##############################################################################
#
# Copyright (c) 2001 Zope Corporation and Contributors. All Rights Reserved.
#
# This software is subject to the provisions of the Zope Public License,
# Version 2.0 (ZPL). A copy of the ZPL should accompany this distribution.
# THIS SOFTWARE IS PROVIDED "AS IS" AND ANY AND ALL EXPRESS OR IMPLIED
# WARRANTIES ARE DISCLAIMED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
# WARRANTIES OF TITLE, MERCHANTABILITY, AGAINST INFRINGEMENT, AND FITNESS
# FOR A PARTICULAR PURPOSE
#
##############################################################################
import re, ST, STDOM
from STletters import letters
from types import StringType,UnicodeType,ListType
StringTypes = (StringType, UnicodeType)
class StructuredTextExample(ST.StructuredTextParagraph):
"""Represents a section of document with literal text, as for examples"""
def __init__(self, subs, **kw):
t=[]; a=t.append
for s in subs: a(s.getNodeValue())
apply(ST.StructuredTextParagraph.__init__,
(self, '\n\n'.join(t), ()),
kw)
def getColorizableTexts(self): return ()
def setColorizableTexts(self, src): pass # never color examples
class StructuredTextBullet(ST.StructuredTextParagraph):
"""Represents a section of a document with a title and a body"""
class StructuredTextNumbered(ST.StructuredTextParagraph):
"""Represents a section of a document with a title and a body"""
class StructuredTextDescriptionTitle(ST.StructuredTextParagraph):
"""Represents a section of a document with a title and a body"""
class StructuredTextDescriptionBody(ST.StructuredTextParagraph):
"""Represents a section of a document with a title and a body"""
class StructuredTextDescription(ST.StructuredTextParagraph):
"""Represents a section of a document with a title and a body"""
def __init__(self, title, src, subs, **kw):
apply(ST.StructuredTextParagraph.__init__, (self, src, subs), kw)
self._title=title
def getColorizableTexts(self): return self._title, self._src
def setColorizableTexts(self, src): self._title, self._src = src
def getChildren(self):
return (StructuredTextDescriptionTitle(self._title),
StructuredTextDescriptionBody(self._src, self._subs))
class StructuredTextSectionTitle(ST.StructuredTextParagraph):
"""Represents a section of a document with a title and a body"""
class StructuredTextSection(ST.StructuredTextParagraph):
"""Represents a section of a document with a title and a body"""
def __init__(self, src, subs=None, **kw):
apply(ST.StructuredTextParagraph.__init__,
(self, StructuredTextSectionTitle(src), subs),
kw)
def getColorizableTexts(self):
return self._src.getColorizableTexts()
def setColorizableTexts(self,src):
self._src.setColorizableTexts(src)
# a StructuredTextTable holds StructuredTextRows
class StructuredTextTable(ST.StructuredTextDocument):
"""
rows is a list of lists containing tuples, which
represent the columns/cells in each rows.
EX
rows = [[('row 1:column1',1)],[('row2:column1',1)]]
"""
def __init__(self, rows, src, subs, **kw):
apply(ST.StructuredTextDocument.__init__,(self,subs),kw)
self._rows = []
for row in rows:
if row:
self._rows.append(StructuredTextRow(row,kw))
def getRows(self):
return [self._rows]
def _getRows(self):
return self.getRows()
def getColorizableTexts(self):
"""
return a tuple where each item is a column/cell's
contents. The tuple, result, will be of this format.
("r1 col1", "r1=col2", "r2 col1", "r2 col2")
"""
#result = ()
result = []
for row in self._rows:
for column in row.getColumns()[0]:
#result = result[:] + (column.getColorizableTexts(),)
result.append(column.getColorizableTexts()[0])
return result
def setColorizableTexts(self,texts):
"""
texts is going to a tuple where each item is the
result of being mapped to the colortext function.
Need to insert the results appropriately into the
individual columns/cells
"""
for row_index in range(len(self._rows)):
for column_index in range(len(self._rows[row_index]._columns)):
self._rows[row_index]._columns[column_index].setColorizableTexts((texts[0],))
texts = texts[1:]
def _getColorizableTexts(self):
return self.getColorizableTexts()
def _setColorizableTexts(self):
return self.setColorizableTexts()
# StructuredTextRow holds StructuredTextColumns
class StructuredTextRow(ST.StructuredTextDocument):
def __init__(self,row,kw):
"""
row is a list of tuples, where each tuple is
the raw text for a cell/column and the span
of that cell/column.
