possible issue withC formatted input/output

possible issue withC formatted input/output

[a.rburgers]

[-- Attachment #1: Type: text/plain, Size: 1212 bytes --]

While preparing gsl packages, one of the check fails.
The failure is reproduced by attached .c file, which writes
doubles to a file, and reads them back.

On linux this is the output (all zero differences in the last column).

ivals[0]=1: 0.002518891687657430663 0.002518891687657430663 0
ivals[1]=6: 0.01511335012594458398 0.01511335012594458398 0
ivals[2]=41: 0.1032745591939546598 0.1032745591939546598 0
ivals[3]=45: 0.113350125944584379 0.113350125944584379 0
ivals[4]=46: 0.1158690176322418192 0.1158690176322418192 0
dawn.ad.intra{~/tmp }75: uname -a
Linux dawn.ad.intra 2.6.9-89.31.1.ELsmp #1 SMP Mon Oct 4 21:41:59 EDT 
2010 x86_64 x86_64 x86_64 GNU/Linux

on cygwin (both 32 and 64 bit):

$ ./tst
ivals[0]=1: 0.002518891687657430663 0.002518891687657430663 0
ivals[1]=6: 0.01511335012594458398 0.01511335012594458224 1.7347e-18
ivals[2]=41: 0.1032745591939546598 0.1032745591939546459 1.3878e-17
ivals[3]=45: 0.113350125944584379 0.1133501259445843651 1.3878e-17
ivals[4]=46: 0.1158690176322418192 0.1158690176322418053 1.3878e-17

the gsl test expects 0 differences and reports a fail.
I am not sure whether this is a bug.  Maybe something with
extended precision registers, when I wild guess.

Teun


[-- Attachment #2: tst.c --]
[-- Type: text/plain, Size: 462 bytes --]

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

#define NVALS 5

int main(void) {
    int N = 397;
    int i, ivals[NVALS] = {1, 6, 41, 45, 46};
    double xr, x;
    FILE *f;
    for (i = 0; i < NVALS; i++) {
	x = ivals[i]/ ((double) N);
	f = fopen("test.txt", "w");
	fprintf(f, "%.19e\n", x);
	fclose(f);
	f = fopen("test.txt", "r");
	fscanf(f, "%lg", &xr);
	fprintf(stderr, "ivals[%d]=%d: %.19g %.19g %.5g\n",
		i, ivals[i], x, xr, fabs(x-xr));
    }
    fclose(f);
}


[-- Attachment #3: Type: text/plain, Size: 218 bytes --]

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple

Re: possible issue withC formatted input/output

[Corinna Vinschen]

On Mar 25 21:44, a.rburgers@quicknet.nl wrote:
> While preparing gsl packages, one of the check fails.
> The failure is reproduced by attached .c file, which writes
> doubles to a file, and reads them back.
> 
> On linux this is the output (all zero differences in the last column).
> 
> ivals[0]=1: 0.002518891687657430663 0.002518891687657430663 0
> ivals[1]=6: 0.01511335012594458398 0.01511335012594458398 0
> ivals[2]=41: 0.1032745591939546598 0.1032745591939546598 0
> ivals[3]=45: 0.113350125944584379 0.113350125944584379 0
> ivals[4]=46: 0.1158690176322418192 0.1158690176322418192 0
> dawn.ad.intra{~/tmp }75: uname -a
> Linux dawn.ad.intra 2.6.9-89.31.1.ELsmp #1 SMP Mon Oct 4 21:41:59
> EDT 2010 x86_64 x86_64 x86_64 GNU/Linux
> 
> on cygwin (both 32 and 64 bit):
> 
> $ ./tst
> ivals[0]=1: 0.002518891687657430663 0.002518891687657430663 0
> ivals[1]=6: 0.01511335012594458398 0.01511335012594458224 1.7347e-18
> ivals[2]=41: 0.1032745591939546598 0.1032745591939546459 1.3878e-17
> ivals[3]=45: 0.113350125944584379 0.1133501259445843651 1.3878e-17
> ivals[4]=46: 0.1158690176322418192 0.1158690176322418053 1.3878e-17
> 
> the gsl test expects 0 differences and reports a fail.
> I am not sure whether this is a bug.  Maybe something with
> extended precision registers, when I wild guess.

Thanks for the testcase.  This problem has been introduced in newlib a
while back and made it into 1.7.17, unfortunately.  However, it's fixed
in the latest snapshots from http://cygwin.com/snapshots/ and will be in
1.7.18.


Corinna

-- 
Corinna Vinschen                  Please, send mails regarding Cygwin to
Cygwin Maintainer                 cygwin AT cygwin DOT com
Red Hat

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple