[Python-Dev] [Python-checkins] r51525 - in python/trunk:Lib/test/test_float.py Objects/floatobject.c

[Alex Martelli]

On Aug 23, 2006, at 3:13 PM, martin at v.loewis.de wrote:

> Zitat von Tim Peters <tim.peters at gmail.com>:
>
>> Huh.  It's been a (mildly controversial, but intentional all the  
>> same)
>> feature that Python tries to raise raise OverflowError on overflowing
>> libm operations.  Doesn't work all that well, since   there's no
>> consistency across platforms about when libm sets errno, or to what
>> (when it does).
>
> The other question is whether pow() is guaranteed to return inf
> when it sets errno to ERANGE; I believe there is no such guarantee.

Summarizing my longer mail that crossed with yours: the guarantee is  
there in the 1999 C Standard; I don't have access any more to a copy  
of the 1990 C Standard to check how things were back then.

> So I think the change should be reverted.

I have reverted the change, but I think it (or rather "something  
like" it -- *SOME* fix for x**2 raising OverflowError while x*x  
proceeds happily for a certain range of floats) is important and  
should go back in again.  Using Py_HUGE_VAL instead of trusting pow's  
result might be OK, for example, if the 1990 C Standard is that much  
weaker on this than 1999's and we need to stick to it anyway.  But  
the problem is that testing this sensibly requires fpconst (from PEP  
754) or the equivalent, otherwise it may be impossible to write  
unittest to show the presence or absence of this bug.


Alex