The patch modifies configure.in to add PY_FORMAT_SIZE_T
to configure.in (meaning you need to run autoheader on
configure.in) so that if %zd is supported for size_t it
sets PY_FORMAT_SIZE_T to "z", otherwise it goes
undefined and the preprocessor trickery in
Include/pyport.h kicks in.

This fix removes compiler warnings on OS X 10.4.6 with
gcc 4.0.1 thanks to PY_FORMAT_SIZE_T being set to "".

Initially assigned to Martin v. Loewis since he said
this would be good to do and the Py_ssize_t stuff is
his invention.

Logged In: YES 
user_id=21627

The patch seems to rely on printf returning <0 for the
unrecognized format. That seems unreliable: atleast on
Linux, printf just outputs the format as-is for unrecognized
formats. Instead, I think it should use sprintf, and then
check whether the result is the string "0" (in addition to
checking whether the printf call itself failed).

Logged In: YES 
user_id=357491

OK, uploaded a new version that uses strchr to check for
'%', 'z', and 'd'.  If it looks reasonable I will apply it
and hope I don't break the buildbot.  =)

Logged In: YES 
user_id=357491

Realized there is a better way: just strncmp() for the
expected result.  Uploaded a new version.

Logged In: YES 
user_id=21627

Looks fine to me, although it has "unusual" style of C:

- sizeof(char) is guaranteed to be 1 by the C standard. The
C standard defines "char" and "byte" as synonyms, even if
that means that "byte" has more than 8 bits. sizeof gives
the number of bytes, so for char, it is always 1.

- for a fixed-size array, people would normally make this an
automatic (stack) variable, instead of bothering with
explicit memory allocation, i.e.

  char str_buffer[4]

Just out of fear of buffer overruns, many people would also
add some horrendous overallocation, such as str_buffer[1024] :-)

Logged In: YES 
user_id=357491

Yeah, I tried to use a string constant as a stack value, but
that didn't work.  =)  My brain just was not thinking in C
when I first came up with the patch.

I have a new version that uses a char array as the buffer. 
I am on vacation so I don't have the time to apply it and
break buildbot, so I will hold off on applying if no one
finds problems with this version.

Logged In: YES 
user_id=357491

Applied in r45960 .