headers
Brian Gough | 1 Jun 2010 18:34
Picon

Re: gsl-config --libs incorrectly returns -lm on Haiku.

At Wed, 26 May 2010 12:35:05 +0100,
Jerome Kelleher wrote:
> > I think the patch enclosed would fix this problem, but I haven't tested
> > it on any Haiku machine. If you could please try it out and it will be
> > easier for Brian to consider the patch.
> > 
> Thanks for the patch. I've applied it to the latest git version (gsl-1.14+), 
> ran make dist and tried it on the Haiku VM. It works perfectly.
> 

I have applied the patch. Thanks Peter.

--

-- 
Brian Gough
Permalink | Reply | 0 comments |
headers
Huan Wu | 9 Jun 2010 00:06
Favicon

RE: gsl_linalg_complex_cholesky_invert

Please find attached the Cholesky invert for complex matrices and the testing functions.
I used gcc to compile and test it.

Regards,
Huan

-----Original Message-----
From: Brian Gough [mailto:bjg <at> gnu.org] 
Sent: Monday, May 24, 2010 1:56 PM
To: Huan Wu
Cc: Patrick Alken
Subject: Re: gsl_linalg_complex_cholesky_invert

At Fri, 14 May 2010 11:18:37 -0400,
Huan Wu wrote:
> I noticed that the Cholesky invert function for complex Hermitian matrix
> is not available yet from release 1.14. I'm working on a project that
> needs it so I did the extension of the available function for real
> symmetric matrix (gsl_linalg_cholesky_invert, added to GSL in release
> 1.12), to the one for complex Hermitian matrix. I'd like to contribute
> the new one (gsl_linalg_complex_cholesky_invert) to the GSL but not sure
> if anybody else has already planned to do that (forgive my ignorance to
> the GSL community if I missed any should-to-know information as a first
> time user).
> Thanks,
> 
> Huan
> 
> Huan Wu
>
(Continue reading)

Permalink | Reply | 1 comment |
headers
Peter Johansson | 11 Jun 2010 04:16
Picon

autogen.sh overwrites file INSTALL

Hi,

I noticed that when I run `./autogen.sh' it will run delete the file 
`INSTALL' and instead create a link to 
`/usr/local/share/automake-1.11/INSTALL'. This is because automake is 
run with '--foce-missing'. If I instead run 'autoreconf -ivf' as 
recommended by autogen.sh, the same thing happens: the INSTALL file is 
replaced by a link to...

I understand the --force option is desired in order to avoid missing 
upgrades of other automake provided files such as the script missing. On 
the other hand, the current behavior is a bit annoying as it took me 
some time to get back the INSTALL file (not very familiar with git:( ), 
and I figure the risk that the generic FSF INSTALL goes into a release 
tarball rather than GSL's INSTALL is quite significant.

AFAIK there is no option in Automake to '--install' only some files and 
not, e.g., `INSTALL'. The simplest solution, IMVHO, would be to rename 
the file INSTALL and distribute both this renamed file together with the 
generic INSTALL. But it's likely that the Automake people have a better 
solution.

Thanks,
Peter

--

-- 
Peter Johansson

svndigest maintainer, http://dev.thep.lu.se/svndigest
yat maintainer,       http://dev.thep.lu.se/yat
(Continue reading)

Permalink | Reply | 4 comments |
headers
Brian Gough | 13 Jun 2010 17:40
Picon

Re: gsl_linalg_complex_cholesky_invert

At Tue, 8 Jun 2010 18:06:52 -0400,
Huan Wu wrote:
> 
> [1  <text/plain; us-ascii (quoted-printable)>]
> Please find attached the Cholesky invert for complex matrices and the testing functions.
> I used gcc to compile and test it.
> 

Thank you. I have committed it to the repository.

--

-- 
Brian Gough
Permalink | Reply | 0 comments |
headers
Brian Gough | 13 Jun 2010 22:16
Picon

Re: autogen.sh overwrites file INSTALL

At Thu, 10 Jun 2010 22:16:58 -0400,
Peter Johansson wrote:
> I noticed that when I run `./autogen.sh' it will run delete the file 
> `INSTALL' and instead create a link to 
> `/usr/local/share/automake-1.11/INSTALL'. This is because automake is 
> run with '--foce-missing'. If I instead run 'autoreconf -ivf' as 
> recommended by autogen.sh, the same thing happens: the INSTALL file is 
> replaced by a link to...
> 
> I understand the --force option is desired in order to avoid missing 
> upgrades of other automake provided files such as the script missing. On 
> the other hand, the current behavior is a bit annoying as it took me 
> some time to get back the INSTALL file (not very familiar with git:( ), 
> and I figure the risk that the generic FSF INSTALL goes into a release 
> tarball rather than GSL's INSTALL is quite significant.
> 
> AFAIK there is no option in Automake to '--install' only some files and 
> not, e.g., `INSTALL'. The simplest solution, IMVHO, would be to rename 
> the file INSTALL and distribute both this renamed file together with the 
> generic INSTALL. But it's likely that the Automake people have a better 
> solution.

