tokens and values that make sense out of it
|
|
|
|
class Dictionary(object):
|
|
def __init__(self, dictFile):
|
|
self.filename = dictFile
|
|
self.size = 0
|
|
self.fo = file(dictFile,'rb')
|
|
self.stable = []
|
|
self.size = readEncodedNumber(self.fo)
|
|
for i in xrange(self.size):
|
|
self.stable.append(self.escapestr(readString(self.fo)))
|
|
self.pos = 0
|
|
|
|
def escapestr(self, str):
|
|
str = str.replace('&','&')
|
|
str = str.replace('<','<')
|
|
str = str.replace('>','>')
|
|
str = str.replace('=','=')
|
|
return str
|
|
|
|
def lookup(self,val):
|
|
if ((val >= 0) and (val < self.size)) :
|
|
self.pos = val
|
|
return self.stable[self.pos]
|
|
else:
|
|
print "Error - %d outside of string table limits" % val
|
|
sys.exit(-1)
|
|
|
|
def getSize(self):
|
|
return self.size
|
|
|
|
def getPos(self):
|
|
return self.pos
|
|
|
|
def dumpDict(self):
|
|
for i in xrange(self.size):
|
|
print "%d %s %s" % (i, convert(i), self.stable[i])
|
|
return
|
|
|
|
# parses the xml snippets that are represented by each page*.dat file.
|
|
# also parses the other0.dat file - the main stylesheet
|
|
# and information used to inject the xml snippets into page*.dat files
|
|
|
|
class PageParser(object):
|
|
def __init__(self, filename, dict, debug, flat_xml):
|
|
self.fo = file(filename,'rb')
|
|
self.id = os.path.basename(filename).replace('.dat','')
|
|
self.dict = dict
|
|
self.debug = debug
|
|
self.flat_xml = flat_xml
|
|
self.tagpath = []
|
|
self.doc = []
|
|
self.snippetList = []
|
|
|
|
|
|
# hash table used to enable the decoding process
|
|
# This has all been developed by trial and error so it may still have omissions or
|
|
# contain errors
|
|
# Format:
|
|
# tag : (number of arguments, argument type, subtags present, special case of subtags presents when escaped)
|
|
|
|
token_tags = {
|
|
'x' : (1, 'scalar_number', 0, 0),
|
|
'y' : (1, 'scalar_number', 0, 0),
|
|
'h' : (1, 'scalar_number', 0, 0),
|
|
'w' : (1, 'scalar_number', 0, 0),
|
|
'firstWord' : (1, 'scalar_number', 0, 0),
|
|
'lastWord' : (1, 'scalar_number', 0, 0),
|
|
'rootID' : (1, 'scalar_number', 0, 0),
|
|
'stemID' : (1, 'scalar_number', 0, 0),
|
|
'type' : (1, 'scalar_text', 0, 0),
|
|
|
|
'info' : (0, 'number', 1, 0),
|
|
|
|
'info.word' : (0, 'number', 1, 1),
|
|
'info.word.ocrText' : (1, 'text', 0, 0),
|
|
'info.word.firstGlyph' : (1, 'raw', 0, 0),
|
|
'info.word.lastGlyph' : (1, 'raw', 0, 0),
|
|
'info.word.bl' : (1, 'raw', 0, 0),
|
|
'info.word.link_id' : (1, 'number', 0, 0),
|
|
|
|
'glyph' : (0, 'number', 1, 1),
|
|
'glyph.x' : (1, 'number', 0, 0),
|
|
'glyph.y' : (1, 'number', 0, 0),
|
|
'glyph.glyphID' : (1, 'number', 0, 0),
|
|
|
|
'dehyphen' : (0, 'number', 1, 1),
|
|
'dehyphen.rootID' : (1, 'number', 0, 0),
|
|
'dehyphen.stemID' : (1, 'number', 0, 0),
|
|
'dehyphen.stemPage' : (1, 'number', 0, 0),
|
|
'dehyphen.sh' : (1, 'number', 0, 0),
|
|
|
|
'links' : (0, 'number', 1, 1),
