[Python-Dev] Backport new float repr to Python 2.7? (original) (raw)

Raymond Hettinger python at rcn.com
Sun Oct 11 21:44:01 CEST 2009

[Mark Dickinson]

- string to float and float to string conversions are both guaranteed correctly rounded in 3.x: David Gay's code implements the conversion in both directions, and having correctly rounded string -> float conversions is essential to ensure that eval(repr(x)) recovers x exactly.

IMO, that is so important that it should be considered a bug fix. Recently, I lost a lot of time in a discussion about a piece of mathematical code returning a wrong answer. The problem wasn't the Python code; instead, it was the str to float conversion (that we inherit from libc) giving the wrong answer. The code worked correctly under Py3.1 but not Py2.6.

IDLE 2.6.2

float('-30000000000000000477086733292797541405082425691835844169343982673243516316511143182440409957570445312.00000000000000000009000000000000000258695827614150149965556917147323765827915398318737061345018446445465087890625') -2.9999999999999999e+100

Python 3.1.1 (r311:74483, Aug 17 2009, 17:02:12) [MSC v.1500 32 bit (Intel)] on win32 Type "copyright", "credits" or "license()" for more information.

float('-30000000000000000477086733292797541405082425691835844169343982673243516316511143182440409957570445312.00000000000000000009000000000000000258695827614150149965556917147323765827915398318737061345018446445465087890625') -3.0000000000000002e+100

- the repr of round(x, n) really does have at most n digits after the point, giving the semi-illusion that x really has been rounded exactly, and eliminating one of the most common user complaints about the round function.

This is also an important improvement and makes round() do what people expect.

- side effects like finding that float(x) rounds correctly for Decimal instances x.

This is also important because other decimal calculations can often be done exactly and it's a bummer to have an accurate result thrown off by an incorrect rounding to float.

- the float <-> string conversions are under our control, so any bugs found can be fixed in the Python source. There's no shortage of conversion bugs in the wild, and certainly bugs have been observed in OS X, Linux and Windows. (The ones I found in OS X 10.5 have been fixed in OS X 10.6, though.)

Nice win.

Thoughts?

+1

I've worked with the 3.1 float reprs for a while and have been delighted. It was a great improvement.

Raymond

More information about the Python-Dev mailing list