test release gsl-1.9.90.tar.gz

test release gsl-1.9.90.tar.gz

[Brian Gough]

I have made a test release for gsl-1.10. 

I have only checked it on GNU/Linux -- please report any problems you
find in compiling it, running make check or against existing
applications (it should be backwards compatible).

It is available at the following location:

http://www.network-theory.co.uk/download/gsl/gsl-1.9.90.tar.gz
http://www.network-theory.co.uk/download/gsl/gsl-1.9.90.tar.gz.sig

The main new features from the ChangeLog are:

** License updated to GNU GPL version 3.

** Added support for generalized eigensystems (Patrick Alken)

** Plus bug fixes for all problems reported since the last release.

-- 
Brian Gough

Re: test release gsl-1.9.90.tar.gz

[Patrick Alken]

When compiling on:

Linux 2.4.21-50.EL #1 SMP Tue May 8 17:08:28 EDT 2007 ia64 ia64 ia64
GNU/Linux

compile goes fine, make check gives this error:

make[2]: Entering directory `/home/palken/gsl-1.9.90/specfunc'
FAIL: gsl_sf_legendre_sphPlm_e(140,135,0.99998689456491752,&r) [2229]
  expected: -6.542652532690932e-305
  obtained:                    0 +/- 0 (rel=nan)
  fracdiff:                    1
  value/expected not consistent within reported error
  value not within tolerance of expected value
                       0                       0
FAIL: Legendre Functions [2439]
FAIL: test
===================
1 of 1 tests failed
===================

I used no special flags for configure. Some more info:

> cat /proc/cpuinfo 
processor  : 0
vendor     : GenuineIntel
arch       : IA-64
family     : Itanium 2
model      : 2
revision   : 1
archrev    : 0
features   : branchlong
cpu number : 0
cpu regs   : 4
cpu MHz    : 1595.647000
itc MHz    : 1595.647000
BogoMIPS   : 2390.75
siblings   : 1


On Tue, Sep 11, 2007 at 10:17:14AM +0100, Brian Gough wrote:
> I have made a test release for gsl-1.10. 
> 
> I have only checked it on GNU/Linux -- please report any problems you
> find in compiling it, running make check or against existing
> applications (it should be backwards compatible).
> 
> It is available at the following location:
> 
> http://www.network-theory.co.uk/download/gsl/gsl-1.9.90.tar.gz
> http://www.network-theory.co.uk/download/gsl/gsl-1.9.90.tar.gz.sig
> 
> The main new features from the ChangeLog are:
> 
> ** License updated to GNU GPL version 3.
> 
> ** Added support for generalized eigensystems (Patrick Alken)
> 
> ** Plus bug fixes for all problems reported since the last release.
> 
> -- 
> Brian Gough
> 

Re: test release gsl-1.9.90.tar.gz

[Aleksej Saushev]

[-- Attachment #1: Type: text/plain, Size: 780 bytes --]

Brian Gough <bjg@network-theory.co.uk> writes:

> I have made a test release for gsl-1.10. 

gcc -DHAVE_CONFIG_H -I. -I.. -I..     -g -O2 -c test.c
/bin/ksh ../libtool --tag=CC    --mode=link gcc  -g -O2    -o test test.o libgslsys.la ../ieee-utils/libgslieeeutils.la ../err/libgslerr.la ../test/libgsltest.la libgslsys.la ../utils/libutils.la -lm
gcc -g -O2 -o test test.o  ../ieee-utils/.libs/libgslieeeutils.a ../err/.libs/libgslerr.a ../test/.libs/libgsltest.a ./.libs/libgslsys.a ../utils/.libs/libutils.a -lm
make  check-TESTS
FAIL: gsl_isinf(-inf) (1 observed vs -1 expected) [112]
FAIL: test
===================
1 of 1 tests failed
===================

Fix attached.

NetBSD 4.99.31 i386
cc (GCC) 4.1.3 20070620 prerelease (NetBSD nb1 20070620)

With fix all tests pass.


