ODT Input: Speed up conversion of ODT files that define huge amounts of redundant style information. Fixes #777468 (Conversion from ODT to EPUB extremely slow)

src/calibre/ebooks/odt/input.py

@@ -7,6 +7,8 @@ __docformat__ = 'restructuredtext en'
 Convert an ODT file into a Open Ebook
 '''
 import os
+
+from lxml import etree
 from odf.odf2xhtml import ODF2XHTML
 
 from calibre import CurrentDir, walk
@@ -23,7 +25,48 @@ class Extract(ODF2XHTML):
                 with open(name, 'wb') as f:
                     f.write(data)
 
-    def __call__(self, stream, odir):
+    def filter_css(self, html, log):
+        root = etree.fromstring(html)
+        style = root.xpath('//*[local-name() = "style" and @type="text/css"]')
+        if style:
+            style = style[0]
+            css = style.text
+            if css:
+                style.text, sel_map = self.do_filter_css(css)
+                for x in root.xpath('//*[@class]'):
+                    extra = []
+                    orig = x.get('class')
+                    for cls in orig.split():
+                        extra.extend(sel_map.get(cls, []))
+                    if extra:
+                        x.set('class', orig + ' ' + ' '.join(extra))
+                html = etree.tostring(root, encoding='utf-8',
+                        xml_declaration=True)
+        return html
+
+    def do_filter_css(self, css):
+        from cssutils import parseString
+        from cssutils.css import CSSRule
+        sheet = parseString(css)
+        rules = list(sheet.cssRules.rulesOfType(CSSRule.STYLE_RULE))
+        sel_map = {}
+        count = 0
+        for r in rules:
+            # Check if we have only class selectors for this rule
+            nc = [x for x in r.selectorList if not
+                    x.selectorText.startswith('.')]
+            if len(r.selectorList) > 1 and not nc:
+                replace_name = 'c_odt%d'%count
+                count += 1
+                for sel in r.selectorList:
+                    s = sel.selectorText[1:]
+                    if s not in sel_map:
+                        sel_map[s] = []
+                    sel_map[s].append(replace_name)
+                r.selectorText = '.'+replace_name
+        return sheet.cssText, sel_map
+
+    def __call__(self, stream, odir, log):
         from calibre.utils.zipfile import ZipFile
         from calibre.ebooks.metadata.meta import get_metadata
         from calibre.ebooks.metadata.opf2 import OPFCreator
@@ -32,13 +75,17 @@ class Extract(ODF2XHTML):
         if not os.path.exists(odir):
             os.makedirs(odir)
         with CurrentDir(odir):
-            print 'Extracting ODT file...'
+            log('Extracting ODT file...')
             html = self.odf2xhtml(stream)
             # A blanket img specification like this causes problems
-            # with EPUB output as the contaiing element often has
+            # with EPUB output as the containing element often has
             # an absolute height and width set that is larger than
             # the available screen real estate
             html = html.replace('img { width: 100%; height: 100%; }', '')
+            try:
+                html = self.filter_css(html, log)
+            except:
+                log.exception('Failed to filter CSS, conversion may be slow')
             with open('index.xhtml', 'wb') as f:
                 f.write(html.encode('utf-8'))
             zf = ZipFile(stream, 'r')
@@ -67,7 +114,7 @@ class ODTInput(InputFormatPlugin):
 
     def convert(self, stream, options, file_ext, log,
                 accelerators):
-        return Extract()(stream, '.')
+        return Extract()(stream, '.', log)
 
     def postprocess_book(self, oeb, opts, log):
         # Fix <p><div> constructs as the asinine epubchecker complains

src/odf/odf2xhtml.py

@@ -841,11 +841,19 @@ ol, ul { padding-left: 2em; }
             self.styledict[name] = styles
         # Write the styles to HTML
         self.writeout(self.default_styles)
+        # Changed by Kovid to not write out endless copies of the same style
+        css_styles = {}
         for name in self.stylestack:
             styles = self.styledict.get(name)
-            css2 = self.cs.convert_styles(styles)
+            css2 = tuple(self.cs.convert_styles(styles).iteritems())
-            self.writeout("%s {\n" % name)
+            if css2 in css_styles:
-            for style, val in css2.items():
+                css_styles[css2].append(name)
+            else:
+                css_styles[css2] = [name]
+
+        for css2, names in css_styles.iteritems():
+            self.writeout("%s {\n" % ', '.join(names))
+            for style, val in css2:
                 self.writeout("\t%s: %s;\n" % (style, val) )
             self.writeout("}\n")