1 | # Originally contributed by Sjoerd Mullender.
|
---|
2 | # Significantly modified by Jeffrey Yasskin <jyasskin at gmail.com>.
|
---|
3 |
|
---|
4 | """Rational, infinite-precision, real numbers."""
|
---|
5 |
|
---|
6 | from __future__ import division
|
---|
7 | import math
|
---|
8 | import numbers
|
---|
9 | import operator
|
---|
10 | import re
|
---|
11 |
|
---|
12 | __all__ = ['Fraction', 'gcd']
|
---|
13 |
|
---|
14 | Rational = numbers.Rational
|
---|
15 |
|
---|
16 |
|
---|
17 | def gcd(a, b):
|
---|
18 | """Calculate the Greatest Common Divisor of a and b.
|
---|
19 |
|
---|
20 | Unless b==0, the result will have the same sign as b (so that when
|
---|
21 | b is divided by it, the result comes out positive).
|
---|
22 | """
|
---|
23 | while b:
|
---|
24 | a, b = b, a%b
|
---|
25 | return a
|
---|
26 |
|
---|
27 |
|
---|
28 | _RATIONAL_FORMAT = re.compile(r"""
|
---|
29 | \A\s* # optional whitespace at the start, then
|
---|
30 | (?P<sign>[-+]?) # an optional sign, then
|
---|
31 | (?=\d|\.\d) # lookahead for digit or .digit
|
---|
32 | (?P<num>\d*) # numerator (possibly empty)
|
---|
33 | (?: # followed by an optional
|
---|
34 | /(?P<denom>\d+) # / and denominator
|
---|
35 | | # or
|
---|
36 | \.(?P<decimal>\d*) # decimal point and fractional part
|
---|
37 | )?
|
---|
38 | \s*\Z # and optional whitespace to finish
|
---|
39 | """, re.VERBOSE)
|
---|
40 |
|
---|
41 |
|
---|
42 | class Fraction(Rational):
|
---|
43 | """This class implements rational numbers.
|
---|
44 |
|
---|
45 | Fraction(8, 6) will produce a rational number equivalent to
|
---|
46 | 4/3. Both arguments must be Integral. The numerator defaults to 0
|
---|
47 | and the denominator defaults to 1 so that Fraction(3) == 3 and
|
---|
48 | Fraction() == 0.
|
---|
49 |
|
---|
50 | Fractions can also be constructed from strings of the form
|
---|
51 | '[-+]?[0-9]+((/|.)[0-9]+)?', optionally surrounded by spaces.
|
---|
52 |
|
---|
53 | """
|
---|
54 |
|
---|
55 | __slots__ = ('_numerator', '_denominator')
|
---|
56 |
|
---|
57 | # We're immutable, so use __new__ not __init__
|
---|
58 | def __new__(cls, numerator=0, denominator=1):
|
---|
59 | """Constructs a Fraction.
|
---|
60 |
|
---|
61 | Takes a string like '3/2' or '1.5', another Fraction, or a
|
---|
62 | numerator/denominator pair.
|
---|
63 |
|
---|
64 | """
|
---|
65 | self = super(Fraction, cls).__new__(cls)
|
---|
66 |
|
---|
67 | if type(numerator) not in (int, long) and denominator == 1:
|
---|
68 | if isinstance(numerator, basestring):
|
---|
69 | # Handle construction from strings.
|
---|
70 | input = numerator
|
---|
71 | m = _RATIONAL_FORMAT.match(input)
|
---|
72 | if m is None:
|
---|
73 | raise ValueError('Invalid literal for Fraction: %r' % input)
|
---|
74 | numerator = m.group('num')
|
---|
75 | decimal = m.group('decimal')
|
---|
76 | if decimal:
|
---|
77 | # The literal is a decimal number.
|
---|
78 | numerator = int(numerator + decimal)
|
---|
79 | denominator = 10**len(decimal)
|
---|
80 | else:
|
---|
81 | # The literal is an integer or fraction.
|
---|
82 | numerator = int(numerator)
|
---|
83 | # Default denominator to 1.
|
---|
84 | denominator = int(m.group('denom') or 1)
|
---|
85 |
|
---|
86 | if m.group('sign') == '-':
|
---|
87 | numerator = -numerator
|
---|
88 |
|
---|
89 | elif isinstance(numerator, Rational):
|
---|
90 | # Handle copies from other rationals. Integrals get
|
---|
91 | # caught here too, but it doesn't matter because
|
---|
92 | # denominator is already 1.
