Pelles C forum

Pelles C => Bug reports => Topic started by: loo on May 21, 2015, 11:21:29 PM

Title: problem with double conversion ?
Post by: loo on May 21, 2015, 11:21:29 PM

Hi,

I compiled a json parser (parson) which has also a test suite included. I wondered why two test cases failed and reduced it to the following:

Code: [Select]

printf("%f\n", (double)-9876.543210);
printf("%f\n", (double)0.123456789e-12);
printf("%f\n", (double)1.234567890E+34);
printf("%f\n", (double)23456789012E66);

The output with Pelles C 8.00.60 under Win 8.1 is:
-9876.543210
0.000000
12345678900000001303851000000000000.000000
23456789012000001093800000000000000000000000000000000000000000000000000000000.000000

I was expecting something like:
-9876.543210
0.000000000000123456789
12345678900000000000000000000000000.000000
23456789012000000000000000000000000000000000000000000000000000000000000000000.000000

Any ideas ?

Regards
Otto

Title: Re: problem with double conversion ?
Post by: MichaelW on May 22, 2015, 01:31:27 AM

The default precision is 6 decimal places, but you can specify a greater precision (for a double the realistic limit is ~15 significant digits):

Code: [Select]

printf("%.16f\n", (double)0.123456789e-12);

Result: 0.0000000000001235

Title: Re: problem with double conversion ?
Post by: loo on May 22, 2015, 09:19:15 AM

Right. My mistake. Meanwhile I got this too   ;)
But what I'm really interested in is the 3rd and the 4th line. Where are the ghost digits in the middle of the number coming from ?  :o

Title: Re: problem with double conversion ?
Post by: frankie on May 22, 2015, 01:50:01 PM

Quote from: loo on May 22, 2015, 09:19:15 AM

But what I'm really interested in is the 3rd and the 4th line. Where are the ghost digits in the middle of the number coming from ?  :o

Rounding and float irriducible fractions (numbers that expressed in base 2 give irrational or periodic fractional number).