kopia lustrzana https://github.com/noDRM/DeDRM_tools
256 wiersze
10 KiB
Python
256 wiersze
10 KiB
Python
#! /usr/bin/python
|
|
# vim:ts=4:sw=4:softtabstop=4:smarttab:expandtab
|
|
|
|
import sys
|
|
import csv
|
|
import os
|
|
import getopt
|
|
from struct import pack
|
|
from struct import unpack
|
|
|
|
|
|
class PParser(object):
|
|
def __init__(self, gd, flatxml, meta_array):
|
|
self.gd = gd
|
|
self.flatdoc = flatxml.split(b'\n')
|
|
self.docSize = len(self.flatdoc)
|
|
self.temp = []
|
|
|
|
self.ph = -1
|
|
self.pw = -1
|
|
startpos = self.posinDoc('page.h') or self.posinDoc('book.h')
|
|
for p in startpos:
|
|
(name, argres) = self.lineinDoc(p)
|
|
self.ph = max(self.ph, int(argres))
|
|
startpos = self.posinDoc('page.w') or self.posinDoc('book.w')
|
|
for p in startpos:
|
|
(name, argres) = self.lineinDoc(p)
|
|
self.pw = max(self.pw, int(argres))
|
|
|
|
if self.ph <= 0:
|
|
self.ph = int(meta_array.get('pageHeight', '11000'))
|
|
if self.pw <= 0:
|
|
self.pw = int(meta_array.get('pageWidth', '8500'))
|
|
|
|
res = []
|
|
startpos = self.posinDoc('info.glyph.x')
|
|
for p in startpos:
|
|
argres = self.getDataatPos('info.glyph.x', p)
|
|
res.extend(argres)
|
|
self.gx = res
|
|
|
|
res = []
|
|
startpos = self.posinDoc('info.glyph.y')
|
|
for p in startpos:
|
|
argres = self.getDataatPos('info.glyph.y', p)
|
|
res.extend(argres)
|
|
self.gy = res
|
|
|
|
res = []
|
|
startpos = self.posinDoc('info.glyph.glyphID')
|
|
for p in startpos:
|
|
argres = self.getDataatPos('info.glyph.glyphID', p)
|
|
res.extend(argres)
|
|
self.gid = res
|
|
|
|
|
|
# return tag at line pos in document
|
|
def lineinDoc(self, pos) :
|
|
if (pos >= 0) and (pos < self.docSize) :
|
|
item = self.flatdoc[pos]
|
|
if item.find(b'=') >= 0:
|
|
(name, argres) = item.split(b'=',1)
|
|
else :
|
|
name = item
|
|
argres = b''
|
|
return name, argres
|
|
|
|
# find tag in doc if within pos to end inclusive
|
|
def findinDoc(self, tagpath, pos, end) :
|
|
result = None
|
|
if end == -1 :
|
|
end = self.docSize
|
|
else:
|
|
end = min(self.docSize, end)
|
|
foundat = -1
|
|
for j in range(pos, end):
|
|
item = self.flatdoc[j]
|
|
if item.find(b'=') >= 0:
|
|
(name, argres) = item.split(b'=',1)
|
|
else :
|
|
name = item
|
|
argres = b''
|
|
if (isinstance(tagpath,str)):
|
|
tagpath = tagpath.encode('utf-8')
|
|
if name.endswith(tagpath) :
|
|
result = argres
|
|
foundat = j
|
|
break
|
|
return foundat, result
|
|
|
|
# return list of start positions for the tagpath
|
|
def posinDoc(self, tagpath):
|
|
startpos = []
|
|
pos = 0
|
|
res = ""
|
|
while res != None :
|
|
(foundpos, res) = self.findinDoc(tagpath, pos, -1)
|
|
if res != None :
|
|
startpos.append(foundpos)
|
|
pos = foundpos + 1
|
|
return startpos
|
|
|
|
def getData(self, path):
|
|
result = None
|
|
cnt = len(self.flatdoc)
|
|
for j in range(cnt):
|
|
item = self.flatdoc[j]
|
|
if item.find(b'=') >= 0:
|
|
(name, argt) = item.split(b'=')
|
|
argres = argt.split(b'|')
|
|
else:
|
|
name = item
|
|
argres = []
|
|
if (name.endswith(path)):
|
|
result = argres
|
|
break
|
|
if (len(argres) > 0) :
|
|
for j in range(0,len(argres)):
|
|
argres[j] = int(argres[j])
|
|
return result
|
|
|
|