|
|
'links.page' : (1, 'number', 0, 0),
|
|
'links.rel' : (1, 'number', 0, 0),
|
|
'links.row' : (1, 'number', 0, 0),
|
|
'links.title' : (1, 'text', 0, 0),
|
|
'links.href' : (1, 'text', 0, 0),
|
|
'links.type' : (1, 'text', 0, 0),
|
|
|
|
'paraCont' : (0, 'number', 1, 1),
|
|
'paraCont.rootID' : (1, 'number', 0, 0),
|
|
'paraCont.stemID' : (1, 'number', 0, 0),
|
|
'paraCont.stemPage' : (1, 'number', 0, 0),
|
|
|
|
'paraStems' : (0, 'number', 1, 1),
|
|
'paraStems.stemID' : (1, 'number', 0, 0),
|
|
|
|
'wordStems' : (0, 'number', 1, 1),
|
|
'wordStems.stemID' : (1, 'number', 0, 0),
|
|
|
|
'empty' : (1, 'snippets', 1, 0),
|
|
|
|
'page' : (1, 'snippets', 1, 0),
|
|
'page.pageid' : (1, 'scalar_text', 0, 0),
|
|
'page.pagelabel' : (1, 'scalar_text', 0, 0),
|
|
'page.type' : (1, 'scalar_text', 0, 0),
|
|
'page.h' : (1, 'scalar_number', 0, 0),
|
|
'page.w' : (1, 'scalar_number', 0, 0),
|
|
'page.startID' : (1, 'scalar_number', 0, 0),
|
|
|
|
'group' : (1, 'snippets', 1, 0),
|
|
'group.type' : (1, 'scalar_text', 0, 0),
|
|
|
|
'region' : (1, 'snippets', 1, 0),
|
|
'region.type' : (1, 'scalar_text', 0, 0),
|
|
'region.x' : (1, 'scalar_number', 0, 0),
|
|
'region.y' : (1, 'scalar_number', 0, 0),
|
|
'region.h' : (1, 'scalar_number', 0, 0),
|
|
'region.w' : (1, 'scalar_number', 0, 0),
|
|
|
|
'img' : (1, 'snippets', 1, 0),
|
|
'img.x' : (1, 'scalar_number', 0, 0),
|
|
'img.y' : (1, 'scalar_number', 0, 0),
|
|
'img.h' : (1, 'scalar_number', 0, 0),
|
|
'img.w' : (1, 'scalar_number', 0, 0),
|
|
'img.src' : (1, 'scalar_number', 0, 0),
|
|
|
|
'paragraph' : (1, 'snippets', 1, 0),
|
|
'paragraph.class' : (1, 'scalar_text', 0, 0),
|
|
'paragraph.firstWord' : (1, 'scalar_number', 0, 0),
|
|
'paragraph.lastWord' : (1, 'scalar_number', 0, 0),
|
|
|
|
'word_semantic' : (1, 'snippets', 1, 1),
|
|
'word_semantic.type' : (1, 'scalar_text', 0, 0),
|
|
'word_semantic.firstWord' : (1, 'scalar_number', 0, 0),
|
|
'word_semantic.lastWord' : (1, 'scalar_number', 0, 0),
|
|
|
|
'word' : (1, 'snippets', 1, 0),
|
|
'word.type' : (1, 'scalar_text', 0, 0),
|
|
'word.class' : (1, 'scalar_text', 0, 0),
|
|
'word.firstGlyph' : (1, 'scalar_number', 0, 0),
|
|
'word.lastGlyph' : (1, 'scalar_number', 0, 0),
|
|
|
|
'_span' : (1, 'snippets', 1, 0),
|
|
'_span.firstWord' : (1, 'scalar_number', 0, 0),
|
|
'-span.lastWord' : (1, 'scalar_number', 0, 0),
|
|
|
|
'span' : (1, 'snippets', 1, 0),
|
|
'span.firstWord' : (1, 'scalar_number', 0, 0),
|
|
'span.lastWord' : (1, 'scalar_number', 0, 0),
|
|
|
|
'extratokens' : (1, 'snippets', 1, 0),
|
|
'extratokens.type' : (1, 'scalar_text', 0, 0),
|
|
'extratokens.firstGlyph' : (1, 'scalar_number', 0, 0),
|
|
'extratokens.lastGlyph' : (1, 'scalar_number', 0, 0),
|
|
|
|
'glyph.h' : (1, 'number', 0, 0),
|
|
'glyph.w' : (1, 'number', 0, 0),
|
|
'glyph.use' : (1, 'number', 0, 0),
|
|
'glyph.vtx' : (1, 'number', 0, 1),
|
|
'glyph.len' : (1, 'number', 0, 1),
|
|
'glyph.dpi' : (1, 'number', 0, 0),
|
|
'vtx' : (0, 'number', 1, 1),
|
|
'vtx.x' : (1, 'number', 0, 0),
|
|
'vtx.y' : (1, 'number', 0, 0),
|
|
'len' : (0, 'number', 1, 1),
|
|
'len.n' : (1, 'number', 0, 0),
|
|
|
|
'book' : (1, 'snippets', 1, 0),
|
|
'version' : (1, 'snippets', 1, 0),
|
|
'version.FlowEdit_1_id' : (1, 'scalar_text', 0, 0),
|
|
'version.FlowEdit_1_version' : (1, 'scalar_text', 0, 0),
|
|
'version.Schema_id' : (1, 'scalar_text', 0, 0),
|
|
'version.Schema_version' : (1, 'scalar_text', 0, 0),
|
|
'version.Topaz_version' : (1, 'scalar_text', 0, 0),
|
|
'version.WordDetailEdit_1_id' : (1, 'scalar_text', 0, 0),
|
|
'version.WordDetailEdit_1_version' : (1, 'scalar_text', 0, 0),
|
|
'version.ZoneEdit_1_id' : (1, 'scalar_text', 0, 0),
|
|
'version.ZoneEdit_1_version' : (1, 'scalar_text', 0, 0),
|
|
'version.chapterheaders' : (1, 'scalar_text', 0, 0),
|
|
'version.creation_date' : (1, 'scalar_text', 0, 0),
|
|
'version.header_footer' : (1, 'scalar_text', 0, 0),
|
|
'version.init_from_ocr' : (1, 'scalar_text', 0, 0),
|
|
'version.letter_insertion' : (1, 'scalar_text', 0, 0),
|
|
'version.xmlinj_convert' : (1, 'scalar_text', 0, 0),
|
|
'version.xmlinj_reflow' : (1, 'scalar_text', 0, 0),
|
|
'version.xmlinj_transform' : (1, 'scalar_text', 0, 0),
|
|
'version.findlists' : (1, 'scalar_text', 0, 0),
|
|
'version.page_num' : (1, 'scalar_text', 0, 0),
|
|
'version.page_type' : (1, 'scalar_text', 0, 0),
|
|
|
|
'stylesheet' : (1, 'snippets', 1, 0),
|
|
'style' : (1, 'snippets', 1, 0),
|
|
'style._tag' : (1, 'scalar_text', 0, 0),
|
|
'style.type' : (1, 'scalar_text', 0, 0),
|
|
'style._parent_type' : (1, 'scalar_text', 0, 0),
|
|
'style.class' : (1, 'scalar_text', 0, 0),
|
|
'style._after_class' : (1, 'scalar_text', 0, 0),
|
|
'rule' : (1, 'snippets', 1, 0),
|
|
'rule.attr' : (1, 'scalar_text', 0, 0),
|
|
'rule.value' : (1, 'scalar_text', 0, 0),
|
|
|
|
'original' : (0, 'number', 1, 1),
|
|
'original.pnum' : (1, 'number', 0, 0),
|
|
'original.pid' : (1, 'text', 0, 0),
|
|
'pages' : (0, 'number', 1, 1),
|
|
'pages.ref' : (1, 'number', 0, 0),
|
|
'pages.id' : (1, 'number', 0, 0),
|
|
'startID' : (0, 'number', 1, 1),
|
|
'startID.page' : (1, 'number', 0, 0),
|
|
'startID.id' : (1, 'number', 0, 0),
|
|
|
|
}
|
|
|
|
|
|
# full tag path record keeping routines
|
|
def tag_push(self, token):
|
|
self.tagpath.append(token)
|
|
def tag_pop(self):
|
|
if len(self.tagpath) > 0 :
|
|
self.tagpath.pop()
|
|
def tagpath_len(self):
|
|
return len(self.tagpath)
|
|
def get_tagpath(self, i):
|
|
cnt = len(self.tagpath)
|
|
if i < cnt : result = self.tagpath[i]
|
|
for j in xrange(i+1, cnt) :
|
|
result += '.' + self.tagpath[j]
|
|
return result
|
|
|
|
|
|
# list of absolute command byte values values that indicate
|
|
# various types of loop meachanisms typically used to generate vectors
|
|
|
|
cmd_list = (0x76, 0x76)
|
|
|
|
# peek at and return 1 byte that is ahead by i bytes
|
