The patch unifies the creation and representation of "inf", "-inf" and
"nan" on all platforms.

>>> float("inf")
inf
>>> float("-inf")
-inf
>>> float("nan")
nan
>>> repr(1e300 * 1e300)
'inf'
>>> repr(1e300 * 1e300 * 0)
'nan'
>>> repr(1e300 * 1e300 * -1)
'-inf'

Update:

I've added a platform independent implementation of stricmp and strnicmp
(case insensitive compare). It could be useful for other parts of the
code as well.

You have:
#define Py_NAN Py_HUGE_VAL * 0
I think this would be safer as:
#define Py_NAN (Py_HUGE_VAL * 0)

For instance, in code that may do "a / Py_NAN".

Those manual string copies (*cp++ = 'n';) are ugly. Can't you use
strcpy() instead?

Adam Olsen wrote:

You have:
#define Py_NAN Py_HUGE_VAL * 0
I think this would be safer as:
#define Py_NAN (Py_HUGE_VAL * 0)

For instance, in code that may do "a / Py_NAN".

You are right! Fixed

Those manual string copies (*cp++ = 'n';) are ugly. Can't you use
strcpy() instead?

Done

Christian

I'm posting a combined patch for all features at bpo-1640.

Mostly looks good. Here are some nits.

(1) You shouldn't have to add pystrcmp.c to the VC project files since
on Windows it isn't used, right?

(2) Will the Windows input routine still accept the *old*
representations for INF and NAN? IMO that's important (a) so as to be
able to read old pickles or marshalled data, (b) so as to be able to
read data files written by C programs.

(3) Shouldn't you be using Py_HUGE_VAL instead of HUGE_VAL in the chunk
starting at line 187 in floatobject.c?

[Guido]

...
(2) Will the Windows input routine still accept the *old*
representations for INF and NAN? IMO that's important (a) so as to be
able to read old pickles or marshalled data, (b) so as to be able to
read data files written by C programs.

Ha! You're such an optimist ;-) The remarkable truth is that Windows
has never been able to read its own representations for INF and NAN:

'1.#INF'
>>> float(_)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ValueError: invalid literal for float(): 1.#INF
>>> repr(nan)
'-1.#IND'
>>> float(_)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ValueError: invalid literal for float(): -1.#IND

This has nothing to do with Python -- same thing from C, etc.

Guido van Rossum wrote:

(1) You shouldn't have to add pystrcmp.c to the VC project files since
on Windows it isn't used, right?

It was the easiest way to test the functions in pystrcmp. Linux doesn't
have stricmp but it also doesn't require the additional code to mangle
1.#INF into inf.

(2) Will the Windows input routine still accept the *old*
representations for INF and NAN? IMO that's important (a) so as to be
able to read old pickles or marshalled data, (b) so as to be able to
read data files written by C programs.

See Tim's answer.
Pickles and other C programs aren't an issue. Internally NaNs and INFs
are represented with a special bit pattern (all bits of the exponent are
set). As long as users don't use str() or repr() on floats it still
works. The pickle module uses struct.

(3) Shouldn't you be using Py_HUGE_VAL instead of HUGE_VAL in the chunk
starting at line 187 in floatobject.c?

I missed the spot, thanks.

Christian

Historical note: Guido is probably thinking of "the old" pickle and
marshal here, which did have problems with inf and NaN on Windows (as in
they didn't work at all). Michael Hudson changed them to use special
bit patterns instead, IIRC for Python 2.5.

Historical note: Guido is probably thinking of "the old" pickle and
marshal here, which did have problems with inf and NaN on Windows (as in
they didn't work at all). Michael Hudson changed them to use special
bit patterns instead, IIRC for Python 2.5.

In pickle.py, protocol 0 (still the default in 2.6) uses repr(x) to
write a float and float(s) to convert that back to input. Maybe you're
thinking of marshal, which is more sophisticated.

I suggest you check this in (with the Py_HUGE_VAL fix) and then see how
much of bpo-1640 we still need.

Applied in r59558 to the trunk