aboutsummaryrefslogtreecommitdiffstats
path: root/markdown.py
diff options
context:
space:
mode:
authorArtem Yunusov <nedrlab@gmail.com>2008-08-01 02:58:48 +0500
committerArtem Yunusov <nedrlab@gmail.com>2008-08-01 02:58:48 +0500
commit13d25f86f61bbce247a12b845c3675b57274a13e (patch)
treecb8affc75b6715b87cda1f6a495a66a48ef4e79e /markdown.py
parentbd185087dc899b6157f821fdd29561caffcc4682 (diff)
downloadmarkdown-13d25f86f61bbce247a12b845c3675b57274a13e.tar.gz
markdown-13d25f86f61bbce247a12b845c3675b57274a13e.tar.bz2
markdown-13d25f86f61bbce247a12b845c3675b57274a13e.zip
Some cleanups.
Diffstat (limited to 'markdown.py')
-rwxr-xr-xmarkdown.py78
1 files changed, 36 insertions, 42 deletions
diff --git a/markdown.py b/markdown.py
index 3891ac3..12fd70d 100755
--- a/markdown.py
+++ b/markdown.py
@@ -63,8 +63,8 @@ def isstr(s):
return isinstance(s, unicode) or isinstance(s, str)
def importETree():
- """ Importing best variant of ElementTree
- and returning module object """
+ """ Imports best variant of ElementTree
+ and returns module object """
try:
# Python 2.5+
@@ -527,10 +527,6 @@ else:
LINK_RE = NOIMG + BRK + \
r'''\(\s*(<.*?>|((?:(?:\(.*?\))|[^\(\)]))*?)\s*((['"])(.*)\12)?\)''' # [text](url) or [text](<url>)
-#r'''\(\s*(<.*?>|((?:(?:\(.*?\))|[^\(\)]))*)\s*((['"])(.*)\12)?\)''' # [text](url) or [text](<url>)
-
-#LINK_RE2 = NOIMG + BRK + \
-#r'''\(\s*(((?:(?:\(.*?\))|[^\(\)]))*?|<\10*?>)\s*((['"])(.*)\12)?\)''' # [text](url) or [text](<url>)
IMAGE_LINK_RE = r'\!' + BRK + r'\s*\((<.*?>|([^\)]*))\)' # ![alttxt](http://x.com/) or ![alttxt](<http://x.com/>)
REFERENCE_RE = NOIMG + BRK+ r'\s*\[([^\]]*)\]' # [Google][3]
@@ -1116,7 +1112,6 @@ class CorePatterns:
'isline2': r'(\-*)', # ---
'isline3': r'(\_*)', # ___
'tabbed': r'((\t)|( ))(.*)', # an indented line
- 'tabbed2': r'(()( ))(.*)', # an indented line
'quoted': r'[ ]{0,2}> ?(.*)', # a quoted block ("> ...")
}
@@ -1393,32 +1388,32 @@ class Markdown:
if lines and not lines[0].strip():
lines = lines[1:] # skip the first (blank) line
- def _processHR(self, parent_elem):
- hr = etree.SubElement(parent_elem, "hr")
+ def _processHR(self, parentElem):
+ hr = etree.SubElement(parentElem, "hr")
- def _processHeader(self, parent_elem, paragraph):
+ def _processHeader(self, parentElem, paragraph):
m = RE.regExp['header'].match(paragraph[0])
if m:
level = len(m.group(1))
- h = etree.SubElement(parent_elem, "h%d" % level)
+ h = etree.SubElement(parentElem, "h%d" % level)
inline = etree.SubElement(h, "inline")
inline.text = m.group(2).strip()
else:
message(CRITICAL, "We've got a problem header!")