|
def peek(self, aheadi):
|
|
c = self.fo.read(aheadi)
|
|
if (len(c) == 0):
|
|
return None
|
|
self.fo.seek(-aheadi,1)
|
|
c = c[-1:]
|
|
return ord(c)
|
|
|
|
|
|
# get the next value from the file being processed
|
|
def getNext(self):
|
|
nbyte = self.peek(1);
|
|
if (nbyte == None):
|
|
return None
|
|
val = readEncodedNumber(self.fo)
|
|
return val
|
|
|
|
|
|
# format an arg by argtype
|
|
def formatArg(self, arg, argtype):
|
|
if (argtype == 'text') or (argtype == 'scalar_text') :
|
|
result = self.dict.lookup(arg)
|
|
elif (argtype == 'raw') or (argtype == 'number') or (argtype == 'scalar_number') :
|
|
result = arg
|
|
elif (argtype == 'snippets') :
|
|
result = arg
|
|
else :
|
|
print "Error Unknown argtype %s" % argtype
|
|
sys.exit(-2)
|
|
return result
|
|
|
|
|
|
# process the next tag token, recursively handling subtags,
|
|
# arguments, and commands
|
|
def procToken(self, token):
|
|
|
|
known_token = False
|
|
self.tag_push(token)
|
|
|
|
if self.debug : print 'Processing: ', self.get_tagpath(0)
|
|
cnt = self.tagpath_len()
|
|
for j in xrange(cnt):
|
|
tkn = self.get_tagpath(j)
|
|
if tkn in self.token_tags :
|
|
num_args = self.token_tags[tkn][0]
|
|
argtype = self.token_tags[tkn][1]
|
|
subtags = self.token_tags[tkn][2]
|
|
splcase = self.token_tags[tkn][3]
|
|
ntags = -1
|
|
known_token = True
|
|
break
|
|
|
|
if known_token :
|
|
|
|
# handle subtags if present
|
|
subtagres = []
|
|
if (splcase == 1):
|
|
# this type of tag uses of escape marker 0x74 indicate subtag count
|
|
if self.peek(1) == 0x74:
|
|
skip = readEncodedNumber(self.fo)
|
|
subtags = 1
|
|
num_args = 0
|
|
|
|
if (subtags == 1):
|
|
ntags = readEncodedNumber(self.fo)
|
|
if self.debug : print 'subtags: ' + token + ' has ' + str(ntags)
|
|
for j in xrange(ntags):
|
|
val = readEncodedNumber(self.fo)
|
|
subtagres.append(self.procToken(self.dict.lookup(val)))
|
|
|
|
# arguments can be scalars or vectors of text or numbers
|
|
argres = []
|
|
if num_args > 0 :
|
|
firstarg = self.peek(1)
|
|
if (firstarg in self.cmd_list) and (argtype != 'scalar_number') and (argtype != 'scalar_text'):
|
|
# single argument is a variable length vector of data
|
|
arg = readEncodedNumber(self.fo)
|
|
argres = self.decodeCMD(arg,argtype)
|
|
else :
|
|
# num_arg scalar arguments
|
|
for i in xrange(num_args):
|
|
argres.append(self.formatArg(readEncodedNumber(self.fo), argtype))
|
|
|
|
# build the return tag
|
|
result = []
|
|
tkn = self.get_tagpath(0)
|
|
result.append(tkn)
|
|
result.append(subtagres)
|
|
result.append(argtype)
|
|
result.append(argres)
|
|
self.tag_pop()
|
|
return result
|
|
|
|
# all tokens that need to be processed should be in the hash
|
|
# table if it may indicate a problem, either new token
|
|
# or an out of sync condition
|
|
else:
|
|
result = []
|
|
if (self.debug):
|
|
print 'Unknown Token:', token
|
|
self.tag_pop()
