openmedialibrary_platform/Linux/lib/python2.7/site-packages/twisted/python/text.py

209 lines
5.3 KiB
Python
Raw Normal View History

2013-10-11 17:28:32 +00:00
# -*- test-case-name: twisted.test.test_text -*-
#
# Copyright (c) Twisted Matrix Laboratories.
# See LICENSE for details.
"""
Miscellany of text-munging functions.
"""
def stringyString(object, indentation=''):
"""
Expansive string formatting for sequence types.
C{list.__str__} and C{dict.__str__} use C{repr()} to display their
elements. This function also turns these sequence types
into strings, but uses C{str()} on their elements instead.
Sequence elements are also displayed on seperate lines, and nested
sequences have nested indentation.
"""
braces = ''
sl = []
if type(object) is dict:
braces = '{}'
for key, value in object.items():
value = stringyString(value, indentation + ' ')
if isMultiline(value):
if endsInNewline(value):
value = value[:-len('\n')]
sl.append("%s %s:\n%s" % (indentation, key, value))
else:
# Oops. Will have to move that indentation.
sl.append("%s %s: %s" % (indentation, key,
value[len(indentation) + 3:]))
elif type(object) is tuple or type(object) is list:
if type(object) is tuple:
braces = '()'
else:
braces = '[]'
for element in object:
element = stringyString(element, indentation + ' ')
sl.append(element.rstrip() + ',')
else:
sl[:] = map(lambda s, i=indentation: i + s,
str(object).split('\n'))
if not sl:
sl.append(indentation)
if braces:
sl[0] = indentation + braces[0] + sl[0][len(indentation) + 1:]
sl[-1] = sl[-1] + braces[-1]
s = "\n".join(sl)
if isMultiline(s) and not endsInNewline(s):
s = s + '\n'
return s
def isMultiline(s):
"""
Returns C{True} if this string has a newline in it.
"""
return (s.find('\n') != -1)
def endsInNewline(s):
"""
Returns C{True} if this string ends in a newline.
"""
return (s[-len('\n'):] == '\n')
def greedyWrap(inString, width=80):
"""
Given a string and a column width, return a list of lines.
Caveat: I'm use a stupid greedy word-wrapping
algorythm. I won't put two spaces at the end
of a sentence. I don't do full justification.
And no, I've never even *heard* of hypenation.
"""
outLines = []
#eww, evil hacks to allow paragraphs delimited by two \ns :(
if inString.find('\n\n') >= 0:
paragraphs = inString.split('\n\n')
for para in paragraphs:
outLines.extend(greedyWrap(para, width) + [''])
return outLines
inWords = inString.split()
column = 0
ptr_line = 0
while inWords:
column = column + len(inWords[ptr_line])
ptr_line = ptr_line + 1
if (column > width):
if ptr_line == 1:
# This single word is too long, it will be the whole line.
pass
else:
# We've gone too far, stop the line one word back.
ptr_line = ptr_line - 1
(l, inWords) = (inWords[0:ptr_line], inWords[ptr_line:])
outLines.append(' '.join(l))
ptr_line = 0
column = 0
elif not (len(inWords) > ptr_line):
# Clean up the last bit.
outLines.append(' '.join(inWords))
del inWords[:]
else:
# Space
column = column + 1
# next word
return outLines
wordWrap = greedyWrap
def removeLeadingBlanks(lines):
ret = []
for line in lines:
if ret or line.strip():
ret.append(line)
return ret
def removeLeadingTrailingBlanks(s):
lines = removeLeadingBlanks(s.split('\n'))
lines.reverse()
lines = removeLeadingBlanks(lines)
lines.reverse()
return '\n'.join(lines)+'\n'
def splitQuoted(s):
"""
Like a string split, but don't break substrings inside quotes.
>>> splitQuoted('the "hairy monkey" likes pie')
['the', 'hairy monkey', 'likes', 'pie']
Another one of those "someone must have a better solution for
this" things. This implementation is a VERY DUMB hack done too
quickly.
"""
out = []
quot = None
phrase = None
for word in s.split():
if phrase is None:
if word and (word[0] in ("\"", "'")):
quot = word[0]
word = word[1:]
phrase = []
if phrase is None:
out.append(word)
else:
if word and (word[-1] == quot):
word = word[:-1]
phrase.append(word)
out.append(" ".join(phrase))
phrase = None
else:
phrase.append(word)
return out
def strFile(p, f, caseSensitive=True):
"""
Find whether string C{p} occurs in a read()able object C{f}.
@rtype: C{bool}
"""
buf = ""
buf_len = max(len(p), 2**2**2**2)
if not caseSensitive:
p = p.lower()
while 1:
r = f.read(buf_len-len(p))
if not caseSensitive:
r = r.lower()
bytes_read = len(r)
if bytes_read == 0:
return False
l = len(buf)+bytes_read-buf_len
if l <= 0:
buf = buf + r
else:
buf = buf[l:] + r
if buf.find(p) != -1:
return True