repr/str diff between Python 2 and 3

[Terry Reedy]

On 10/11/2016 8:59 PM, Skip Montanaro wrote:
> I'm trying to port some code from Python 2 to 3, and beyond the usual
> mechanical stuff, I've encountered a difference between the str() of
> floats. Here's an example. In Python 3 I get:
>
>>>> print(repr(27.04 - 0.01))
> 27.029999999999998
>>>> print(str(27.04 - 0.01))
> 27.029999999999998
>
> but in Python 2 str() behaves differently:
>
>>>> print repr(27.04 - 0.01)
> 27.029999999999998
>>>> print str(27.04 - 0.01)
> 27.03
>
> My test case writes through a csv writer, which writes the str() of each
> element to the output. My test assertions are looking at the output, not
> the numeric input, so the option to test if the number is "close enough"
> isn't readily available. I suppose I should adjust my test case, as in this
> case 27.04-0.01 is different than 27.03.

Or you can make the text version-specific.

Basing tests on strings generated by Python is fragile because they are 
subject to change.  For instance, 3.1 has a new algorithm for 
repr(float), discussed in What's New in 3.1.  The repr for other objects 
can and does change with new versions, as to exception messages.  (They 
are only changed in bugfix releases if there is a real and significant 
but.)

> Is there documentation of this particular change? My searching turned up
> documentation of plenty of other changes, but not this particular one.

There was discussed on the development lists.  I believe this change was 
in 3.0, in which case there would be no mention in the 3.x docs.  I did 
not see mention in What New in 3.0, though.  The change would certainly 
be in the NEWS file that lists all commits.  This is now reformatted 
into an html changelog linked from What's New, but this started after in 
3.0.

-- 
Terry Jan Reedy