EX
[('this is column one',1), ('this is column two',1)]
"""
apply(ST.StructuredTextDocument.__init__,(self,[]),kw)
self._columns = []
for column in row:
self._columns.append(StructuredTextColumn(column[0],column[1],kw))
def getColumns(self):
return [self._columns]
def _getColumns(self):
return [self._columns]
# this holds the raw text of a table cell
class StructuredTextColumn(ST.StructuredTextParagraph):
"""
StructuredTextColumn is a cell/column in a table.
This contains the actual text of a column and is
thus a StructuredTextParagraph. A StructuredTextColumn
also holds the span of its column
"""
def __init__(self,text,span,kw):
apply(ST.StructuredTextParagraph.__init__,(self,text,[]),kw)
self._span = span
def getSpan(self):
return self._span
def _getSpan(self):
return self._span
class StructuredTextMarkup(STDOM.Element):
def __init__(self, v, **kw):
self._value=v
self._attributes=kw.keys()
for k, v in kw.items(): setattr(self, k, v)
def getChildren(self, type=type, lt=type([])):
v=self._value
if type(v) is not lt: v=[v]
return v
def getColorizableTexts(self): return self._value,
def setColorizableTexts(self, v): self._value=v[0]
def __repr__(self):
return '%s(%s)' % (self.__class__.__name__, `self._value`)
class StructuredTextLiteral(StructuredTextMarkup):
def getColorizableTexts(self): return ()
def setColorizableTexts(self, v): pass
class StructuredTextEmphasis(StructuredTextMarkup): pass
class StructuredTextStrong(StructuredTextMarkup): pass
class StructuredTextInnerLink(StructuredTextMarkup): pass
class StructuredTextNamedLink(StructuredTextMarkup): pass
class StructuredTextUnderline(StructuredTextMarkup): pass
class StructuredTextLink(StructuredTextMarkup):
"A simple hyperlink"
class DocumentClass:
"""
Class instance calls [ex.=> x()] require a structured text
structure. Doc will then parse each paragraph in the structure
and will find the special structures within each paragraph.
Each special structure will be stored as an instance. Special
structures within another special structure are stored within
the 'top' structure
EX : '-underline this-' => would be turned into an underline
instance. '-underline **this**' would be stored as an underline
instance with a strong instance stored in its string
"""
paragraph_types = [
'doc_bullet',
'doc_numbered',
'doc_description',
'doc_header',
'doc_table',
]
text_types = [
'doc_href1',
'doc_href2',
'doc_strong',
'doc_emphasize',
'doc_literal',
'doc_inner_link',
'doc_named_link',
'doc_underline',
]
def __call__(self, doc):
if type(doc) in StringTypes:
doc=ST.StructuredText(doc)
doc.setSubparagraphs(self.color_paragraphs(
doc.getSubparagraphs()))
else:
doc=ST.StructuredTextDocument(self.color_paragraphs(
doc.getSubparagraphs()))
return doc
def parse(self, raw_string, text_type,
type=type, sts=StringTypes, lt=type([])):
"""
Parse accepts a raw_string, an expr to test the raw_string,
and the raw_string's subparagraphs.
Parse will continue to search through raw_string until
all instances of expr in raw_string are found.
If no instances of expr are found, raw_string is returned.
