[Python-Dev] snprintf breaks build

[Tim Peters]

[Moshe Zadka]
> (Tidbit: since sizeof(long) measures in bytes, and %d prints in
> decimals, then a buffer of length sizeof(long) is enough to hold
> a decimal represntation of a long).

[Greg Ewing]
> Pardon? I think you're a bit out in your calculation there!
>
> 3*sizeof(long) should be enough, though (unless some weird C
> implementation measures sizes in units of more than 8 bits).

Getting closer, but the sign bit can consume a character all by itself, so
3*sizeof(long) sitll isn't enough.  To do this correctly and minimally
requires that we implement an arbitrary-precision log10 function, use the
platform MIN/MIX #define's for longs and chars, and malloc the buffers at
runtime.

Note that instead I boosted the buffer sizes in the module from 80 to 250.
That's obviously way more than enough for 64-bit platforms, and "obviously
way more" is the correct thing to do for programmers <wink>.  If one of the
principled alternatives is ever checked in (be it an snprintf or /F's custom
solution (which I like better)), we can go back and use those instead.