+"""String utility functions for collation."""
+
+__all__ = ["sortemes", "numeric", "normalize_number", "deroman"]
+
import unicodedata
CONTINUE_ON = frozenset([
UNKNOWN, LETTER, NUMBER = range(3)
-BREAKER = u"\u2029" # Paragraph break character
-
-def sortemes(string):
+BREAKER = u"\u2028" # Line break character
+HBREAKER = u"\u2029" # Paragraph break character
+INFINITY = float('inf')
+
+KEEP_IN_NUMBERS = u"'.,"
+ALLOWED_IN_NUMBERS = KEEP_IN_NUMBERS + u"_"
+
+ROMAN = {
+ u"i": 1,
+ u"v": 5,
+ u"x": 10,
+ u"l": 50,
+ u"c": 100,
+ u"d": 500,
+ u"m": 1000,
+ u"\u2180": 1000,
+ u"\u2181": 5000,
+ u"\u2182": 10000,
+ u"\u2183": 100,
+ u"\u2184": 100,
+ u"\u2185": 6,
+ u"\u2186": 50,
+ u"\u2187": 50000,
+ u"\u2188": 100000,
+ }
+
+def stripends(word):
+ """Strip punctuation and symbols from the ends of a string."""
+ while word and unicodedata.category(word[0])[0] in "PS":
+ word = word[1:]
+ while word and unicodedata.category(word[-1])[0] in "PS":
+ word = word[:-1]
+ return word
+
+def sortemes(string, key=lambda s: s):
"""Generate a list of sortemes for the string.
A sorteme, by analogy with grapheme/morpheme/etc. is an atom of
sort information. This is larger than a word boundry but smaller
than a sentence boundry; roughly, a sorteme boundry occurs between
- letters and numbers, between numbers and numbrs if 'too much'
+ letters and numbers, between numbers and numbers if 'too much'
punctuation exists in between, between lines.
There is no formal specification for sortemes; the goal of this
function is to provide good output for Collator.sortemekey.
+
"""
+ if not string:
+ return []
+
words = []
letters = []
digits = []
- if not string:
- return words
+ lappend = letters.append
+ dappend = digits.append
string = unicode(string)
categories = map(unicodedata.category, string)
previous = UNKNOWN
- types = []
-
- def stripends(word):
- while word and unicodedata.category(word[0])[0] in "PS":
- word = word[1:]
- while word and unicodedata.category(word[-1])[0] in "PS":
- word = word[:-1]
- return word
+ wappend = words.append
+ join = u"".join
+ i = 0
- # TODO(jfw): This kind of evolved over time, there's probably a much
- # faster / more concise way to express it now.
- for i, (c, category) in enumerate(zip(string, categories)):
+ for uchar in string:
+ category = categories[i]
if letters and previous == LETTER and words:
- word = stripends(words.pop().strip())
- letters = list(stripends(word).strip() + BREAKER) + letters
+ word = stripends(words.pop()[1].strip()) + BREAKER
+ letters.insert(0, word)
previous = UNKNOWN
# Split at the first letter following a number or
# non-continuing character.
if category[0] == "L":
- letters.append(c)
+ lappend(uchar)
if digits:
- words.append(u"".join(digits).strip())
+ words.append((numeric(join(digits).strip()), u''))
+ del(digits[:])
previous = NUMBER
- digits = []
# Split at the first number following a non-number or
# non-continuing character.
elif category[0] == "N":
- digits.append(c)
+ dappend(uchar)
if letters:
- words.append(u"".join(letters))
+ if unicodedata.category(letters[-1])[0] == "L":
+ lappend(HBREAKER)
+ wappend((INFINITY, stripends(join(letters))))
+ del(letters[:])
previous = LETTER
- letters = []
# Only certain punctuation allowed in numbers.
- elif digits and c not in "',._":
- words.append(u"".join(digits))
+ elif digits and uchar not in ALLOWED_IN_NUMBERS:
+ words.append((numeric(join(digits)), u''))
+ del(digits[:])
previous = NUMBER
- digits = []
# Split if we find a non-continuing character ("weird" ones).
- elif letters and category not in CONTINUE_ON:
+ elif category not in CONTINUE_ON:
if letters:
- words.append(u"".join(letters).strip() + BREAKER)
+ wappend(
+ (INFINITY,
+ stripends(join(letters).strip() + BREAKER)))
+ del(letters[:])
previous = LETTER
- letters = []
if digits:
- words.append(u"".join(digits).strip() + BREAKER)
+ words.append((numeric(join(digits)), u''))
+ del(digits[:])
previous = NUMBER
- digits = []
