gh-72902: speedup Fraction.from_decimal/float in typical cases by skirpichev · Pull Request #133251 · python/cpython

skirpichev · 2025-05-01T13:49:58Z

Issue: Fractions instantiation revisited #72902

skirpichev · 2025-05-01T13:53:39Z

Benchmark	ref	patch	patch2
Fraction.from_decimal(1)	15.5 us	6.59 us: 2.35x faster	10.4 us: 1.49x faster
Fraction.from_decimal(myint)	16.3 us	11.4 us: 1.43x faster	10.8 us: 1.51x faster
Fraction.from_decimal(Decimal('1'))	10.9 us	7.94 us: 1.37x faster	7.88 us: 1.38x faster
Fraction.from_float(1)	7.06 us	4.25 us: 1.66x faster	7.68 us: 1.09x slower
Fraction.from_float(myint)	8.87 us	9.92 us: 1.12x slower	9.36 us: 1.06x slower
Fraction.from_float(1.1)	7.37 us	4.68 us: 1.57x faster	4.68 us: 1.58x faster
Geometric mean	(ref)	1.49x faster	1.27x faster

Here patch2 is the current version. v1 had a hack to check isinstance(op, (int, Integral)) instead.

Details

# bench.py
import pyperf
from fractions import Fraction as F
from decimal import Decimal as D
from numbers import Integral

runner = pyperf.Runner()
s = 'Fraction.from_decimal'
f = F.from_decimal
class myint:
    numerator = 123
    denominator = 1
    def __int__(self):
        return 123
    def __repr__(self):
        return "myint"
Integral.register(myint)
for v in [1, myint(), D(1)]:
    r = s + '(' + repr(v) + ')'
    runner.bench_func(r, f, v)
s = 'Fraction.from_float'
f = F.from_float
for v in [1, myint(), 1.1]:
    r = s + '(' + repr(v) + ')'
    runner.bench_func(r, f, v)

diff --git a/Lib/fractions.py b/Lib/fractions.py
index fa722589fb..d7887af9f8 100644
--- a/Lib/fractions.py
+++ b/Lib/fractions.py
@@ -335,23 +335,23 @@ def from_float(cls, f):
         Beware that Fraction.from_float(0.3) != Fraction(3, 10).
 
         """
-        if isinstance(f, numbers.Integral):
+        if not isinstance(f, float):
+            if not isinstance(f, (int, numbers.Integral)):
+                raise TypeError("%s.from_float() only takes floats, not %r (%s)" %
+                                (cls.__name__, f, type(f).__name__))
             return cls(f)
-        elif not isinstance(f, float):
-            raise TypeError("%s.from_float() only takes floats, not %r (%s)" %
-                            (cls.__name__, f, type(f).__name__))
         return cls._from_coprime_ints(*f.as_integer_ratio())
 
     @classmethod
     def from_decimal(cls, dec):
         """Converts a finite Decimal instance to a rational number, exactly."""
         from decimal import Decimal
-        if isinstance(dec, numbers.Integral):
-            dec = Decimal(int(dec))
-        elif not isinstance(dec, Decimal):
-            raise TypeError(
-                "%s.from_decimal() only takes Decimals, not %r (%s)" %
-                (cls.__name__, dec, type(dec).__name__))
+        if not isinstance(dec, Decimal):
+            if not isinstance(dec, (int, numbers.Integral)):
+                raise TypeError(
+                    "%s.from_decimal() only takes Decimals, not %r (%s)" %
+                    (cls.__name__, dec, type(dec).__name__))
+            dec = int(dec)
         return cls._from_coprime_ints(*dec.as_integer_ratio())
 
     @classmethod

skirpichev · 2025-05-16T04:06:45Z

This is a second part of improvements, cherry-picked from the issue thread.

I think this is less controversial, as speedup affects main use-cases (i.e. Decimal's or float's, respectively). With v2 version we haven't speed regressions for integer case. Though, I'm not sure about using that hack.

neonene · 2025-05-17T04:45:47Z

Please consider import decimal; isinstance(dec, decimal.Decimal) as well.

This is slightly slower:

sk@note:~ $ python3.13 -m timeit -s 'import decimal;o1=decimal.Decimal(1);o2=1j' \
    'isinstance(o1, decimal.Decimal)'
5000000 loops, best of 5: 82.1 nsec per loop
sk@note:~ $ python3.13 -m timeit -s 'import decimal;o1=decimal.Decimal(1);o2=1j' \
    'isinstance(o2, decimal.Decimal)'
2000000 loops, best of 5: 159 nsec per loop
sk@note:~ $ python3.13 -m timeit -s 'from decimal import Decimal;o1=Decimal(1);o2=1j' \
    'isinstance(o1, Decimal)'
5000000 loops, best of 5: 65.5 nsec per loop
sk@note:~ $ python3.13 -m timeit -s 'from decimal import Decimal;o1=Decimal(1);o2=1j' \
    'isinstance(o2, Decimal)'
2000000 loops, best of 5: 139 nsec per loop

github-actions · 2026-04-24T06:38:07Z

This PR is stale because it has been open for 30 days with no activity.

read-the-docs-community · 2026-05-23T07:45:42Z

Documentation build overview

📚 cpython-previews | 🛠️ Build #32820272 | 📁 Comparing 1ea8f6a against main (a7d5a6c)

🔍 Preview build

483 files changed · ± 444 modified · - 39 deleted

± Modified

- Deleted

eendebakpt

Minor nit, but otherwise approved.

Co-authored-by: Pieter Eendebak <pieter.eendebak@gmail.com>

serhiy-storchaka

LGTM.

scoder

This change is meant as an optimisation but it impacts the correctness. The code should be reverted to its previous state and then the two special cases can be prepended as suggested below. They should not replace the original code.

scoder · 2026-06-26T18:18:10Z

+        if isinstance(f, float):
+            return cls._from_coprime_ints(*f.as_integer_ratio())


For a subclass of float, f.as_integer_ratio() doesn't necessarily return a normalised integer pair. I think we should restrict this optimisation to exact floats.

Suggested change

if isinstance(f, float):

return cls._from_coprime_ints(*f.as_integer_ratio())

if type(f) is float:

return cls._from_coprime_ints(*f.as_integer_ratio())

The intention is to optimise the common case, and that is certainly (and by a margin) exact float, not arbitrary subclasses.
We then need to keep the fallback code that handles subclasses, as it was before.

This affects only float subclasses. The float.as_integer_ration was already documented to return coprime ints.

scoder · 2026-06-26T18:18:42Z

+        if isinstance(dec, Decimal):
+            return cls._from_coprime_ints(*dec.as_integer_ratio())


Suggested change

if isinstance(dec, Decimal):

return cls._from_coprime_ints(*dec.as_integer_ratio())

if type(dec) is Decimal:

return cls._from_coprime_ints(*dec.as_integer_ratio())

scoder · 2026-06-26T18:54:52Z

Hmm, I looked through the code a bit more and found that this assumption is made at other places as well, including the constructor and .from_number(). I'll move the discussion to the ticket since it's not PR specific.

pythongh-72902: speedup Fraction.from_Decimal/float in typical cases

caa608f

bedevere-app Bot mentioned this pull request May 1, 2025

Fractions instantiation revisited #72902

Open

bedevere-app Bot added the awaiting review label May 1, 2025

skirpichev changed the title ~~gh-72902: speedup Fraction.from_Decimal/float in typical cases~~ gh-72902: speedup Fraction.from_decimal/float in typical cases May 1, 2025

+1

1ea8f6a

skirpichev requested a review from serhiy-storchaka May 16, 2025 04:02

github-actions Bot added the stale Stale PR or inactive for long period of time. label Apr 24, 2026

skirpichev closed this Apr 24, 2026

skirpichev deleted the opt-Fraction.from_Decimal/72902 branch April 24, 2026 23:03

skirpichev restored the opt-Fraction.from_Decimal/72902 branch May 23, 2026 07:43

skirpichev reopened this May 23, 2026

Merge branch 'main' into opt-Fraction.from_Decimal/72902

23d0b52

github-actions Bot removed the stale Stale PR or inactive for long period of time. label May 23, 2026

eendebakpt approved these changes May 23, 2026

View reviewed changes

Comment thread Misc/NEWS.d/next/Library/2025-05-01-16-44-16.gh-issue-72902.19qwJW.rst Outdated

bedevere-app Bot added awaiting core review and removed awaiting review labels May 23, 2026

Apply suggestion from @eendebakpt

bf0842d

Co-authored-by: Pieter Eendebak <pieter.eendebak@gmail.com>

serhiy-storchaka approved these changes May 24, 2026

View reviewed changes

Comment thread Lib/fractions.py Outdated

Comment thread Misc/NEWS.d/next/Library/2025-05-01-16-44-16.gh-issue-72902.19qwJW.rst Outdated

bedevere-app Bot added awaiting merge and removed awaiting core review labels May 24, 2026

skirpichev added 2 commits May 24, 2026 17:38

+ cleanup news

ba889d8

address review: rewrite without nested if

cbb2d3f

serhiy-storchaka merged commit 7e5d1d8 into python:main May 25, 2026
52 checks passed

bedevere-app Bot removed the awaiting merge label May 25, 2026

skirpichev deleted the opt-Fraction.from_Decimal/72902 branch May 25, 2026 07:19

scoder reviewed Jun 26, 2026

View reviewed changes

		if isinstance(f, float):
		return cls._from_coprime_ints(*f.as_integer_ratio())

		if isinstance(dec, Decimal):
		return cls._from_coprime_ints(*dec.as_integer_ratio())

Uh oh!

Uh oh!

Conversation

skirpichev commented May 1, 2025 • edited by bedevere-app Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

skirpichev commented May 1, 2025

Uh oh!

skirpichev commented May 16, 2025

Uh oh!

neonene commented May 17, 2025 • edited by skirpichev Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot commented Apr 24, 2026

Uh oh!

read-the-docs-community Bot commented May 23, 2026

Documentation build overview

Uh oh!

eendebakpt left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

serhiy-storchaka left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

scoder left a comment

Choose a reason for hiding this comment

Uh oh!

scoder Jun 26, 2026

Choose a reason for hiding this comment

Uh oh!

skirpichev Jun 27, 2026

Choose a reason for hiding this comment

Uh oh!

scoder Jun 26, 2026

Choose a reason for hiding this comment

Uh oh!

skirpichev Jun 27, 2026

Choose a reason for hiding this comment

Uh oh!

scoder commented Jun 26, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

skirpichev commented May 1, 2025 •

edited by bedevere-app Bot

Loading

neonene commented May 17, 2025 •

edited by skirpichev

Loading