|
|
return result
|
|
|
|
|
|
# special loop used to process code snippets
|
|
# it is NEVER used to format arguments.
|
|
# builds the snippetList
|
|
def doLoop72(self, argtype):
|
|
cnt = readEncodedNumber(self.fo)
|
|
if self.debug :
|
|
result = 'Set of '+ str(cnt) + ' xml snippets. The overall structure \n'
|
|
result += 'of the document is indicated by snippet number sets at the\n'
|
|
result += 'end of each snippet. \n'
|
|
print result
|
|
for i in xrange(cnt):
|
|
if self.debug: print 'Snippet:',str(i)
|
|
snippet = []
|
|
snippet.append(i)
|
|
val = readEncodedNumber(self.fo)
|
|
snippet.append(self.procToken(self.dict.lookup(val)))
|
|
self.snippetList.append(snippet)
|
|
return
|
|
|
|
|
|
|
|
# general loop code gracisouly submitted by "skindle" - thank you!
|
|
def doLoop76Mode(self, argtype, cnt, mode):
|
|
result = []
|
|
adj = 0
|
|
if mode & 1:
|
|
adj = readEncodedNumber(self.fo)
|
|
mode = mode >> 1
|
|
x = []
|
|
for i in xrange(cnt):
|
|
x.append(readEncodedNumber(self.fo) - adj)
|
|
for i in xrange(mode):
|
|
for j in xrange(1, cnt):
|
|
x[j] = x[j] + x[j - 1]
|
|
for i in xrange(cnt):
|
|
result.append(self.formatArg(x[i],argtype))
|
|
return result
|
|
|
|
|
|
# dispatches loop commands bytes with various modes
|
|
# The 0x76 style loops are used to build vectors
|
|
|
|
# This was all derived by trial and error and
|
|
# new loop types may exist that are not handled here
|
|
# since they did not appear in the test cases
|
|
|
|
def decodeCMD(self, cmd, argtype):
|
|
if (cmd == 0x76):
|
|
|
|
# loop with cnt, and mode to control loop styles
|
|
cnt = readEncodedNumber(self.fo)
|
|
mode = readEncodedNumber(self.fo)
|
|
|
|
if self.debug : print 'Loop for', cnt, 'with mode', mode, ': '
|
|
return self.doLoop76Mode(argtype, cnt, mode)
|
|
|
|
if self.dbug: print "Unknown command", cmd
|
|
result = []
|
|
return result
|
|
|
|
|
|
|
|
# add full tag path to injected snippets
|
|
def updateName(self, tag, prefix):
|
|
name = tag[0]
|
|
subtagList = tag[1]
|
|
argtype = tag[2]
|
|
argList = tag[3]
|
|
nname = prefix + '.' + name
|
|
nsubtaglist = []
|
|
for j in subtagList:
|
|
nsubtaglist.append(self.updateName(j,prefix))
|
|
ntag = []
|
|
ntag.append(nname)
|
|
ntag.append(nsubtaglist)
|
|
ntag.append(argtype)
|
|
ntag.append(argList)
|
|
return ntag
|
|
|
|
|
|
|
|
# perform depth first injection of specified snippets into this one
|
|
def injectSnippets(self, snippet):
|
|
snipno, tag = snippet
|
|
name = tag[0]
|
|
subtagList = tag[1]
|
|
argtype = tag[2]
|
|
argList = tag[3]
|
|
nsubtagList = []
|
|
if len(argList) > 0 :
|
|
for j in argList:
|
|
asnip = self.snippetList[j]
|
|
aso, atag = self.injectSnippets(asnip)
|
|
atag = self.updateName(atag, name)
|
|
nsubtagList.append(atag)
|
|
argtype='number'
|
|
argList=[]
|
|
if len(nsubtagList) > 0 :
|
|
subtagList.extend(nsubtagList)
|
|
tag = []
|
|
tag.append(name)
|
|
tag.append(subtagList)
|
|
tag.append(argtype)
|
|
tag.append(argList)
|
|
snippet = []
|
|
snippet.append(snipno)
|
|
snippet.append(tag)
|
|
return snippet
|
|
|
|
|
|
|
|
# format the tag for output
|
|
def formatTag(self, node):
|
|
name = node[0]
|
|
subtagList = node[1]
|
|
argtype = node[2]
|
|
argList = node[3]
|
|
fullpathname = name.split('.')
