[Python-Dev] PEP 3101: floats format 'f' and 'F'

[Eric Smith]

Thomas Heller wrote:
> Guido van Rossum schrieb:
>> On Thu, Jul 17, 2008 at 9:25 AM, Raymond Hettinger <python at rcn.com> wrote:
>>> From: "Eric Smith" <eric+python-dev at trueblade.com>
>>>> I have this ready for checkin (with docs and tests).  I'd like to get it
>>>> in for this beta, since it does involved changed behavior, no matter how
>>>> small ('1e+100' becomes '1E+100' with '%F').  But it relies on the
>>>> platform's vsnprintf to do the right thing with 'F', so I'm worried
>>>> about breakages on platforms I don't have access to.  Resolving those
>>>> issues might take a few days.
>>>>
>>>> Any advice on checking this in now, or waiting until after this beta is
>>>> released?
>>> I recommend doing it after the release.  It's unlikely to be exercised
>>> much by the beta users so there's no real advantage.  If you wait
>>> until afterwards, then there is time to let the buildbots have a go
>>> at it and reveal any cross-platform issues.
>>>
>>> Besides, Barry said something about getting meaner.
>>> Would hate to find out what he meant the hard way ;-)
>> I'd advise the opposite -- check it in now. It's not going to break
>> anything, and if it is, let's find out sooner rather than later.
>>
> 
> Before we get old waiting for the windows buildbots, here how test_format
> fails now (on XP SP2 x86, trunk rev 65072:
> 
> c:\svn\trunk\PCbuild>.\\python_d  -E -tt ../lib/test/regrtest.py test_format
> test_format
> test test_format produced unexpected output:
> **********************************************************************
> '%F' % (1.0,) == '' != '1.000000'
> u'%F' % (1.0,) == u'' != '1.000000'
> '%f' % (nan,) == '-1.#IND00' != 'nan'
> u'%f' % (nan,) == u'-1.#IND00' != 'nan'
> '%F' % (nan,) == '' != 'NAN'
> u'%F' % (nan,) == u'' != 'NAN'
> '%f' % (inf,) == '1.#INF' != 'inf'
> u'%f' % (inf,) == u'1.#INF' != 'inf'
> '%F' % (inf,) == '1.#INF' != 'INF'
> u'%F' % (inf,) == u'1.#INF' != 'INF'
> 
> **********************************************************************
> 1 test failed:
>     test_format
> [23521 refs]
> 
> c:\svn\trunk\PCbuild>
> 
> Windows vsnprintf does not support '%F' at all, and '%f' does not
> behave as expected with nan and inf.
> 

I was afraid of that.  I'll back these out until I can dig up a Windows 
box to test on.

Thanks for checking this.

Eric.