Fix stdio fmt of "%.0e" and "%.0g"

[G4Vi]

Fixes #543

"%.0e" was fixed by always using dtoa mode 2: max(1,ndigits) significant digits so the number will still be rounded even if no decimal places will be printed, the change should only change behavior when precision is 0.

The fix to "%.0g" is similar, according to man 3 printf "if the precision is zero, it is treated as 1", so perc is now rounded up to 1 instead of 0 if necessary. This also leads to always using dtoa mode 2.

EDIT: Changing this to WIP as "%.0g", 10 and "%.0g", -10. appears to be failing.
EDIT2: Not sure if the behavior required by tests is desired, see next comment.

[G4Vi]

Is there a specific spec of printf cosmo is trying to follow? The failing tests below from fmt_test.c in {fmt, input, expected} form (do ensure that "%.0g" prints the shortest representation of 10 and -10, however it differs from glibc and my interpretation of the linux man 3 printf

{"%.0g", 10., "10"},
{"%.0g", -10., "-10"},

glibc prints

1e+01
-1e+01

for those tests.

My interpretation of the g, G conversion specifier in psuedo code is as follows:

if(precision == undef)
{
    precision = 6;
}
else if(precision == 0)
{
    precision = 1;
}
if((exponent_from_conversion < -4) || (exponent_from_conversion >= precision))
{
    usestylee();
}
else
{
    usestylef();
}

The "%g" patch currently included causes 10 and -10 to match the glibc behavior of encoding them as 1e+01 and -1e+01.

[jart]

Is there a specific spec of printf cosmo is trying to follow?

Glibc is the gold standard of C libraries. Anything that helps us be closer to Glibc in terms of behavior would be very welcome.

If we have tests that specify a behavior that's inconsistent with Glibc, then our tests are wrong. Many of them were borrowed from Paland's printf() implementation which was intended to be a simple solution for embedded systems. We've modified our implementation a lot since then to bring ourselves into closer conformance. For instance, a lot of the double formatting code in our printf() implementation was borrowed from the dtoa authors.

Since you're already a contributor, please ping me when this is ready to merge. Then I'll approve it.

[G4Vi]

@jart Thanks for the quick reviews. I fixed the %g tests to match glibc and added more. LGTM.

[jart]

Thank you!