|
---|
93 | other_rational = numerator
|
---|
94 | numerator = other_rational.numerator
|
---|
95 | denominator = other_rational.denominator
|
---|
96 |
|
---|
97 | if denominator == 0:
|
---|
98 | raise ZeroDivisionError('Fraction(%s, 0)' % numerator)
|
---|
99 | numerator = operator.index(numerator)
|
---|
100 | denominator = operator.index(denominator)
|
---|
101 | g = gcd(numerator, denominator)
|
---|
102 | self._numerator = numerator // g
|
---|
103 | self._denominator = denominator // g
|
---|
104 | return self
|
---|
105 |
|
---|
106 | @classmethod
|
---|
107 | def from_float(cls, f):
|
---|
108 | """Converts a finite float to a rational number, exactly.
|
---|
109 |
|
---|
110 | Beware that Fraction.from_float(0.3) != Fraction(3, 10).
|
---|
111 |
|
---|
112 | """
|
---|
113 | if isinstance(f, numbers.Integral):
|
---|
114 | return cls(f)
|
---|
115 | elif not isinstance(f, float):
|
---|
116 | raise TypeError("%s.from_float() only takes floats, not %r (%s)" %
|
---|
117 | (cls.__name__, f, type(f).__name__))
|
---|
118 | if math.isnan(f) or math.isinf(f):
|
---|
119 | raise TypeError("Cannot convert %r to %s." % (f, cls.__name__))
|
---|
120 | return cls(*f.as_integer_ratio())
|
---|
121 |
|
---|
122 | @classmethod
|
---|
123 | def from_decimal(cls, dec):
|
---|
124 | """Converts a finite Decimal instance to a rational number, exactly."""
|
---|
125 | from decimal import Decimal
|
---|
126 | if isinstance(dec, numbers.Integral):
|
---|
127 | dec = Decimal(int(dec))
|
---|
128 | elif not isinstance(dec, Decimal):
|
---|
129 | raise TypeError(
|
---|
130 | "%s.from_decimal() only takes Decimals, not %r (%s)" %
|
---|
131 | (cls.__name__, dec, type(dec).__name__))
|
---|
132 | if not dec.is_finite():
|
---|
133 | # Catches infinities and nans.
|
---|
134 | raise TypeError("Cannot convert %s to %s." % (dec, cls.__name__))
|
---|
135 | sign, digits, exp = dec.as_tuple()
|
---|
136 | digits = int(''.join(map(str, digits)))
|
---|
137 | if sign:
|
---|
138 | digits = -digits
|
---|
139 | if exp >= 0:
|
---|
140 | return cls(digits * 10 ** exp)
|
---|
141 | else:
|
---|
142 | return cls(digits, 10 ** -exp)
|
---|
143 |
|
---|
144 | def limit_denominator(self, max_denominator=1000000):
|
---|
145 | """Closest Fraction to self with denominator at most max_denominator.
|
---|
146 |
|
---|
147 | >>> Fraction('3.141592653589793').limit_denominator(10)
|
---|
148 | Fraction(22, 7)
|
---|
149 | >>> Fraction('3.141592653589793').limit_denominator(100)
|
---|
150 | Fraction(311, 99)
|
---|
151 | >>> Fraction(4321, 8765).limit_denominator(10000)
|
---|
152 | Fraction(4321, 8765)
|
---|
153 |
|
---|
154 | """
|
---|
155 | # Algorithm notes: For any real number x, define a *best upper
|
---|
156 | # approximation* to x to be a rational number p/q such that:
|
---|
157 | #
|
---|
158 | # (1) p/q >= x, and
|
---|
159 | # (2) if p/q > r/s >= x then s > q, for any rational r/s.
|
---|
160 | #
|
---|
161 | # Define *best lower approximation* similarly. Then it can be
|
---|
162 | # proved that a rational number is a best upper or lower
|
---|
163 | # approximation to x if, and only if, it is a convergent or
|
---|
164 | # semiconvergent of the (unique shortest) continued fraction
|
---|
165 | # associated to x.
|
---|
166 | #
|
---|
167 | # To find a best rational approximation with denominator <= M,
|
---|
168 | # we find the best upper and lower approximations with
|
---|
169 | # denominator <= M and take whichever of these is closer to x.
|
---|
170 | # In the event of a tie, the bound with smaller denominator is
|
---|
171 | # chosen. If both denominators are equal (which can happen
|
---|
172 | # only when max_denominator == 1 and self is midway between
|
---|
173 | # two integers) the lower bound---i.e., the floor of self, is
|
---|
174 | # taken.
|
---|
175 |
|
---|
176 | if max_denominator < 1:
|
---|
177 | raise ValueError("max_denominator should be at least 1")
|
---|
178 | if self._denominator <= max_denominator:
|
---|
179 | return Fraction(self)
|
---|
180 |
|
---|
181 | p0, q0, p1, q1 = 0, 1, 1, 0
|
---|
182 | n, d = self._numerator, self._denominator
|
---|
183 | while True:
|
---|
184 | a = n//d
|
---|
185 | q2 = q0+a*q1
|
---|
186 | if q2 > max_denominator:
|
---|
187 | break
|
---|
188 | p0, q0, p1, q1 = p1, q1, p0+a*p1, q2
|
---|
189 | n, d = d, n-a*d
|
---|
190 |
|
---|
191 | k = (max_denominator-q0)//q1
|
---|
192 | bound1 = Fraction(p0+k*p1, q0+k*q1)
|
---|
193 | bound2 = Fraction(p1, q1)
|
---|
194 | if abs(bound2 - self) <= abs(bound1-self):
|
---|
195 | return bound2
|
---|
196 | else:
|
---|
197 | return bound1
|
---|
198 |
|
---|
199 | @property
|
---|
200 | def numerator(a):
|
---|
201 | return a._numerator
|
---|
202 |
|
---|
203 | @property
|
---|
204 | def denominator(a):
|
---|
205 | return a._denominator
|
---|
206 |
|
---|
207 | def __repr__(self):
|
---|
208 | """repr(self)"""
|
---|
209 | return ('Fraction(%s, %s)' % (self._numerator, self._denominator))
|
---|
210 |
|
---|
211 | def __str__(self):
|
---|
212 | """str(self)"""
|
---|
213 | if self._denominator == 1:
|
---|
214 | return str(self._numerator)
|
---|
215 | else:
|
---|
216 | return '%s/%s' % (self._numerator, self._denominator)
|
---|
217 |
|
---|
218 | def _operator_fallbacks(monomorphic_operator, fallback_operator):
|
---|
219 | """Generates forward and reverse operators given a purely-rational
|
---|
220 | operator and a function from the operator module.
|
---|
221 |
|
---|
222 | Use this like:
|
---|
223 | __op__, __rop__ = _operator_fallbacks(just_rational_op, operator.op)
|
---|
224 |
|
---|
225 | In general, we want to implement the arithmetic operations so
|
---|
226 | that mixed-mode operations either call an implementation whose
|
---|
227 | author knew about the types of both arguments, or convert both
|
---|
228 | to the nearest built in type and do the operation there. In
|
---|
229 | Fraction, that means that we define __add__ and __radd__ as:
|
---|
230 |
|
---|
231 | def __add__(self, other):
|
---|
232 | # Both types have numerators/denominator attributes,
|
---|
233 | # so do the operation directly
|
---|
234 | if isinstance(other, (int, long, Fraction)):
|
---|
235 | return Fraction(self.numerator * other.denominator +
|
---|
236 | other.numerator * self.denominator,
|
---|
237 | self.denominator * other.denominator)
|
---|
238 | # float and complex don't have those operations, but we
|
---|
239 | # know about those types, so special case them.
|
---|
240 | elif isinstance(other, float):
|
---|
241 | return float(self) + other
|
---|
242 | elif isinstance(other, complex):
|
---|
243 | return complex(self) + other
|
---|
244 | # Let the other type take over.
|
---|
245 | return NotImplemented
|
---|
246 |
|
---|
247 | def __radd__(self, other):
|
---|
248 | # radd handles more types than add because there's
|
---|
249 | # nothing left to fall back to.
|
---|
250 | if isinstance(other, Rational):
|
---|
251 | return Fraction(self.numerator * other.denominator +
|
---|
252 | other.numerator * self.denominator,
|
---|
253 | self.denominator * other.denominator)
|
---|
254 | elif isinstance(other, Real):
|
---|
255 | return float(other) + float(self)
|
---|
256 | elif isinstance(other, Complex):
|
---|
257 | return complex(other) + complex(self)
|
---|
258 | return NotImplemented
|
---|
259 |
|
---|
260 |
|
---|
261 | There are 5 different cases for a mixed-type addition on
|
---|
262 | Fraction. I'll refer to all of the above code that doesn't
|
---|
263 | refer to Fraction, float, or complex as "boilerplate". 'r'
|
---|
264 | will be an instance of Fraction, which is a subtype of
|
---|
265 | Rational (r : Fraction <: Rational), and b : B <:
|
---|
266 | Complex. The first three involve 'r + b':
|
---|
267 |
|
---|
268 | 1. If B <: Fraction, int, float, or complex, we handle
|
---|
269 | that specially, and all is well.
|
---|
270 | 2. If Fraction falls back to the boilerplate code, and it
|
---|
271 | were to return a value from __add__, we'd miss the
|
---|
272 | possibility that B defines a more intelligent __radd__,
|
---|
273 | so the boilerplate should return NotImplemented from
|
---|
274 | __add__. In particular, we don't handle Rational
|
---|
275 | here, even though we could get an exact answer, in case
|
---|
276 | the other type wants to do something special.
|
---|
277 | 3. If B <: Fraction, Python tries B.__radd__ before
|
---|
278 | Fraction.__add__. This is ok, because it was
|
---|
279 | implemented with knowledge of Fraction, so it can
|
---|
280 | handle those instances before delegating to Real or
|
---|
281 | Complex.
|
---|
282 |
|
---|
283 | The next two situations describe 'b + r'. We assume that b
|
---|
284 | didn't know about Fraction in its implementation, and that it
|
---|
285 | uses similar boilerplate code:
|
---|
286 |
|
---|
287 | 4. If B <: Rational, then __radd_ converts both to the
|
---|
288 | builtin rational type (hey look, that's us) and
|
---|
289 | proceeds.
|
---|
290 | 5. Otherwise, __radd__ tries to find the nearest common
|
---|
291 | base ABC, and fall back to its builtin type. Since this
|
---|
292 | class doesn't subclass a concrete type, there's no
|
---|
293 | implementation to fall back to, so we need to try as
|
---|
294 | hard as possible to return an actual value, or the user
|
---|
295 | will get a TypeError.
|
---|
296 |
|
---|
297 | """
|
---|
298 | def forward(a, b):
|
---|
299 | if isinstance(b, (int, long, Fraction)):
|
---|
300 | return monomorphic_operator(a, b)
|
---|
301 | elif isinstance(b, float):
|
---|
302 | return fallback_operator(float(a), b)
|
---|
303 | elif isinstance(b, complex):
|
---|
304 | return fallback_operator(complex(a), b)
|
---|
305 | else:
|
---|
306 | return NotImplemented
|
---|
307 | forward.__name__ = '__' + fallback_operator.__name__ + '__'
|
---|
308 | forward.__doc__ = monomorphic_operator.__doc__
|
---|
309 |
|
---|
310 | def reverse(b, a):
|
---|
311 | if isinstance(a, Rational):
|
---|
312 | # Includes ints.
|
---|
313 | return monomorphic_operator(a, b)
|
---|
314 | elif isinstance(a, numbers.Real):
|
---|
315 | return fallback_operator(float(a), float(b))
|
---|
316 | elif isinstance(a, numbers.Complex):
|
---|
317 | return fallback_operator(complex(a), complex(b))
|
---|
318 | else:
|
---|
319 | return NotImplemented
|
---|
320 | reverse.__name__ = '__r' + fallback_operator.__name__ + '__'
|
---|
321 | reverse.__doc__ = monomorphic_operator.__doc__
|
---|
322 |
|
---|
323 | return forward, reverse
|
---|
324 |
|
---|
325 | def _add(a, b):
|
---|
326 | """a + b"""
|
---|
327 | return Fraction(a.numerator * b.denominator +
|
---|
328 | b.numerator * a.denominator,
|
---|
329 | a.denominator * b.denominator)
|
---|
330 |
|
---|
331 | __add__, __radd__ = _operator_fallbacks(_add, operator.add)
|
---|
332 |
|
---|
333 | def _sub(a, b):
|
---|
334 | """a - b"""
|
---|
335 | return Fraction(a.numerator * b.denominator -
|
---|
336 | b.numerator * a.denominator,
|
---|
337 | a.denominator * b.denominator)
|
---|
338 |
|
---|
339 | __sub__, __rsub__ = _operator_fallbacks(_sub, operator.sub)
|
---|
340 |
|
---|
341 | def _mul(a, b):
|
---|
342 | """a * b"""
|
---|
343 | return Fraction(a.numerator * b.numerator, a.denominator * b.denominator)
|
---|
344 |
|
---|
345 | __mul__, __rmul__ = _operator_fallbacks(_mul, operator.mul)
|
---|
346 |
|
---|
347 | def _div(a, b):
|
---|
348 | """a / b"""
|
---|
349 | return Fraction(a.numerator * b.denominator,
|
---|
350 | a.denominator * b.numerator)
|
---|
351 |
|
---|
352 | __truediv__, __rtruediv__ = _operator_fallbacks(_div, operator.truediv)
|
---|
353 | __div__, __rdiv__ = _operator_fallbacks(_div, operator.div)
|
---|
354 |
|
---|
355 | def __floordiv__(a, b):
|
---|
356 | """a // b"""
|
---|
357 | # Will be math.floor(a / b) in 3.0.