[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: Fix wrong isinf ret. value assumption in tests --]
[-- Type: text/x-patch, Size: 377 bytes --]

--- sys/test.c.orig	2007-06-11 12:57:34.000000000 +0400
+++ sys/test.c	2007-06-11 12:57:52.000000000 +0400
@@ -383,7 +383,7 @@
        negative).  (This is all that C99 requires.) */
 
     s = gsl_isinf (-inf);  
-    gsl_test_int (s, -1, "gsl_isinf(-inf)");
+    gsl_test_int (s, 1, "gsl_isinf(-inf)");
 
     s = gsl_isinf (nan);
     gsl_test_int (s, 0, "gsl_isinf(nan)");

Re: test release gsl-1.9.90.tar.gz

[Brian Gough]

At Wed, 12 Sep 2007 00:00:35 +0400,
Aleksej Saushev wrote:
> make  check-TESTS
> FAIL: gsl_isinf(-inf) (1 observed vs -1 expected) [112]
> FAIL: test
> ===================
> 1 of 1 tests failed
> ===================
> 
> Fix attached.
> 
> NetBSD 4.99.31 i386
> cc (GCC) 4.1.3 20070620 prerelease (NetBSD nb1 20070620)

Thanks!

-- 
Brian Gough

Re: test release gsl-1.9.90.tar.gz

[Brian Gough]

At Tue, 11 Sep 2007 12:39:40 -0600,
> make[2]: Entering directory `/home/palken/gsl-1.9.90/specfunc'
> FAIL: gsl_sf_legendre_sphPlm_e(140,135,0.99998689456491752,&r) [2229]
>   expected: -6.542652532690932e-305
>   obtained:                    0 +/- 0 (rel=nan)
>   fracdiff:                    1
>   value/expected not consistent within reported error
>   value not within tolerance of expected value

Thanks I'll look into that. Another of these new underflows I'd guess.

-- 
Brian Gough

Re: test release gsl-1.9.90.tar.gz

[Jari Häkkinen]

Applying Aleksejs patch gives a successful compile and test of the 
package on my Intel Core 2 Duo Mac 10.4.10


gcc --version
i686-apple-darwin8-gcc-4.0.1 (GCC) 4.0.1 (Apple Computer, Inc. build 5363)


Jari


Brian Gough wrote:
> At Wed, 12 Sep 2007 00:00:35 +0400,
> Aleksej Saushev wrote:
>> make  check-TESTS
>> FAIL: gsl_isinf(-inf) (1 observed vs -1 expected) [112]
>> FAIL: test
>> ===================
>> 1 of 1 tests failed
>> ===================
>>
>> Fix attached.
>>
>> NetBSD 4.99.31 i386
>> cc (GCC) 4.1.3 20070620 prerelease (NetBSD nb1 20070620)
> 
> Thanks!
> 

Re: test release gsl-1.9.90.tar.gz

[Tommy Nordgren]

On 11 sep 2007, at 22.00, Aleksej Saushev wrote:

> Brian Gough <bjg@network-theory.co.uk> writes:
>
>> I have made a test release for gsl-1.10.
>
> gcc -DHAVE_CONFIG_H -I. -I.. -I..     -g -O2 -c test.c
> /bin/ksh ../libtool --tag=CC    --mode=link gcc  -g -O2    -o test  
> test.o libgslsys.la ../ieee-utils/libgslieeeutils.la ../err/ 
> libgslerr.la ../test/libgsltest.la libgslsys.la ../utils/ 
> libutils.la -lm
> gcc -g -O2 -o test test.o  ../ieee-utils/.libs/libgslieeeutils.a ../ 
> err/.libs/libgslerr.a ../test/.libs/libgsltest.a ./.libs/ 
> libgslsys.a ../utils/.libs/libutils.a -lm
> make  check-TESTS
> FAIL: gsl_isinf(-inf) (1 observed vs -1 expected) [112]
> FAIL: test
> ===================
> 1 of 1 tests failed
> ===================
>
> Fix attached.
>
> NetBSD 4.99.31 i386
> cc (GCC) 4.1.3 20070620 prerelease (NetBSD nb1 20070620)
>
> With fix all tests pass.
>
> --- sys/test.c.orig	2007-06-11 12:57:34.000000000 +0400
> +++ sys/test.c	2007-06-11 12:57:52.000000000 +0400
> @@ -383,7 +383,7 @@
>         negative).  (This is all that C99 requires.) */
>
>      s = gsl_isinf (-inf);
> -    gsl_test_int (s, -1, "gsl_isinf(-inf)");
> +    gsl_test_int (s, 1, "gsl_isinf(-inf)");
>
>      s = gsl_isinf (nan);
>      gsl_test_int (s, 0, "gsl_isinf(nan)");