def getDataatPos(self, path, pos):
|
|
result = None
|
|
item = self.flatdoc[pos]
|
|
if item.find(b'=') >= 0:
|
|
(name, argt) = item.split(b'=')
|
|
argres = argt.split(b'|')
|
|
else:
|
|
name = item
|
|
argres = []
|
|
if (len(argres) > 0) :
|
|
for j in range(0,len(argres)):
|
|
argres[j] = int(argres[j])
|
|
if (isinstance(path,str)):
|
|
path = path.encode('utf-8')
|
|
if (name.endswith(path)):
|
|
result = argres
|
|
return result
|
|
|
|
def getDataTemp(self, path):
|
|
result = None
|
|
cnt = len(self.temp)
|
|
for j in range(cnt):
|
|
item = self.temp[j]
|
|
if item.find(b'=') >= 0:
|
|
(name, argt) = item.split(b'=')
|
|
argres = argt.split(b'|')
|
|
else:
|
|
name = item
|
|
argres = []
|
|
if (isinstance(path,str)):
|
|
path = path.encode('utf-8')
|
|
if (name.endswith(path)):
|
|
result = argres
|
|
self.temp.pop(j)
|
|
break
|
|
if (len(argres) > 0) :
|
|
for j in range(0,len(argres)):
|
|
argres[j] = int(argres[j])
|
|
return result
|
|
|
|
def getImages(self):
|
|
result = []
|
|
self.temp = self.flatdoc
|
|
while (self.getDataTemp('img') != None):
|
|
h = self.getDataTemp('img.h')[0]
|
|
w = self.getDataTemp('img.w')[0]
|
|
x = self.getDataTemp('img.x')[0]
|
|
y = self.getDataTemp('img.y')[0]
|
|
src = self.getDataTemp('img.src')[0]
|
|
result.append('<image xlink:href="../img/img%04d.jpg" x="%d" y="%d" width="%d" height="%d" />\n' % (src, x, y, w, h))
|
|
return result
|
|
|
|
def getGlyphs(self):
|
|
result = []
|
|
if (self.gid != None) and (len(self.gid) > 0):
|
|
glyphs = []
|
|
for j in set(self.gid):
|
|
glyphs.append(j)
|
|
glyphs.sort()
|
|
for gid in glyphs:
|
|
id='id="gl%d"' % gid
|
|
path = self.gd.lookup(id)
|
|
if path:
|
|
result.append(id + ' ' + path)
|
|
return result
|
|
|
|
|
|
def convert2SVG(gdict, flat_xml, pageid, previd, nextid, svgDir, raw, meta_array, scaledpi):
|
|
mlst = []
|
|
pp = PParser(gdict, flat_xml, meta_array)
|
|
mlst.append('<?xml version="1.0" standalone="no"?>\n')
|
|
if (raw):
|
|
mlst.append('<!DOCTYPE svg PUBLIC "-//W3C/DTD SVG 1.1//EN" "http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd">\n')
|
|
mlst.append('<svg width="%fin" height="%fin" viewBox="0 0 %d %d" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" version="1.1">\n' % (pp.pw / scaledpi, pp.ph / scaledpi, pp.pw -1, pp.ph -1))
|
|
mlst.append('<title>Page %d - %s by %s</title>\n' % (pageid, meta_array['Title'],meta_array['Authors']))
|
|
else:
|
|
mlst.append('<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.1//EN" "http://www.w3.org/TR/xhtml11/DTD/xhtml11.dtd">\n')
|
|
mlst.append('<html xmlns="http://www.w3.org/1999/xhtml" xml:lang="en" ><head>\n')
|
|
mlst.append('<title>Page %d - %s by %s</title>\n' % (pageid, meta_array['Title'],meta_array['Authors']))
|
|
mlst.append('<script><![CDATA[\n')
|
|
mlst.append('function gd(){var p=window.location.href.replace(/^.*\?dpi=(\d+).*$/i,"$1");return p;}\n')
|
|
mlst.append('var dpi=%d;\n' % scaledpi)
|
|
if (previd) :
|
|
mlst.append('var prevpage="page%04d.xhtml";\n' % (previd))
|
|
if (nextid) :
|
|
mlst.append('var nextpage="page%04d.xhtml";\n' % (nextid))
|
|
mlst.append('var pw=%d;var ph=%d;' % (pp.pw, pp.ph))
|
|
mlst.append('function zoomin(){dpi=dpi*(0.8);setsize();}\n')
