[gentoo-science] blas/lapack stuff

[gentoo-science] blas/lapack stuff

[Andrey G. Grozin]

Hello *,

I use the science overlay, and for quite some time I have problems with 
blas/lapack. Each attempt to emerge -uD @world produces something like

================================================================================
[blocks B      ] app-admin/eselect-blas ("app-admin/eselect-blas" is 
blocking sci-libs/lapack-reference-3.3.1-r1, sci-libs/gsl-1.15-r1, 
sci-libs/blas-reference-20110418)
[blocks B      ] app-admin/eselect-cblas ("app-admin/eselect-cblas" is 
blocking sci-libs/lapack-reference-3.3.1-r1, sci-libs/gsl-1.15-r1, 
sci-libs/blas-reference-20110418)
[blocks B      ] app-admin/eselect-lapack ("app-admin/eselect-lapack" is 
blocking sci-libs/lapack-reference-3.3.1-r1, sci-libs/gsl-1.15-r1, 
sci-libs/blas-reference-20110418)

Total: 21 packages (21 upgrades), Size of downloads: 19,795 kB
Conflict: 4 blocks (3 unsatisfied)
Portage tree and overlays:
  [0] /usr/portage
  [1] /var/lib/layman/science

  * Error: The above package list contains packages which cannot be
  * installed at the same time on the same system.

   (sci-libs/lapack-reference-3.3.1-r1::science, ebuild scheduled for 
merge) pulled in by
     >=sci-libs/lapack-reference-3.3 required by 
(virtual/lapack-3.3::science, ebuild scheduled for merge)

   (app-admin/eselect-cblas-0.1::gentoo, installed) pulled in by
     app-admin/eselect-cblas required by 
(sci-libs/blas-atlas-3.9.23-r4::gentoo, installed)

   (sci-libs/blas-reference-20110418::science, ebuild scheduled for merge) 
pulled in by
     >=sci-libs/blas-reference-20110417 required by 
(virtual/blas-2.1::science, ebuild scheduled for merge)

   (app-admin/eselect-blas-0.1::gentoo, installed) pulled in by
     app-admin/eselect-blas required by 
(sci-libs/blas-atlas-3.9.23-r4::gentoo, installed)

   (sci-libs/gsl-1.15-r1::science, ebuild scheduled for merge) pulled in by
     sci-libs/gsl required by (sci-mathematics/giac-0.9.3::grozin, 
installed)
     sci-libs/gsl required by (sci-libs/mathgl-1.11.2::gentoo, installed)
     sci-libs/gsl required by (sci-libs/tamu_anova-0.2.1::gentoo, 
installed)
     sci-libs/gsl required by (sci-visualization/qtiplot-0.9.8.8::gentoo, 
installed)
     sci-libs/gsl required by (media-gfx/asymptote-2.13::grozin, installed)
     sci-libs/gsl required by (media-gfx/inkscape-0.48.2::gentoo, 
installed)
     sci-libs/gsl required by (app-office/kspread-2.3.3::gentoo, installed)
     >=sci-libs/gsl-1.10 required by 
(sci-visualization/pyxplot-0.8.4::gentoo, installed)
     >=sci-libs/gsl-1.14-r2 required by (virtual/cblas-2.0::science, ebuild 
scheduled for merge)

   (app-admin/eselect-lapack-0.1::gentoo, installed) pulled in by
     app-admin/eselect-lapack required by 
(sci-libs/lapack-atlas-3.9.23-r3::gentoo, installed)
================================================================================
Any ideas?

Andrey

Re: [gentoo-science] blas/lapack stuff

[Andrea Arteaga]

Hello Andrey,

the (c)blas/lapack framework has changed a bit. The eselect-blas,
eselect-lapack, eselect-* packages are now deprecated. In order to
access the new versions of these libraries and the new eselect package
you have to remove the old ones (I mean eselect-*), install
eselect-1.2.18-r1, blas-reference-20110418 (or any other
implementation of your choice, like atlas of openblas). In general,
you can safely emerge the blas implementations listed in virtual/blas
[1], virtual/cblas, virtual/lapack,...