|
|
nodename = fullpathname.pop()
|
|
ilvl = len(fullpathname)
|
|
indent = ' ' * (3 * ilvl)
|
|
result = indent + '<' + nodename + '>'
|
|
if len(argList) > 0:
|
|
argres = ''
|
|
for j in argList:
|
|
if (argtype == 'text') or (argtype == 'scalar_text') :
|
|
argres += j + '|'
|
|
else :
|
|
argres += str(j) + ','
|
|
argres = argres[0:-1]
|
|
if argtype == 'snippets' :
|
|
result += 'snippets:' + argres
|
|
else :
|
|
result += argres
|
|
if len(subtagList) > 0 :
|
|
result += '\n'
|
|
for j in subtagList:
|
|
if len(j) > 0 :
|
|
result += self.formatTag(j)
|
|
result += indent + '</' + nodename + '>\n'
|
|
else:
|
|
result += '</' + nodename + '>\n'
|
|
return result
|
|
|
|
|
|
# flatten tag
|
|
def flattenTag(self, node):
|
|
name = node[0]
|
|
subtagList = node[1]
|
|
argtype = node[2]
|
|
argList = node[3]
|
|
result = name
|
|
if (len(argList) > 0):
|
|
argres = ''
|
|
for j in argList:
|
|
if (argtype == 'text') or (argtype == 'scalar_text') :
|
|
argres += j + '|'
|
|
else :
|
|
argres += str(j) + '|'
|
|
argres = argres[0:-1]
|
|
if argtype == 'snippets' :
|
|
result += '.snippets=' + argres
|
|
else :
|
|
result += '=' + argres
|
|
result += '\n'
|
|
for j in subtagList:
|
|
if len(j) > 0 :
|
|
result += self.flattenTag(j)
|
|
return result
|
|
|
|
|
|
# reduce create xml output
|
|
def formatDoc(self, flat_xml):
|
|
result = ''
|
|
for j in self.doc :
|
|
if len(j) > 0:
|
|
if flat_xml:
|
|
result += self.flattenTag(j)
|
|
else:
|
|
result += self.formatTag(j)
|
|
if self.debug : print result
|
|
return result
|
|
|
|
|
|
|
|
# main loop - parse the page.dat files
|
|
# to create structured document and snippets
|
|
|
|
# FIXME: value at end of magic appears to be a subtags count
|
|
# but for what? For now, inject an 'info" tag as it is in
|
|
# every dictionary and seems close to what is meant
|
|
# The alternative is to special case the last _ "0x5f" to mean something
|
|
|
|
def process(self):
|
|
|
|
# peek at the first bytes to see what type of file it is
|
|
magic = self.fo.read(11)
|
|
if (magic[0:1] == 'p') and (magic[2:10] == '__PAGE__'):
|
|
first_token = 'info'
|
|
elif (magic[0:1] == 'g') and (magic[2:11] == '__GLYPH__'):
|
|
skip = self.fo.read(1)
|
|
first_token = 'info'
|
|
else :
|
|
# other0.dat file
|
|
first_token = None
|
|
self.fo.seek(-11,1)
|
|
|
|
|
|
# main loop to read and build the document tree
|
|
while True:
|
|
|
|
if first_token != None :
|
|
# use "inserted" first token 'info' for page and glyph files
|
|
tag = self.procToken(first_token)
|
|
if len(tag) > 0 :
|
|
self.doc.append(tag)
