Bug #5497: Math.log10(10_000) error on HP-UX/PA

• Status: Closed
• Assignee: takano32 (Mitsuhiro TAKANO)
• Target version: -
• ruby -v: UNKNOWN
• Backport: 2.2: UNKNOWN, 2.3: UNKNOWN, 2.4: UNKNOWN

[twwlogin (The Written Word Inc)]

I've built 1.9.2 from svn revision at 33531 on HP-UX/PA and HP-UX/IA. I'm getting two different results for Math.log10(10_000) depending on IA64 and PA-RISC:
(ia64)
p Math.log10(10_000)
4
(pa-risc)
p Math.log10(10_000)
3.9999999999999996

Any idea where in the code I can look to try and fix this?

[naruse (Yui NARUSE)]

The Written Word Inc wrote:

(ia64)
p Math.log10(10_000)
4

Show the result of p 4.0.

(pa-risc)
p Math.log10(10_000)
3.9999999999999996

Build and run following C code:

% cat f.c
#include <math.h>
#include <stdio.h>

int main(void) {
  printf("%a\n", log10(10000));
  return 0;
}
%  gcc -lm f.c && ./a.out
0x1p+2 (or something)

[twwlogin (The Written Word Inc)]

Yui NARUSE wrote:

The Written Word Inc wrote:

(ia64)
p Math.log10(10_000)
4

Show the result of p 4.0.

(ia64)
p 4.0
4.0
(pa-risc)
p 4.0
4.0

(pa-risc)
p Math.log10(10_000)
3.9999999999999996

Build and run following C code:

% cat f.c
#include <math.h>
#include <stdio.h>

int main(void) {
  printf("%a\n", log10(10000));
  return 0;
}
%  gcc -lm f.c && ./a.out
0x1p+2 (or something)

$ cat t.c
#include <math.h>
#include <stdio.h>

int main(void) {
  printf("%f\n", log10(10000));
  printf("%a\n", log10(10000));
  return 0; 
}

(ia64)
  $ cc +O1 +Ofltacc +Olit=all +Oentrysched +Odataprefetch -lm t.c
  $ ./a.out
  4.000000
  0x1.0000000000000p+2

%a is only supported on HP-UX/IA systems, not HP-UX/PA. We're using the HP-UX C compiler to build Ruby as well.

[naruse (Yui NARUSE)]

The Written Word Inc wrote:

Yui NARUSE wrote:

The Written Word Inc wrote:

(ia64)
p Math.log10(10_000)
4

Show the result of p 4.0.

(ia64)
p 4.0
4.0

Hmm, it is strange but I don't have any idea.

(pa-risc)
p Math.log10(10_000)
3.9999999999999996

Build and run following C code:

% cat f.c
#include <math.h>
#include <stdio.h>

int main(void) {
  printf("%a\n", log10(10000));
  return 0;
}
%  gcc -lm f.c && ./a.out
0x1p+2 (or something)

$ cat t.c
#include <math.h>
#include <stdio.h>

int main(void) {
  printf("%f\n", log10(10000));
  printf("%a\n", log10(10000));
  return 0; 
}

%a is only supported on HP-UX/IA systems, not HP-UX/PA. We're using the HP-UX C compiler to build Ruby as well.

OK, can you show a result without %a?
As Bílka says, this is because it is floating point, but it can be fixed by some options to specify libm as IEEE 754 mode.

[twwlogin (The Written Word Inc)]

Yui NARUSE wrote:

The Written Word Inc wrote:

$ cat t.c
#include <math.h>
#include <stdio.h>

int main(void) {
  printf("%f\n", log10(10000));
  printf("%a\n", log10(10000));
  return 0; 
}

%a is only supported on HP-UX/IA systems, not HP-UX/PA. We're using the HP-UX C compiler to build Ruby as well.

OK, can you show a result without %a?
As Bílka says, this is because it is floating point, but it can be fixed by some options to specify libm as IEEE 754 mode.

$ cat t.c
#include <math.h>
#include <stdio.h>

int main(void) {
  printf("%f\n", log10(10000));
  return 0;
} 

$ cc t.c -lm
$ ./a.out
4.000000

[ko1 (Koichi Sasada)]

• Assignee set to mrkn (Kenta Murata)

[shyouhei (Shyouhei Urabe)]

• Status changed from Open to Assigned

[mrkn (Kenta Murata)]

• Status changed from Assigned to Feedback

I don't have any environments to examine this issue.
Please give me a patch to fix this.

[naruse (Yui NARUSE)]

• Target version changed from 1.9.2 to 2.6

[naruse (Yui NARUSE)]

• Assignee changed from mrkn (Kenta Murata) to takano32 (Mitsuhiro TAKANO)

[nobu (Nobuyoshi Nakada)]

Please try more precision.
The default precision doesn't seem enough.

#include <math.h>
#include <stdio.h>

int main(void) {
  printf("%.19f\n", log10(10000));
  return 0;
}

[naruse (Yui NARUSE)]

• Target version deleted (2.6)

[jeremyevans0 (Jeremy Evans)]

• Status changed from Feedback to Closed