# Split if we find two pieces of punctuation in a row, even
# if we should otherwise continue.
- elif i and categories[i-1][0] in "P" and category[0] in "P":
+ elif i and categories[i - 1][0] == category[0] == "P":
if letters:
- words.append(u"".join(letters))
+ wappend((INFINITY, stripends(join(letters))))
+ del(letters[:])
previous = LETTER
- letters = []
if digits:
- words.append(u"".join(digits))
+ words.append((numeric(join(digits)), u''))
+ del(digits[:])
previous = NUMBER
- digits = []
else:
if digits:
- digits.append(c)
+ dappend(uchar)
elif letters:
- letters.append(c)
+ lappend(uchar)
+
+ i += 1
if letters and previous == LETTER and words:
- word = stripends(words.pop().strip())
- letters = list(stripends(word).strip() + BREAKER) + letters
+ word = stripends(words.pop()[1].strip()) + BREAKER
+ letters.insert(0, word)
previous = UNKNOWN
if letters:
- words.append(u"".join(letters))
- letters = []
+ wappend((INFINITY, stripends(join(letters))))
if digits:
- words.append(u"".join(digits))
- digits = []
+ words.append((numeric(join(digits)), u''))
- words = map(stripends, words)
- return words
+ return [(i, key(w)) for i, w in words]
+
+def numeric(orig, invalid=INFINITY):
+ """Parse a number out of a string.
+
+ This function parses a unicode number out of the start of a
+ string. If a number cannot be found at the start, the 'invalid'
+ argument is returned.
+
+ """
-def numeric(orig, invalid=float('inf')):
if not orig:
- return (invalid, '')
+ return invalid
string = unicode(orig)
- for c in string:
- if c.isnumeric():
+ for uchar in string:
+ if uchar.isnumeric():
break
else:
- return (invalid, orig)
+ return invalid
+
+ for char in string:
+ if u"\u2160" <= char <= u"\u2188":
+ return deroman(string)
mult = 1
while string[:1] == u"-" or string[:1] == u"+":
string = string[1:]
if not string[:1].isnumeric():
- return (invalid, orig)
-
- string = normalize_punc(string)
+ return invalid
- # Early out if possible.
- try:
- return (float(string) * mult, orig)
- except ValueError:
- pass
+ string = normalize_number(string)
- # Otherwise we need to do this the hard way.
def _numeric(string):
+ """Interpreter a number as base 10."""
total = 0
- for c in string:
- v = unicodedata.numeric(c)
- if v >= 1 or v == 0:
+ for uchar in string:
+ number = unicodedata.numeric(uchar)
+ if number >= 1 or number == 0:
total *= 10
- total += v
+ total += number
return total
try:
whole, frac = string.split(".")
whole = _numeric(whole)
frac = _numeric(frac) / (10.0 ** len(frac))
- return (mult * (whole + frac), orig)
+ return mult * (whole + frac)
except ValueError:
- return (mult * _numeric(string), orig)
+ return mult * _numeric(string)
+
+def normalize_number(string):
+ """Normalize punctuation in a number.
+
+ This function attempts to guess which characters in a number
+ represent grouping separators and which represent decimal
+ points. It returns a string that is valid to pass to Python's
+ float() routine (potentially, NaN, if nothing like a number is
+ found).
+
+ """
+
+ string = unicode(string)
+ string = filter(lambda u: u.isnumeric() or u in KEEP_IN_NUMBERS, string)
+ string = string.strip(KEEP_IN_NUMBERS)
-def normalize_punc(string):
- string = unicode(string.strip(u",.'"))
- string = filter(lambda u: u.isnumeric() or u in u",.'", string)
commas = string.count(u",")
stops = string.count(u".")
quotes = string.count(u"'")
quotes = 0
def normalize_two(a, b, string):
- # One of each - assume the first is grouping, second is point.
+ """One of each - assume the first is grouping, second is point."""
a_idx = string.rindex(a)
b_idx = string.rindex(b)
if a_idx > b_idx:
# Single stop, but no decimal - probably grouping.
string = string.replace(u".", u"")
- return string
+ return string or "NaN"
+
+def deroman(string):
+ """Turn a Roman numeral into an integer."""
+ string = unicodedata.normalize('NFKD', unicode(string)).lower()
+ previous = 0
+ building = 0
+ for char in reversed(string):
+ try:
+ value = ROMAN[char]
+ except KeyError:
+ continue
+ if value < previous:
+ building -= value
+ else:
+ building += value
+ previous = value
+ return building