|
|
first_token = None
|
|
|
|
v = self.getNext()
|
|
if (v == None):
|
|
break
|
|
|
|
if (v == 0x72):
|
|
self.doLoop72('number')
|
|
elif (v > 0) and (v < self.dict.getSize()) :
|
|
tag = self.procToken(self.dict.lookup(v))
|
|
if len(tag) > 0 :
|
|
self.doc.append(tag)
|
|
else:
|
|
if self.debug:
|
|
print "Main Loop: Unknown value: %x" % v
|
|
|
|
|
|
# now do snippet injection
|
|
if len(self.snippetList) > 0 :
|
|
if self.debug : print 'Injecting Snippets:'
|
|
snippet = self.injectSnippets(self.snippetList[0])
|
|
snipno = snippet[0]
|
|
tag_add = snippet[1]
|
|
if self.debug : print self.formatTag(tag_add)
|
|
if len(tag_add) > 0:
|
|
self.doc.append(tag_add)
|
|
|
|
# handle generation of xml output
|
|
xmlpage = self.formatDoc(self.flat_xml)
|
|
|
|
return xmlpage
|
|
|
|
|
|
|
|
def usage():
|
|
print 'Usage: '
|
|
print ' convert2xml.py dict0000.dat infile.dat '
|
|
print ' '
|
|
print ' Options:'
|
|
print ' -h print this usage help message '
|
|
print ' -d turn on debug output to check for potential errors '
|
|
print ' --flat-xml output the flattened xml page description only '
|
|
print ' '
|
|
print ' This program will attempt to convert a page*.dat file or '
|
|
print ' glyphs*.dat file, using the dict0000.dat file, to its xml description. '
|
|
print ' '
|
|
print ' Use "cmbtc_dump.py" first to unencrypt, uncompress, and dump '
|
|
print ' the *.dat files from a Topaz format e-book.'
|
|
|
|
#
|
|
# Main
|
|
#
|
|
|
|
def main(argv):
|
|
dictFile = ""
|
|
pageFile = ""
|
|
debug = False
|
|
flat_xml = False
|
|
printOutput = False
|
|
if len(argv) == 0:
|
|
printOutput = True
|
|
argv = sys.argv
|
|
else :
|
|
argv = argv.split()
|
|
|
|
try:
|
|
opts, args = getopt.getopt(argv[1:], "hd", ["flat-xml"])
|
|
|
|
except getopt.GetoptError, err:
|
|
|
|
# print help information and exit:
|
|
print str(err) # will print something like "option -a not recognized"
|
|
usage()
|
|
sys.exit(2)
|
|
|
|
if len(opts) == 0 and len(args) == 0 :
|
|
usage()
|
|
sys.exit(2)
|
|
|
|
for o, a in opts:
|
|
if o =="-d":
|
|
debug=True
|
|
if o =="-h":
|
|
usage()
|
|
sys.exit(0)
|
|
if o =="--flat-xml":
|
|
flat_xml = True
|
|
|
|
dictFile, pageFile = args[0], args[1]
|
|
|
|
# read in the string table dictionary
|
|
dict = Dictionary(dictFile)
|
|
# dict.dumpDict()
|
|
|
|
# create a page parser
|
|
pp = PageParser(pageFile, dict, debug, flat_xml)
|
|
|
|
xmlpage = pp.process()
|
|
|
|
if printOutput:
|
|
print xmlpage
|
|
return 0
|
|
|
|
return xmlpage
|
|
|
|
if __name__ == '__main__':
|
|
sys.exit(main(''))
|