Otherwise a list of substrings and instances is returned
"""
tmp = [] # the list to be returned if raw_string is split
append=tmp.append
if type(text_type) in sts: text_type=getattr(self, text_type)
while 1:
t = text_type(raw_string)
if not t: break
#an instance of expr was found
t, start, end = t
if start: append(raw_string[0:start])
tt=type(t)
if tt in sts:
# if we get a string back, add it to text to be parsed
raw_string = t+raw_string[end:len(raw_string)]
else:
if tt is lt:
# is we get a list, append it's elements
tmp[len(tmp):]=t
else:
# normal case, an object
append(t)
raw_string = raw_string[end:len(raw_string)]
if not tmp: return raw_string # nothing found
if raw_string: append(raw_string)
elif len(tmp)==1: return tmp[0]
return tmp
def color_text(self, str, types=None):
"""Search the paragraph for each special structure
"""
if types is None: types=self.text_types
for text_type in types:
if type(str) in StringTypes:
str = self.parse(str, text_type)
elif type(str) is ListType:
r=[]; a=r.append
for s in str:
if type(s) in StringTypes:
s=self.parse(s, text_type)
if type(s) is ListType: r[len(r):]=s
else: a(s)
else:
s.setColorizableTexts(
map(self.color_text,
s.getColorizableTexts()
))
a(s)
str=r
else:
r=[]; a=r.append; color=self.color_text
for s in str.getColorizableTexts():
color(s, (text_type,))
a(s)
str.setColorizableTexts(r)
return str
def color_paragraphs(self, raw_paragraphs,
type=type, sequence_types=(type([]), type(())),
sts=StringTypes):
result=[]
for paragraph in raw_paragraphs:
if paragraph.getNodeName() != 'StructuredTextParagraph':
result.append(paragraph)
continue
for pt in self.paragraph_types:
if type(pt) in sts:
# grab the corresponding function
pt=getattr(self, pt)
# evaluate the paragraph
r=pt(paragraph)
if r:
if type(r) not in sequence_types:
r=r,
new_paragraphs=r
for paragraph in new_paragraphs:
paragraph.setSubparagraphs(self.color_paragraphs(paragraph.getSubparagraphs()))
break
else:
new_paragraphs=ST.StructuredTextParagraph(paragraph.getColorizableTexts()[0],
self.color_paragraphs(paragraph.getSubparagraphs()),
indent=paragraph.indent),
# color the inline StructuredText types
# for each StructuredTextParagraph
for paragraph in new_paragraphs:
paragraph.setColorizableTexts(
map(self.color_text,
paragraph.getColorizableTexts()
))
result.append(paragraph)
return result
def doc_table(self,paragraph, expr = re.compile('(\s*)([||]+)').match):
#print "paragraph=>", type(paragraph), paragraph, paragraph._src
text = paragraph.getColorizableTexts()[0]
m = expr(text)
if not (m):
return None
rows = []
# initial split
for row in text.split("\n"):
rows.append(row)
# clean up the rows
for index in range(len(rows)):
tmp = []
rows[index] = rows[index].strip()
l = len(rows[index])-2
result = rows[index][:l].split("||")
for text in result:
if text:
tmp.append(text)
tmp.append('')
else:
tmp.append(text)
rows[index] = tmp
# remove trailing '''s
for index in range(len(rows)):
l = len(rows[index])-1
rows[index] = rows[index][:l]
result = []
for row in rows:
cspan = 0
tmp = []
for item in row:
if item:
tmp.append((item,cspan))
cspan = 0
else:
cspan = cspan + 1
result.append(tmp)
subs = paragraph.getSubparagraphs()
# indent=paragraph.indent
return StructuredTextTable(result,text,subs,indent=paragraph.indent)
def doc_bullet(self, paragraph, expr = re.compile('\s*[-*o]\s+').match):
top=paragraph.getColorizableTexts()[0]
m=expr(top)
if not m:
return None
subs=paragraph.getSubparagraphs()
if top[-2:]=='::':
subs=[StructuredTextExample(subs)]
top=top[:-1]
return StructuredTextBullet(top[m.span()[1]:], subs,
indent=paragraph.indent,
bullet=top[:m.span()[1]]
)
def doc_numbered(
self, paragraph,
expr = re.compile('(\s*[%s]+\.)|(\s*[0-9]+\.)|(\s*[0-9]+\s+)' % letters).match):
# This is the old expression. It had a nasty habit
# of grabbing paragraphs that began with a single
# letter word even if there was no following period.
#expr = re.compile('\s*'
