Re: Zero-padding and zero-masking fixes for to_char(float)

When you posted this, I made a note to review it.

On Wed, Mar 18, 2015 at 05:52:44PM -0400, Bruce Momjian wrote:
> This "junk" digit zeroing matches the Oracle behavior:
>
> SELECT to_char(1.123456789123456789123456789d, '9.9999999999999999999999999999999999999') as x from dual;
> ------
> 1.1234567891234568000000000000000000000
>
> Our output with the patch would be:
>
> SELECT to_char(float8 '1.123456789123456789123456789', '9.9999999999999999999999999999999999999');
> ------
> 1.1234567891234500000000000000000000000
>
> which is pretty close.

PostgreSQL 9.4 returns "1.12345678912346". Your patch truncates digits past
DBL_DIG, whereas both Oracle and PostgreSQL 9.4 round to the nearest DBL_DIG
digits. PostgreSQL must continue to round.

These outputs show Oracle treating 17 digits as significant while PostgreSQL
treats 15 digits as significant. Should we match Oracle in this respect while
we're breaking compatibility anyway? I tend to think yes.

> *************** int4_to_char(PG_FUNCTION_ARGS)
> *** 5214,5221 ****
> /* we can do it easily because float8 won't lose any precision */
> float8 val = (float8) value;
>
> ! orgnum = (char *) palloc(MAXDOUBLEWIDTH + 1);
> ! snprintf(orgnum, MAXDOUBLEWIDTH + 1, "%+.*e", Num.post, val);
>
> /*
> * Swap a leading positive sign for a space.
> --- 5207,5213 ----
> /* we can do it easily because float8 won't lose any precision */
> float8 val = (float8) value;
>
> ! orgnum = psprintf("%+.*e", Num.post, val);

Neither the submission notes nor the commit messages mentioned it, but this
improves behavior for to_char(int4, text). Specifically, it helps EEEE
formats with more than about 500 decimal digits:

SELECT length(to_char(1, '9D' || repeat('9', 800) || 'EEEE'));