|
|
mlst.append('function zoomout(){dpi=dpi*1.25;setsize();}\n')
|
|
mlst.append('function setsize(){var svg=document.getElementById("svgimg");var prev=document.getElementById("prevsvg");var next=document.getElementById("nextsvg");var width=(pw/dpi)+"in";var height=(ph/dpi)+"in";svg.setAttribute("width",width);svg.setAttribute("height",height);prev.setAttribute("height",height);prev.setAttribute("width","50px");next.setAttribute("height",height);next.setAttribute("width","50px");}\n')
|
|
mlst.append('function ppage(){window.location.href=prevpage+"?dpi="+Math.round(dpi);}\n')
|
|
mlst.append('function npage(){window.location.href=nextpage+"?dpi="+Math.round(dpi);}\n')
|
|
mlst.append('var gt=gd();if(gt>0){dpi=gt;}\n')
|
|
mlst.append('window.onload=setsize;\n')
|
|
mlst.append(']]></script>\n')
|
|
mlst.append('</head>\n')
|
|
mlst.append('<body onLoad="setsize();" style="background-color:#777;text-align:center;">\n')
|
|
mlst.append('<div style="white-space:nowrap;">\n')
|
|
if previd == None:
|
|
mlst.append('<a href="javascript:ppage();"><svg id="prevsvg" viewBox="0 0 100 300" xmlns="http://www.w3.org/2000/svg" version="1.1" style="background-color:#777"></svg></a>\n')
|
|
else:
|
|
mlst.append('<a href="javascript:ppage();"><svg id="prevsvg" viewBox="0 0 100 300" xmlns="http://www.w3.org/2000/svg" version="1.1" style="background-color:#777"><polygon points="5,150,95,5,95,295" fill="#AAAAAA" /></svg></a>\n')
|
|
|
|
mlst.append('<a href="javascript:npage();"><svg id="svgimg" viewBox="0 0 %d %d" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" version="1.1" style="background-color:#FFF;border:1px solid black;">' % (pp.pw, pp.ph))
|
|
if (pp.gid != None):
|
|
mlst.append('<defs>\n')
|
|
gdefs = pp.getGlyphs()
|
|
for j in range(0,len(gdefs)):
|
|
mlst.append(gdefs[j])
|
|
mlst.append('</defs>\n')
|
|
img = pp.getImages()
|
|
if (img != None):
|
|
for j in range(0,len(img)):
|
|
mlst.append(img[j])
|
|
if (pp.gid != None):
|
|
for j in range(0,len(pp.gid)):
|
|
mlst.append('<use xlink:href="#gl%d" x="%d" y="%d" />\n' % (pp.gid[j], pp.gx[j], pp.gy[j]))
|
|
if (img == None or len(img) == 0) and (pp.gid == None or len(pp.gid) == 0):
|
|
xpos = "%d" % (pp.pw // 3)
|
|
ypos = "%d" % (pp.ph // 3)
|
|
mlst.append('<text x="' + xpos + '" y="' + ypos + '" font-size="' + meta_array['fontSize'] + '" font-family="Helvetica" stroke="black">This page intentionally left blank.</text>\n')
|
|
if (raw) :
|
|
mlst.append('</svg>')
|
|
else :
|
|
mlst.append('</svg></a>\n')
|
|
if nextid == None:
|
|
mlst.append('<a href="javascript:npage();"><svg id="nextsvg" viewBox="0 0 100 300" xmlns="http://www.w3.org/2000/svg" version="1.1" style="background-color:#777"></svg></a>\n')
|
|
else :
|
|
mlst.append('<a href="javascript:npage();"><svg id="nextsvg" viewBox="0 0 100 300" xmlns="http://www.w3.org/2000/svg" version="1.1" style="background-color:#777"><polygon points="5,5,5,295,95,150" fill="#AAAAAA" /></svg></a>\n')
|
|
mlst.append('</div>\n')
|
|
mlst.append('<div><a href="javascript:zoomin();">zoom in</a> - <a href="javascript:zoomout();">zoom out</a></div>\n')
|
|
mlst.append('</body>\n')
|
|
mlst.append('</html>\n')
|
|
return "".join(mlst)
|