Notice that most of these packages are ~keyworded, so if you want to
stay with stable packages you will be unable to install the newest
versions of blas and lapack, as they need the patched eselect.

Also notice that now atlas-3.9.* provides blas, cblas and lapack;
blas-atlas, cblas-atlas are not needed anymore.

This also applies to blacs, lapacke, scalapack, mpi.

Cheers
Andrea

[1] http://goo.gl/P8wPN

[gentoo-science] sci-mathematics/reduce on amd64

[Andrey G. Grozin]

Hello *,

Could somebody with amd64 check if sci-mathematics/reduce-20110414 
compiles on this arch? You'll have to keyword it. Upstream says it should 
compile.

Andrey

Re: [gentoo-science] sci-mathematics/reduce on amd64

[Kacper Kowalik]

[-- Attachment #1: Type: text/plain, Size: 492 bytes --]

On 12/21/11 11:39, Andrey G. Grozin wrote:
> Hello *,
> 
> Could somebody with amd64 check if sci-mathematics/reduce-20110414
> compiles on this arch? You'll have to keyword it. Upstream says it
> should compile.

It does and passes test suite. Though it doesn't respect libdir:

/dev/shm/portage/sci-mathematics/reduce-20110414/image/usr/lib/reduce/reduce.fonts/README

Judging by the content it shouldn't be libdir in the first place, rather
/usr/share...
Cheers,
Kacper



[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 900 bytes --]

Re: [gentoo-science] sci-mathematics/reduce on amd64

[Miguel de Val-Borro]

On Wed, Dec 21, 2011 at 12:01:18PM +0100, Kacper Kowalik wrote:
> On 12/21/11 11:39, Andrey G. Grozin wrote:
> > Hello *,
> > 
> > Could somebody with amd64 check if sci-mathematics/reduce-20110414
> > compiles on this arch? You'll have to keyword it. Upstream says it
> > should compile.
> 
> It does and passes test suite. Though it doesn't respect libdir:
> 
> /dev/shm/portage/sci-mathematics/reduce-20110414/image/usr/lib/reduce/reduce.fonts/README
> 
> Judging by the content it shouldn't be libdir in the first place, rather
> /usr/share...

There is also a warning about an unrecognized configure option:
configure: WARNING: unrecognized options: --with-gui

Here is the full build log on amd64:
http://paste.pocoo.org/show/523967/

Miguel

Re: [gentoo-science] sci-mathematics/reduce on amd64

[Andrey G. Grozin]

On Wed, 21 Dec 2011, Miguel de Val-Borro wrote:
> There is also a warning about an unrecognized configure option:
> configure: WARNING: unrecognized options: --with-gui
This package is a QA nightmare :-( It bundles modified sources of 
x11-libs/fox (it does not compile with the system fox). First, it compiles 
this bundled fox, and installs it somewhere inside the build tree. Then it 
configures csl (and thus finds this freshly installed fox) and compiles 
it. Then it compiles reduce with csl. During this process, autoconf is 
called from make ("maintainer mode") many times. And each time, the 
arguments of the top-level configure are passed to all configure scripts 
in subdirectories. --with-gui is needed by one configure script (this is 
controlled by the USE flag X; with -X, fox is not built, and csl and 
reduce work at the command line only). But this --with-gui is passed to 
all configure scripts in various directories (fox sources, for example), 
and these scripts don't recognize this option. Ugly, but this works.

> Here is the full build log on amd64:
> http://paste.pocoo.org/show/523967/
Many thanks. Looks OK.

I really want reduce to become more popular. There are only 3 free 
computer algebra systems: maxima, reduce, axiom (I don't count toys like 
yacas; and axiom has forked into axiom, openaxiom, and fricas). reduce is 
mature (from late 60s), powerful and very efficient. So, I have to live 
somehow with the weirdness of its current build system.