- def _processParagraph(self, parent_elem, paragraph, inList, looseList):
+ def _processParagraph(self, parentElem, paragraph, inList, looseList):
- if ( parent_elem.tag == 'li'
- and not (looseList or parent_elem.getchildren())):
+ if ( parentElem.tag == 'li'
+ and not (looseList or parentElem.getchildren())):
# If this is the first paragraph inside "li", don't
# put <p> around it - append the paragraph bits directly
- # onto parent_elem
- el = parent_elem
+ # onto parentElem
+ el = parentElem
else:
# Otherwise make a "p" element
- el = etree.SubElement(parent_elem, "p")
+ el = etree.SubElement(parentElem, "p")
dump = []
@@ -1436,16 +1431,16 @@ class Markdown:
inline = etree.SubElement(el, "inline")
inline.text = text
- def _processUList(self, parent_elem, lines, inList):
- self._processList(parent_elem, lines, inList,
+ def _processUList(self, parentElem, lines, inList):
+ self._processList(parentElem, lines, inList,
listexpr='ul', tag = 'ul')
- def _processOList(self, parent_elem, lines, inList):
- self._processList(parent_elem, lines, inList,
+ def _processOList(self, parentElem, lines, inList):
+ self._processList(parentElem, lines, inList,
listexpr='ol', tag = 'ol')
- def _processList(self, parent_elem, lines, inList, listexpr, tag):
+ def _processList(self, parentElem, lines, inList, listexpr, tag):
"""
Given a list of document lines starting with a list item,
finds the end of the list, breaks it up, and recursively
@@ -1453,7 +1448,7 @@ class Markdown:
Keyword arguments:
- * parent_elem: A ElementTree element to which the content will be added
+ * parentElem: A ElementTree element to which the content will be added
* lines: a list of lines
* inList: a level
@@ -1461,7 +1456,7 @@ class Markdown:
"""
- ul = etree.SubElement(parent_elem, tag) # ul might actually be '<ol>'
+ ul = etree.SubElement(parentElem, tag) # ul might actually be '<ol>'
looseList = 0
@@ -1502,7 +1497,7 @@ class Markdown:
# Now we need to detect list items (at the current level)
# while also detabing child elements if necessary
- for expr in ['ul', 'ol', 'tabbed', 'tabbed2']:
+ for expr in ['ul', 'ol', 'tabbed']:
m = RE.regExp[expr].match(line)
if m:
@@ -1512,7 +1507,7 @@ class Markdown:
# at the beginning of the list item
items.append([m.group(1)])
item += 1
- elif expr == 'tabbed' or expr == 'tabbed2': # This line needs to be detabbed
+ elif expr == 'tabbed': # This line needs to be detabbed
items[item].append(m.group(4)) #after the 'tab'
i += 1
@@ -1531,7 +1526,7 @@ class Markdown:
# Process the remaining part of the section
- self._processSection(parent_elem, lines[i:], inList)
+ self._processSection(parentElem, lines[i:], inList)
def _linesUntil(self, lines, condition):
@@ -1551,7 +1546,7 @@ class Markdown:
i += 1
return lines[:i], lines[i:]
- def _processQuote(self, parent_elem, lines, inList):
+ def _processQuote(self, parentElem, lines, inList):
"""
Given a list of document lines starting with a quote finds
the end of the quote, unindents it and recursively
@@ -1560,7 +1555,7 @@ class Markdown:
Keyword arguments:
- * parent_elem: ElementTree element to which the content will be added
+ * parentElem: ElementTree element to which the content will be added
* lines: a list of lines
* inList: a level
@@ -1587,15 +1582,15 @@ class Markdown:
else:
break
- blockquote = etree.SubElement(parent_elem, "blockquote")
+ blockquote = etree.SubElement(parentElem, "blockquote")
self._processSection(blockquote, dequoted, inList)
- self._processSection(parent_elem, lines[i:], inList)
+ self._processSection(parentElem, lines[i:], inList)
- def _processCodeBlock(self, parent_elem, lines, inList):
+ def _processCodeBlock(self, parentElem, lines, inList):
"""
Given a list of document lines starting with a code block
finds the end of the block, puts it into the ElementTree verbatim
@@ -1604,7 +1599,7 @@ class Markdown:
Keyword arguments:
- * parent_elem: ElementTree element to which the content will be added
+ * parentElem: ElementTree element to which the content will be added
* lines: a list of lines
* inList: a level
@@ -1614,13 +1609,12 @@ class Markdown:
detabbed, theRest = self.blockGuru.detectTabbed(lines)
- pre = etree.SubElement(parent_elem, "pre")
+ pre = etree.SubElement(parentElem, "pre")
code = etree.SubElement(pre, "code")
text = "\n".join(detabbed).rstrip()+"\n"
- #text = text.replace("&", "&amp;")
code.text = text
- self._processSection(parent_elem, theRest, inList)
+ self._processSection(parentElem, theRest, inList)
def _handleInline(self, data, patternIndex=0):
"""
@@ -1791,7 +1785,7 @@ class Markdown:
Keyword arguments:
- * el - parent element of Document.
+ * el - parent element of ElementTree.
Returns: ElementTree object with applied inline patterns.
"""
@@ -1843,9 +1837,7 @@ class Markdown:
Returns: ElementTree object.
"""
-
-
-
+
el = markdownTree.getroot()
self._processTree(el)
@@ -1855,8 +1847,9 @@ class Markdown:
def markdownToTree(self, source=None):
"""
- Retrun ElementTree, without applying
- inline paterns
+ Retrun ElementTree, without applying inline paterns,
+ all data, that should be processed with
+ inline patterns included in <inline></inline> sections.
Keyword arguments:
@@ -1920,6 +1913,7 @@ class Markdown:
if self.stripTopLevelTags:
xml = xml.strip()[44:-7] + "\n"
+ # Run the text post-processors
for pp in self.textPostprocessors:
xml = pp.run(xml)