Printing rationals as (rounded) floats

Hi,

This problem showed up, trying to display a rational with two decimals:

	?- format('~2f', [444996 rdiv 1000000]).
	0.45

Using normal rounding conventions would have expected 0.44.  The way
this is handled is currently as follows:

	- First, the rational is converted into a GMP float with 4
	times the number of requested digits as bits.  4 times because
	that gives us more (binary) precision than the requested
	decimal.

	- Next, gmp_snprintf() is used to print the result.

Now, following an experiment by Keri Harris, gmp_snprintf() isn't good
at rounding.  Using the attached program, we get this output:

$ gcc -o test test.c -lgmp -lmpfr
$ ./test
mpq_t:  444996/1000000

mpf_t:  0.45
mpf_t:  0.445
mpf_t:  0.4450
mpf_t:  0.44500
mpf_t:  0.444996

mpf_get_str() 2dp: 44, exponent: 0
mpf_get_str() 3dp: 445, exponent: 0
mpf_get_str() 4dp: 445, exponent: 0
mpf_get_str() 5dp: 445, exponent: 0
mpf_get_str() 6dp: 444996, exponent: 0

mpfr_t: 0.44
mpfr_t: 0.445
mpfr_t: 0.4450
mpfr_t: 0.44500
mpfr_t: 0.444996

This may suggest using mpf_get_str or mpfr, but I think the problem
is more fundamental.  When we convert to a float, the above may round
to basically 0.445, after which subsequent rounding to two digits
will also produce 0.45.

Does anyone has a glue how to do this job correctly?  Is the only way
to do the math in rationals, computing the next digit?

	Thanks --- Jan
-------------- next part --------------
A non-text attachment was scrubbed...
Name: test.c
Type: text/x-csrc
Size: 1147 bytes
Desc: test.c
URL: <https://lists.iai.uni-bonn.de/pipermail/swi-prolog/attachments/20131104/5ac46563/attachment.bin>