diff options
author | Yuri Takhteyev <yuri@freewisdom.org> | 2008-10-07 02:00:58 -0700 |
---|---|---|
committer | Yuri Takhteyev <yuri@freewisdom.org> | 2008-10-07 02:00:58 -0700 |
commit | 9a7f507d5f40e44f9573347955e9f70169ed5990 (patch) | |
tree | a7632372272921b3602524720ba52af775efe159 /markdown.py | |
parent | 40b8986ccf0ea3fa37dda469b46261dfbf0c25a4 (diff) | |
download | markdown-9a7f507d5f40e44f9573347955e9f70169ed5990.tar.gz markdown-9a7f507d5f40e44f9573347955e9f70169ed5990.tar.bz2 markdown-9a7f507d5f40e44f9573347955e9f70169ed5990.zip |
More cleanup.
Diffstat (limited to 'markdown.py')
-rwxr-xr-x | markdown.py | 102 |
1 files changed, 34 insertions, 68 deletions
diff --git a/markdown.py b/markdown.py index 95bf61d..562380a 100755 --- a/markdown.py +++ b/markdown.py @@ -1172,42 +1172,6 @@ class Markdown: for pattern in self.inlinePatterns: pattern.safe_mode = self.safeMode - - def _transform(self): - """Transform the Markdown text into a XHTML body document. - - Returns: ElementTree object - - """ - # Setup the document - - self.root = etree.Element("div") - - # Split into lines and run the preprocessors that will work with - # self.lines - - self.lines = self.source.split("\n") - - # Run the pre-processors on the lines - for prep in self.preprocessors : - self.lines = prep.run(self.lines) - - # Create a ElementTree from the lines - - buffer = [] - for line in self.lines: - if line.startswith("#"): - - self._processSection(self.root, buffer) - buffer = [line] - else: - buffer.append(line) - - self._processSection(self.root, buffer) - - return etree.ElementTree(self.root) - - def _processSection(self, parent_elem, lines, inList=0, looseList=0): """ @@ -1591,17 +1555,13 @@ class Markdown: if isinstance(data, AtomicString): return data - startIndex = 0 - + startIndex = 0 while patternIndex < len(self.inlinePatterns): - data, matched, startIndex = self._applyInline( self.inlinePatterns[patternIndex], data, patternIndex, startIndex) - if not matched: patternIndex += 1 - return data def _applyInline(self, pattern, data, patternIndex, startIndex=0): @@ -1709,14 +1669,11 @@ class Markdown: strartIndex = 0 while data: - index = data.find(prefix, strartIndex) if index != -1: - id, phEndIndex = self.inlineStash.extractId(data, index) - + if self.inlineStash.isin(id): - node = self.inlineStash.get(id) if index > 0: @@ -1724,19 +1681,14 @@ class Markdown: linkText(text) if not isString(node): # it's Element - for child in [node] + node.getchildren(): - if child.tail: if child.tail.strip(): self._processElementText(node, child, False) - if child.text: if child.text.strip(): self._processElementText(child, child) - else: # it's just a string - linkText(node) strartIndex = phEndIndex continue @@ -1749,23 +1701,22 @@ class Markdown: linkText(data[strartIndex:end]) strartIndex = end else: - text = data[strartIndex:] linkText(text) data = "" return result - def applyInlinePatterns(self, markdownTree): """ - Iterate over ElementTree, find elements with inline tag, - apply inline patterns and append newly created Elements to tree. - If you don't whant process your data with inline paterns, - instead of normal string, use subclass AtomicString: - `node.text = AtomicString("data won't be processed with inline patterns")` + Iterate over ElementTree, find elements with inline tag, apply inline + patterns and append newly created Elements to tree. If you don't + want process your data with inline paterns, instead of normal string, + use subclass AtomicString: + + node.text = AtomicString("data won't be processed with inline patterns") - Keyword arguments: + Arguments: * markdownTree: ElementTree object, representing Markdown tree. @@ -1829,27 +1780,42 @@ class Markdown: for pp in self.textPreprocessors: self.source = pp.run(self.source) - - markdownTree = self._transform() - - return markdownTree + + # Split into lines and run the preprocessors that will work with + # self.lines + self.lines = self.source.split("\n") + for prep in self.preprocessors : + self.lines = prep.run(self.lines) + + # Create a ElementTree from the lines + self.root = etree.Element("div") + buffer = [] + for line in self.lines: + if line.startswith("#"): + self._processSection(self.root, buffer) + buffer = [line] + else: + buffer.append(line) + + self._processSection(self.root, buffer) + + return etree.ElementTree(self.root) + def convert (self, source): - """Create the document in XHTML format. + """Convert markdown to serialized XHTML. Keyword arguments: * source: An ascii or unicode string of Markdown formated text. - Returns: A serialized XHTML body. - """ self.source = source if not self.source: - return u"" + return u"" # a blank unicode string + # Build a tree from the Markdown source and get its root. tree = self.markdownToTree(source) - root = self.applyInlinePatterns(tree).getroot() # Run the post-processors @@ -1859,8 +1825,8 @@ class Markdown: if newRoot: root = newRoot + # Serialize _properly_. Strip top-level tags. xml, length = codecs.utf_8_decode(etree.tostring(root, encoding="utf8")) - if self.stripTopLevelTags: xml = xml.strip()[44:-7] + "\n" |