	The following alternate patch will fix the test failure, without  
modifying the test cases.
--- /Users/emac/Desktop/infnan.c        2007-09-12 22:37:19.000000000  
+0200
+++ /Softsys/gsl-1.9.90/sys/infnan.c    2007-09-12 22:44:30.000000000  
+0200
@@ -118,7 +118,11 @@
  int
  gsl_isinf (const double x)
  {
-    return isinf(x);
+       int result = isinf(x);
+       if (result > 0 && x < 0) {
+               result = - result;
+       }
+       return result;
  }
  # else
Note: the one on the desktop is the unmodified file.
Also, with make -k check all other test cases pass.
-------------------------------------
This sig is dedicated to the advancement of Nuclear Power
Tommy Nordgren
tommy.nordgren@comhem.se

Re: test release gsl-1.9.90.tar.gz

[Tommy Nordgren]

On 12 sep 2007, at 23.01, Tommy Nordgren wrote:

>
> On 11 sep 2007, at 22.00, Aleksej Saushev wrote:
>
>> Brian Gough <bjg@network-theory.co.uk> writes:
>>
>>> I have made a test release for gsl-1.10.
>>
>> gcc -DHAVE_CONFIG_H -I. -I.. -I..     -g -O2 -c test.c
>> /bin/ksh ../libtool --tag=CC    --mode=link gcc  -g -O2    -o test  
>> test.o libgslsys.la ../ieee-utils/libgslieeeutils.la ../err/ 
>> libgslerr.la ../test/libgsltest.la libgslsys.la ../utils/ 
>> libutils.la -lm
>> gcc -g -O2 -o test test.o  ../ieee-utils/.libs/ 
>> libgslieeeutils.a ../err/.libs/libgslerr.a ../test/.libs/ 
>> libgsltest.a ./.libs/libgslsys.a ../utils/.libs/libutils.a -lm
>> make  check-TESTS
>> FAIL: gsl_isinf(-inf) (1 observed vs -1 expected) [112]
>> FAIL: test
>> ===================
>> 1 of 1 tests failed
>> ===================
>>
>> Fix attached.
>>
>> NetBSD 4.99.31 i386
>> cc (GCC) 4.1.3 20070620 prerelease (NetBSD nb1 20070620)
>>
>> With fix all tests pass.
>>
>> --- sys/test.c.orig	2007-06-11 12:57:34.000000000 +0400
>> +++ sys/test.c	2007-06-11 12:57:52.000000000 +0400
>> @@ -383,7 +383,7 @@
>>         negative).  (This is all that C99 requires.) */
>>
>>      s = gsl_isinf (-inf);
>> -    gsl_test_int (s, -1, "gsl_isinf(-inf)");
>> +    gsl_test_int (s, 1, "gsl_isinf(-inf)");
>>
>>      s = gsl_isinf (nan);
>>      gsl_test_int (s, 0, "gsl_isinf(nan)");
>
> 	The following alternate patch will fix the test failure, without  
> modifying the test cases.
> --- /Users/emac/Desktop/infnan.c        2007-09-12  
> 22:37:19.000000000 +0200
> +++ /Softsys/gsl-1.9.90/sys/infnan.c    2007-09-12  
> 22:44:30.000000000 +0200
> @@ -118,7 +118,11 @@
>  int
>  gsl_isinf (const double x)
>  {
> -    return isinf(x);
> +       int result = isinf(x);
> +       if (result > 0 && x < 0) {
> +               result = - result;
> +       }
> +       return result;
>  }
>  # else
> Note: the one on the desktop is the unmodified file.
> Also, with make -k check all other test cases pass.
>
	 OOPS forgot specifying My OS: Mac OS X 10.4.10 (darwin) G4 PowerPC  
(emac)
-------------------------------------
This sig is dedicated to the advancement of Nuclear Power
Tommy Nordgren
tommy.nordgren@comhem.se

