gh-72902: improve Fraction constructor speed for typical inputs

[skirpichev]

This moves abc check for numbers.Rational - down.

• Issue: Fractions instantiation revisited #72902

---

Benchmark	ref	patch
Fraction(myint)	5.62 us	7.03 us: 1.25x slower
Fraction(Fraction(1, 2))	5.89 us	4.88 us: 1.21x faster
Fraction(Decimal('0.25'))	11.1 us	7.98 us: 1.39x faster
Fraction(IntSubclass(123))	5.93 us	5.05 us: 1.17x faster
Fraction(FloatSubclass(0.25))	7.69 us	4.26 us: 1.81x faster
Fraction('123')	19.2 us	16.1 us: 1.19x faster
Fraction('1/3')	19.6 us	16.4 us: 1.20x faster
Fraction('1.2e-3')	26.1 us	23.0 us: 1.14x faster
Fraction('-.2')	23.4 us	20.2 us: 1.16x faster
Geometric mean	(ref)	1.21x faster

Details

# bench.pyimportpyperffromfractionsimportFractionasFfromdecimalimportDecimalasDfromnumbersimportIntegralrunner=pyperf.Runner()
s='Fraction'classMyInt:
numerator=123denominator=1def__int__(self):
return123def__repr__(self):
return"myint"classIntSubclass(int):
def__repr__(self):
return'IntSubclass('+super().__repr__()+')'classFloatSubclass(float):
def__repr__(self):
return'FloatSubclass('+super().__repr__()+')'Integral.register(MyInt)
forvin [MyInt(), F(1, 2), D("0.25"),
IntSubclass(123), FloatSubclass(0.25),
"123", "1/3", "1.2e-3", "-.2"]:
r=s+'('+repr(v) +')'runner.bench_func(r, F, v)

[eendebakpt]

There is a slight behaviour change here: for Fraction subclasses the construction now goes via .as_integer_ratio, instead of via .numerator and .denumerator (for this reason the benchmark Fraction(Fraction(1, 2)) is improving).

A minimal example:

import fractions
class MyFraction(Fraction):
@property
def numerator(self):
print('he!')
return super().numerator
f = MyFraction(1,2)
Fraction(f)

I think the change is acceptable and would probably accept the PR.

[skirpichev]

>>> help(fractions.Fraction.as_integer_ratio)
Help on function as_integer_ratio in module fractions:
as_integer_ratio(self)
Return a pair of integers, whose ratio is equal to the original Fraction.
The ratio is in lowest terms and has a positive denominator.

[serhiy-storchaka]

LGTM. 👍