I agree, it's a minor annoyance that it get overwritten.  I've been
leaving it alone because there are existing references to the GSL
specific info in INSTALL out there which can't be changed (e.g. in the
printed manual).  I think the best thing would be for automake to
handle the INSTALL file in a different way (for example, only updating
the generic part), the generic instructions aren't always that useful
by themselves.
(Continue reading)

Permalink | Reply | 3 comments |
headers
Luciano Ribichini | 16 Jun 2010 14:16
Picon
Favicon

Trascendental funtion (a check suggested by Monniaux)

Hallo GSL team,

I am studying the GSL with the goal of learning to avoid pitfall in computer science.
I found a very nice article by Monniaux on the arxiv server cs/0701192 and on page 18
I found a very nice check of the sinus function.

y=sin(14885392687.0)

I wrote a very small program, and I got the wrong result

y=1.48E-10

I am using gsl 1.11 on Debian 5.0.x 64bit.
Maybe you already corrected this.
Neverthless I report it anyway since, due to cross-dependecies, 
it seems problematic for me to remove gsl from my system, and upgrade.

Last but not least:
Thanks for your fine piece of software.

Cheers
Luciano Ribichini
Permalink | Reply | 3 comments |
headers
Brian Gough | 17 Jun 2010 19:59
Picon

Re: Trascendental funtion (a check suggested by Monniaux)

At Wed, 16 Jun 2010 14:16:47 +0200,
Luciano Ribichini wrote:
> I am studying the GSL with the goal of learning to avoid pitfall in computer science.
> I found a very nice article by Monniaux on the arxiv server cs/0701192 and on page 18
> I found a very nice check of the sinus function.
> 
> y=sin(14885392687.0)
> 
> I wrote a very small program, and I got the wrong result
> 
> y=1.48E-10
> 

Thanks for the email. What sin function do you use - gsl_sf_sin?

This is what I get in Pari with 64-digits of precision:

$ gp
? default(realprecision,64)
%1 = 64

? 14885392687.0/Pi
%3 = 4738167652.000000000047103786916527482860432105403202244461035583
? 14885392687.0-4738167652*Pi
%4 = 1.479809109332217594562699619166045849796144302348 E-10
? sin(14885392687.0-4738167652*Pi)
%5 = 1.479809109332217594557298722864333901077382651758 E-10

Maybe you can explain the difference from the paper for me.
(Continue reading)

Permalink | Reply | 2 comments |
headers
Xuebin Wu | 17 Jun 2010 23:37
Picon

Re: Trascendental funtion (a check suggested by Monniaux)

Hi,

What is gp? Is it some software that can compute with arbitary precision?

I write some program with gmp, and now I have to admit that 1.48E10 is the
more accurate answer now.
It is awkward that the library sinl() function, which works on long double
data, has a less accuracy than sin()...

On Thu, Jun 17, 2010 at 10:59 AM, Brian Gough <bjg <at> gnu.org> wrote:

> At Wed, 16 Jun 2010 14:16:47 +0200,
> Luciano Ribichini wrote:
> > I am studying the GSL with the goal of learning to avoid pitfall in
> computer science.
> > I found a very nice article by Monniaux on the arxiv server cs/0701192
> and on page 18
> > I found a very nice check of the sinus function.
> >
> > y=sin(14885392687.0)
> >
> > I wrote a very small program, and I got the wrong result
> >
> > y=1.48E-10
> >
>
> Thanks for the email. What sin function do you use - gsl_sf_sin?
>
> This is what I get in Pari with 64-digits of precision:
>
(Continue reading)

Permalink | Reply | 1 comment |
headers
Peter Johansson | 20 Jun 2010 04:26
Picon

Re: autogen.sh overwrites file INSTALL

Hello,

[adding automake; replies can drop bug-gsl]

This is

http://lists.gnu.org/archive/html/bug-gsl/2010-06/msg00002.html

On 6/13/10 4:16 PM, Brian Gough wrote:
> At Thu, 10 Jun 2010 22:16:58 -0400,
> Peter Johansson wrote:
>    
>> AFAIK there is no option in Automake to '--install' only some files and
>> not, e.g., `INSTALL'. The simplest solution, IMVHO, would be to rename
>> the file INSTALL and distribute both this renamed file together with the
>> generic INSTALL. But it's likely that the Automake people have a better
>> solution.
>>      
> I agree, it's a minor annoyance that it get overwritten.  I've been
> leaving it alone because there are existing references to the GSL
> specific info in INSTALL out there which can't be changed (e.g. in the
> printed manual).  I think the best thing would be for automake to
> handle the INSTALL file in a different way (for example, only updating
> the generic part), the generic instructions aren't always that useful
> by themselves.
>
>    
Perhaps someone with more experience with Automake knows some trick to 
achieve what we want.
(Continue reading)

Permalink | Reply | 2 comments |
headers
Maurício Antunes | 21 Jun 2010 01:08
Picon

CBLAS_INDEX

Hi,

I see that CBLAS_INDEX is defined in cblas.h using the preprocessor:

    /*
     * Enumerated and derived types
     */
    #define CBLAS_INDEX size_t  /* this may vary between platforms */

Would you mind changing that #define to a typedef? I've just run
into a small problem writing a language binding (to Haskell) that
I believe others may run into as well. I use an actual C program to
generate those bindings, and it contains macros working like that:

    #define make_equivalent_type c_type \
       do_lots_of_stuff...

    make_equivalent_type (CBLAS_INDEX)

This macro writes code that checks that type size, and also use
its name as a string to be used to actually write bindings files.
But since it's defined with the preprocessor, the type ended up
beeing considered a redefinition of size_t.

If there's no problem with the typedef, maybe it could help
prevent others from equivalent problems.

Thanks for your attention,

Maurício
(Continue reading)

Permalink | Reply | 1 comment |

Gmane