Re: test release gsl-1.9.90.tar.gz

[Oliver Jennrich]

On 9/11/07, Brian Gough <bjg@network-theory.co.uk> wrote:
> I have made a test release for gsl-1.10.

Compiles fine on cygwin:

uname -m = i686
uname -r = 1.5.24(0.156/4/2)
uname -s = CYGWIN_NT-5.1
uname -v = 2007-01-31 10:57

gcc version 4.2.1

CFLAGS = -O2

make check fails, though:

make[2]: Entering directory `/home/ojennric/temp/gsl-1.9.90/sys'
FAIL: gsl_isinf(-inf) (1 observed vs -1 expected) [112]
FAIL: test.exe

-- 
Space -- the final frontier

Re: test release gsl-1.9.90.tar.gz

[Aleksej Saushev]

Tommy Nordgren <tommy.nordgren@comhem.se> writes:

>> FAIL: gsl_isinf(-inf) (1 observed vs -1 expected) [112]
>> FAIL: test

> 	The following alternate patch will fix the test failure,
> without  modifying the test cases.
> --- /Users/emac/Desktop/infnan.c        2007-09-12
> 22:37:19.000000000  +0200
> +++ /Softsys/gsl-1.9.90/sys/infnan.c    2007-09-12
> 22:44:30.000000000  +0200
> @@ -118,7 +118,11 @@
>  int
>  gsl_isinf (const double x)
>  {
> -    return isinf(x);
> +       int result = isinf(x);
> +       if (result > 0 && x < 0) {
> +               result = - result;
> +       }
> +       return result;
>  }
>  # else
> Note: the one on the desktop is the unmodified file.
> Also, with make -k check all other test cases pass.

As for me, I don't like the proposal.

It is clear, that the test case is wrong, since in C truth is
denoted by "1", and not by "-1".  This is more convenient from
scientist's point of view too.

If you really want to fix test code, you'd better make it deal
with standard semantic:

   The isinf() macro returns a non-zero value if the value of x
   is an infinity.  Otherwise 0 is returned.

Obviously, test code doesn't behave well, when isinf returns
12345 as its value.

Re: test release gsl-1.9.90.tar.gz

[Tommy Nordgren]

On 13 sep 2007, at 10.03, Aleksej Saushev wrote:

> Tommy Nordgren <tommy.nordgren@comhem.se> writes:
>
>>> FAIL: gsl_isinf(-inf) (1 observed vs -1 expected) [112]
>>> FAIL: test
>
>> 	The following alternate patch will fix the test failure,
>> without  modifying the test cases.
>> --- /Users/emac/Desktop/infnan.c        2007-09-12
>> 22:37:19.000000000  +0200
>> +++ /Softsys/gsl-1.9.90/sys/infnan.c    2007-09-12
>> 22:44:30.000000000  +0200
>> @@ -118,7 +118,11 @@
>>  int
>>  gsl_isinf (const double x)
>>  {
>> -    return isinf(x);
>> +       int result = isinf(x);
>> +       if (result > 0 && x < 0) {
>> +               result = - result;
>> +       }
>> +       return result;
>>  }
>>  # else
>> Note: the one on the desktop is the unmodified file.
>> Also, with make -k check all other test cases pass.
>
> As for me, I don't like the proposal.
>
> It is clear, that the test case is wrong, since in C truth is
> denoted by "1", and not by "-1".  This is more convenient from
> scientist's point of view too.
>
> If you really want to fix test code, you'd better make it deal
> with standard semantic:
>
>    The isinf() macro returns a non-zero value if the value of x
>    is an infinity.  Otherwise 0 is returned.
>
> Obviously, test code doesn't behave well, when isinf returns
> 12345 as its value.
	It's still necessary to fix 'isfnan.c' somewhere though. There are  
two alternative
conditionally compiled versions of gsl_isinf.
	Here is an alternate fix, that will pass your updated test case,  
when isinf is not defined on the system.
--- /Users/emac/Desktop/infnan.c        2007-09-13 11:33:25.000000000  
+0200
+++ /Softsys/gsl-1.9.90/sys/infnan.c    2007-09-13 11:34:57.000000000  
+0200
@@ -127,7 +127,7 @@
  {
    if (! gsl_finite(x) && ! gsl_isnan(x))
      {
-      return (x > 0 ? +1 : -1);
+      return 1;
      }
    else
      {