|
---|
358 | div = a / b
|
---|
359 | if isinstance(div, Rational):
|
---|
360 | # trunc(math.floor(div)) doesn't work if the rational is
|
---|
361 | # more precise than a float because the intermediate
|
---|
362 | # rounding may cross an integer boundary.
|
---|
363 | return div.numerator // div.denominator
|
---|
364 | else:
|
---|
365 | return math.floor(div)
|
---|
366 |
|
---|
367 | def __rfloordiv__(b, a):
|
---|
368 | """a // b"""
|
---|
369 | # Will be math.floor(a / b) in 3.0.
|
---|
370 | div = a / b
|
---|
371 | if isinstance(div, Rational):
|
---|
372 | # trunc(math.floor(div)) doesn't work if the rational is
|
---|
373 | # more precise than a float because the intermediate
|
---|
374 | # rounding may cross an integer boundary.
|
---|
375 | return div.numerator // div.denominator
|
---|
376 | else:
|
---|
377 | return math.floor(div)
|
---|
378 |
|
---|
379 | def __mod__(a, b):
|
---|
380 | """a % b"""
|
---|
381 | div = a // b
|
---|
382 | return a - b * div
|
---|
383 |
|
---|
384 | def __rmod__(b, a):
|
---|
385 | """a % b"""
|
---|
386 | div = a // b
|
---|
387 | return a - b * div
|
---|
388 |
|
---|
389 | def __pow__(a, b):
|
---|
390 | """a ** b
|
---|
391 |
|
---|
392 | If b is not an integer, the result will be a float or complex
|
---|
393 | since roots are generally irrational. If b is an integer, the
|
---|
394 | result will be rational.
|
---|
395 |
|
---|
396 | """
|
---|
397 | if isinstance(b, Rational):
|
---|
398 | if b.denominator == 1:
|
---|
399 | power = b.numerator
|
---|
400 | if power >= 0:
|
---|
401 | return Fraction(a._numerator ** power,
|
---|
402 | a._denominator ** power)
|
---|
403 | else:
|
---|
404 | return Fraction(a._denominator ** -power,
|
---|
405 | a._numerator ** -power)
|
---|
406 | else:
|
---|
407 | # A fractional power will generally produce an
|
---|
408 | # irrational number.
|
---|
409 | return float(a) ** float(b)
|
---|
410 | else:
|
---|
411 | return float(a) ** b
|
---|
412 |
|
---|
413 | def __rpow__(b, a):
|
---|
414 | """a ** b"""
|
---|
415 | if b._denominator == 1 and b._numerator >= 0:
|
---|
416 | # If a is an int, keep it that way if possible.
|
---|
417 | return a ** b._numerator
|
---|
418 |
|
---|
419 | if isinstance(a, Rational):
|
---|
420 | return Fraction(a.numerator, a.denominator) ** b
|
---|
421 |
|
---|
422 | if b._denominator == 1:
|
---|
423 | return a ** b._numerator
|
---|
424 |
|
---|
425 | return a ** float(b)
|
---|
426 |
|
---|
427 | def __pos__(a):
|
---|
428 | """+a: Coerces a subclass instance to Fraction"""
|
---|
429 | return Fraction(a._numerator, a._denominator)
|
---|
430 |
|
---|
431 | def __neg__(a):
|
---|
432 | """-a"""
|
---|
433 | return Fraction(-a._numerator, a._denominator)
|
---|
434 |
|
---|
435 | def __abs__(a):
|
---|
436 | """abs(a)"""
|
---|
437 | return Fraction(abs(a._numerator), a._denominator)
|
---|
438 |
|
---|
439 | def __trunc__(a):
|
---|
440 | """trunc(a)"""
|
---|
441 | if a._numerator < 0:
|
---|
442 | return -(-a._numerator // a._denominator)
|
---|
443 | else:
|
---|
444 | return a._numerator // a._denominator
|
---|
445 |
|
---|
446 | def __hash__(self):
|
---|
447 | """hash(self)
|
---|
448 |
|
---|
449 | Tricky because values that are exactly representable as a
|
---|
450 | float must have the same hash as that float.
|
---|
451 |
|
---|
452 | """
|
---|
453 | # XXX since this method is expensive, consider caching the result
|
---|
454 | if self._denominator == 1:
|
---|
455 | # Get integers right.
|
---|
456 | return hash(self._numerator)
|
---|
457 | # Expensive check, but definitely correct.
|
---|
458 | if self == float(self):
|
---|
459 | return hash(float(self))
|
---|
460 | else:
|
---|
461 | # Use tuple's hash to avoid a high collision rate on
|
---|
462 | # simple fractions.
|
---|
463 | return hash((self._numerator, self._denominator))
|
---|
464 |
|
---|
465 | def __eq__(a, b):
|
---|
466 | """a == b"""
|
---|
467 | if isinstance(b, Rational):
|
---|
468 | return (a._numerator == b.numerator and
|
---|
469 | a._denominator == b.denominator)
|
---|
470 | if isinstance(b, numbers.Complex) and b.imag == 0:
|
---|
471 | b = b.real
|
---|
472 | if isinstance(b, float):
|
---|
473 | return a == a.from_float(b)
|
---|
474 | else:
|
---|
475 | # XXX: If b.__eq__ is implemented like this method, it may
|
---|
476 | # give the wrong answer after float(a) changes a's
|
---|
477 | # value. Better ways of doing this are welcome.
|
---|
478 | return float(a) == b
|
---|
479 |
|
---|
480 | def _subtractAndCompareToZero(a, b, op):
|
---|
481 | """Helper function for comparison operators.
|
---|
482 |
|
---|
483 | Subtracts b from a, exactly if possible, and compares the
|
---|
484 | result with 0 using op, in such a way that the comparison
|
---|
485 | won't recurse. If the difference raises a TypeError, returns
|
---|
486 | NotImplemented instead.
|
---|
487 |
|
---|
488 | """
|
---|
489 | if isinstance(b, numbers.Complex) and b.imag == 0:
|
---|
490 | b = b.real
|
---|
491 | if isinstance(b, float):
|
---|
492 | b = a.from_float(b)
|
---|
493 | try:
|
---|
494 | # XXX: If b <: Real but not <: Rational, this is likely
|
---|
495 | # to fall back to a float. If the actual values differ by
|
---|
496 | # less than MIN_FLOAT, this could falsely call them equal,
|
---|
497 | # which would make <= inconsistent with ==. Better ways of
|
---|
498 | # doing this are welcome.
|
---|
499 | diff = a - b
|
---|
500 | except TypeError:
|
---|
501 | return NotImplemented
|
---|
502 | if isinstance(diff, Rational):
|
---|
503 | return op(diff.numerator, 0)
|
---|
504 | return op(diff, 0)
|
---|
505 |
|
---|
506 | def __lt__(a, b):
|
---|
507 | """a < b"""
|
---|
508 | return a._subtractAndCompareToZero(b, operator.lt)
|
---|
509 |
|
---|
510 | def __gt__(a, b):
|
---|
511 | """a > b"""
|
---|
512 | return a._subtractAndCompareToZero(b, operator.gt)
|
---|
513 |
|
---|
514 | def __le__(a, b):
|
---|
515 | """a <= b"""
|
---|
516 | return a._subtractAndCompareToZero(b, operator.le)
|
---|
517 |
|
---|
518 | def __ge__(a, b):
|
---|
519 | """a >= b"""
|
---|
520 | return a._subtractAndCompareToZero(b, operator.ge)
|
---|
521 |
|
---|
522 | def __nonzero__(a):
|
---|
523 | """a != 0"""
|
---|
524 | return a._numerator != 0
|
---|
525 |
|
---|
526 | # support for pickling, copy, and deepcopy
|
---|
527 |
|
---|
528 | def __reduce__(self):
|
---|
529 | return (self.__class__, (str(self),))
|
---|
530 |
|
---|
531 | def __copy__(self):
|
---|
532 | if type(self) == Fraction:
|
---|
533 | return self # I'm immutable; therefore I am my own clone
|
---|
534 | return self.__class__(self._numerator, self._denominator)
|
---|
535 |
|
---|
536 | def __deepcopy__(self, memo):
|
---|
537 | if type(self) == Fraction:
|
---|
538 | return self # My components are also immutable
|
---|
539 | return self.__class__(self._numerator, self._denominator)
|
---|