# '(([a-zA-Z]|[0-9]+|[ivxlcdmIVXLCDM]+)\.)*'
# '([a-zA-Z]|[0-9]+|[ivxlcdmIVXLCDM]+)\.?'
# '\s+').match):
top=paragraph.getColorizableTexts()[0]
m=expr(top)
if not m: return None
subs=paragraph.getSubparagraphs()
if top[-2:]=='::':
subs=[StructuredTextExample(subs)]
top=top[:-1]
return StructuredTextNumbered(top[m.span()[1]:], subs,
indent=paragraph.indent,
number=top[:m.span()[1]])
def doc_description(
self, paragraph,
delim = re.compile('\s+--\s+').search,
nb=re.compile(r'[^\000- ]').search,
):
top=paragraph.getColorizableTexts()[0]
d=delim(top)
if not d: return None
start, end = d.span()
title=top[:start]
if title.find('\n') >= 0: return None
if not nb(title): return None
d=top[start:end]
top=top[end:]
subs=paragraph.getSubparagraphs()
if top[-2:]=='::':
subs=[StructuredTextExample(subs)]
top=top[:-1]
return StructuredTextDescription(
title, top, subs,
indent=paragraph.indent,
delim=d)
def doc_header(self, paragraph,
expr = re.compile('[ %s0-9.:/,-_*<>\?\'\"]+' % letters).match
):
subs=paragraph.getSubparagraphs()
if not subs: return None
top=paragraph.getColorizableTexts()[0]
if not top.strip(): return None
if top[-2:]=='::':
subs=StructuredTextExample(subs)
if top.strip()=='::': return subs
return ST.StructuredTextParagraph(top[:-1],
[subs],
indent=paragraph.indent,
level=paragraph.level)
if top.find('\n') >= 0: return None
return StructuredTextSection(top, subs, indent=paragraph.indent, level=paragraph.level)
def doc_literal(
self, s,
expr=re.compile(
"(?:\s|^)'" # open
"([^ \t\n\r\f\v']|[^ \t\n\r\f\v'][^\n']*[^ \t\n\r\f\v'])" # contents
"'(?:\s|[,.;:!?]|$)" # close
).search):
r=expr(s)
if r:
start, end = r.span(1)
return (StructuredTextLiteral(s[start:end]), start-1, end+1)
else:
return None
def doc_emphasize(
self, s,
expr = re.compile('\s*\*([ \n%s0-9.:/;,\'\"\?\=\-\>\<\(\)]+)\*(?!\*|-)' % letters).search
):
r=expr(s)
if r:
start, end = r.span(1)
return (StructuredTextEmphasis(s[start:end]), start-1, end+1)
else:
return None
def doc_inner_link(self,
s,
expr1 = re.compile("\.\.\s*").search,
expr2 = re.compile("\[[%s0-9]+\]" % letters).search):
# make sure we dont grab a named link
if expr2(s) and expr1(s):
start1,end1 = expr1(s).span()
start2,end2 = expr2(s).span()
if end1 == start2:
# uh-oh, looks like a named link
return None
else:
# the .. is somewhere else, ignore it
return (StructuredTextInnerLink(s[start2+1,end2-1]),start2,end2)
return None
elif expr2(s) and not expr1(s):
start,end = expr2(s).span()
return (StructuredTextInnerLink(s[start+1:end-1]),start,end)
return None
def doc_named_link(self,
s,
expr=re.compile(r"(\.\.\s)(\[[%s0-9]+\])" % letters).search):
result = expr(s)
if result:
start,end = result.span(2)
str = s[start+1:end-1]
st,en = result.span()
return (StructuredTextNamedLink(str),st,en)
return None
def doc_underline(self,
s,
expr=re.compile("_([%s0-9\s\.,\?\/]+)_" % letters).search):
result = expr(s)
if result:
start,end = result.span(1)
st,e = result.span()
return (StructuredTextUnderline(s[start:end]),st,e)
else:
return None
def doc_strong(self,
s,
expr = re.compile('\s*\*\*([ \n%s0-9.:/;\-,!\?\'\"]+)\*\*' % letters).search
):
r=expr(s)
if r:
start, end = r.span(1)
return (StructuredTextStrong(s[start:end]), start-2, end+2)
else:
return None
def doc_href1(self, s,
expr=re.compile("(\"[ %s0-9\n\-\.\,\;\(\)\/\:\/\*\']+\")(:)([a-zA-Z0-9\@\.\,\?\!\/\:\;\-\#\~]+)([,]*\s*)" % letters).search
):
return self.doc_href(s, expr)
def doc_href2(self, s,
expr=re.compile('(\"[ %s0-9\n\-\.\:\;\(\)\/\*\']+\")([,]+\s+)([a-zA-Z0-9\@\.\,\?\!\/\:\;\-\#\~]+)(\s*)' % letters).search
):
return self.doc_href(s, expr)
def doc_href(self, s, expr, punctuation = re.compile("[\,\.\?\!\;]+").match):
r=expr(s)
if r:
# need to grab the href part and the
# beginning part
start,e = r.span(1)
name = s[start:e]
name = name.replace('"','',2)
#start = start + 1
st,end = r.span(3)
if punctuation(s[end-1:end]):
end = end -1
link = s[st:end]
#end = end - 1
# name is the href title, link is the target
# of the href
return (StructuredTextLink(name, href=link),
start, end)
#return (StructuredTextLink(s[start:end], href=s[start:end]),
# start, end)
else:
return None
|