------
What is a woman that you forsake her, and the hearth fire and the  
home acre,
to go with the old grey Widow Maker.  --Kipling, harp song of the  
Dane women
Tommy Nordgren
tommy.nordgren@comhem.se

Re: test release gsl-1.9.90.tar.gz

[Brian Gough]

At Wed, 12 Sep 2007 21:25:13 +0200,
Oliver Jennrich wrote:
> make check fails, though:
> 
> make[2]: Entering directory `/home/ojennric/temp/gsl-1.9.90/sys'
> FAIL: gsl_isinf(-inf) (1 observed vs -1 expected) [112]

Thanks, I've fixed the test to check for a non-zero value, rather than
-1.

-- 
Brian Gough

Re: test release gsl-1.9.90.tar.gz

[Brian Gough]

At Tue, 11 Sep 2007 12:39:40 -0600,
Patrick Alken wrote:
> compile goes fine, make check gives this error:
> 
> make[2]: Entering directory `/home/palken/gsl-1.9.90/specfunc'
> FAIL: gsl_sf_legendre_sphPlm_e(140,135,0.99998689456491752,&r) [2229]
>   expected: -6.542652532690932e-305
>   obtained:                    0 +/- 0 (rel=nan)
>   fracdiff:                    1

I've commented out this test, as it's on the borderline of underflow -
the initial values in the recurrence are less than 2.22e-308.

-- 
Brian Gough

Re: test release gsl-1.9.90.tar.gz

[Patrick Alken]

I am wondering if maybe the version of gcc (3.2.3) is causing the
problem. I tried make check again on another 64 bit machine with gcc
4.1.1 and everything went ok.

On Thu, Sep 13, 2007 at 02:45:41PM +0100, Brian Gough wrote:
> At Tue, 11 Sep 2007 12:39:40 -0600,
> Patrick Alken wrote:
> > compile goes fine, make check gives this error:
> > 
> > make[2]: Entering directory `/home/palken/gsl-1.9.90/specfunc'
> > FAIL: gsl_sf_legendre_sphPlm_e(140,135,0.99998689456491752,&r) [2229]
> >   expected: -6.542652532690932e-305
> >   obtained:                    0 +/- 0 (rel=nan)
> >   fracdiff:                    1
> 
> I've commented out this test, as it's on the borderline of underflow -
> the initial values in the recurrence are less than 2.22e-308.
> 
> -- 
> Brian Gough
> 

Re: test release gsl-1.9.90.tar.gz

[Brian Gough]

At Thu, 13 Sep 2007 10:30:14 -0600,
Patrick Alken wrote:
> I am wondering if maybe the version of gcc (3.2.3) is causing the
> problem. I tried make check again on another 64 bit machine with gcc
> 4.1.1 and everything went ok.

It's possible, but it doesn't necessarily indicate a problem with the
compiler itself.  It depends whether the value was computed with SSE
or X87 instructions.

The starting values of the recurrence were O(1e-310).  These would
underflow to zero in SSE arithmetic, which does not have denormalised
numbers, so the final result would be zero.  With X87 instructions,
there wouldn't be any underflow, and the final result comes out ok.

-- 
Brian Gough