large integer formatting errors

[Andrew J. Schorr]

In case people are still following this thread, I am attaching
the latest version of my integer formatting patch that is
being committed to xgawk (http://sourceforge.net/projects/xmlgawk).
This fixes problems that are demonstrated by the attached test
case.

So, for example, unpatched gawk 3.1.5 on 32-bit linux does the following:

bash-2.05b$ ./gawk --version | head -1
GNU Awk 3.1.5

bash-2.05b$ ./gawk -f test/intformat.awk | diff test/intformat.ok -
gawk: /home/users/schorr/src/xgawk/cvs/test/intformat.awk:73: fatal error: 
internal error
0a1,3
> (sprintf(0x%x,2^64) = 0x0)-1.84467e+19 = -1.84467e+19
> (sprintf(%d,2^64) = 0)-1.84467e+19 = -1.84467e+19
> (sprintf(0%o,2^64) = 00)-1.84467e+19 = -1.84467e+19
2d4
< 0000000000000000000000000000000000000000000000000000001

whereas with the patched version there should be no output.

Mon Mar  6 22:18:38 UTC 2006 Andrew J. Schorr <address@hidden>

        * node.c (format_val): Test whether a numeric value is integral
        simply by converting it to long and then back again, and checking
        if the value matches.  That's more robust than trying to test
        whether the floating-point value is representable as a long.
        * builtin.c (format_tree): Fix buffer overflow bug, off-by-one errors
        in checking snprintf return codes, and use "%.0f" to implement "%d"
        formatting.  For octal or hex formatting, test whether the value
        is integral by trying to convert back to floating point and seeing
        if the value matches.  This is more robust than trying to test
        whether the floating-point value is representable as an integer.

Regards,
Andy

large_integer.patch
Description: Text document

intformat.awk
Description: Text document

intformat.ok
Description: Text document