I suppose I can keyword it ~amd64, so that more users will have a chance 
to install it.

Andrey

Re: [gentoo-science] sci-mathematics/reduce on amd64

[Burcin Erocal]

On Wed, 21 Dec 2011 20:53:27 +0700 (NOVT)
"Andrey G. Grozin" <A.G.Grozin@inp.nsk.su> wrote:

> I really want reduce to become more popular. There are only 3 free 
> computer algebra systems: maxima, reduce, axiom (I don't count toys
> like yacas; and axiom has forked into axiom, openaxiom, and fricas).
> reduce is mature (from late 60s), powerful and very efficient. So, I
> have to live somehow with the weirdness of its current build system.

There is also Sage:

http://sagemath.org/

It's far from a "toy" or a specialized system (such as pari/gp,
Singular, Macaulay2, etc.). So it should be mentioned.


Cheers,
Burcin

Re: [gentoo-science] sci-mathematics/reduce on amd64

[Andrey G. Grozin]

On Wed, 21 Dec 2011, Burcin Erocal wrote:
> There is also Sage:
>
> http://sagemath.org/
>
> It's far from a "toy" or a specialized system (such as pari/gp,
> Singular, Macaulay2, etc.). So it should be mentioned.
Of course, it's not a toy. To a large degree, it's a large collection of 
existing software, "Gentoo of maths", with its own package system. And it 
includes maxima, fricas, pari/gp, gap, singular, R, and many other 
systems. They are all integrated together to some degree, so, sage is 
somewhere in between a monolythic CAS (maxima, reduce, fricas,...) and 
just a collection.

It would be interesting to make reduce one of its packages, too.

Andrey

Re: [gentoo-science] sci-mathematics/reduce on amd64

[Christopher Schwan]

On Wednesday 21 December 2011 12:13:17 Miguel de Val-Borro wrote:
> On Wed, Dec 21, 2011 at 12:01:18PM +0100, Kacper Kowalik wrote:
> > On 12/21/11 11:39, Andrey G. Grozin wrote:
> > > Hello *,
> > > 
> > > Could somebody with amd64 check if sci-mathematics/reduce-20110414
> > > compiles on this arch? You'll have to keyword it. Upstream says it
> > > should compile.
> > 
> > It does and passes test suite. Though it doesn't respect libdir:
> > 
> > /dev/shm/portage/sci-mathematics/reduce-20110414/image/usr/lib/reduce/re
> > duce.fonts/README
> > 
> > Judging by the content it shouldn't be libdir in the first place, rather
> > /usr/share...
> 
> There is also a warning about an unrecognized configure option:
> configure: WARNING: unrecognized options: --with-gui
> 
> Here is the full build log on amd64:
> http://paste.pocoo.org/show/523967/
> 
> Miguel

I also experienced a (random) build failure, see 
http://paste.pocoo.org/show/524020/ . I guess the build failure is caused by 
parallel make, using MAKEOPTS=-j1 I was able to compile reduce. This showed 
some QA failures, see http://paste.pocoo.org/show/524007/ .

Christopher

Re: [gentoo-science] blas/lapack stuff

[Andrey G. Grozin]

Hello *,

I don't understand how the blas stuff (from the science overlay) works. I 
have blas-reference and atlas (and also eigen) installed, atlas eselected. 
But /usr/lib/libblas.so is a file (not symlink) owned by blas-reference. 
So, a program demanding -lblas will be linked to this library from 
blas-reference, right? And what eselect blas then is doing?

This question is related to the new gsl-shell package. Its Makefile just 
says -lblas. It works on my computer (linking to /usr/lib/libblas.so from 
blas-reference, I suppose). But what should I write as the dependency? 
virtual/blas? (docs say that atlas should be OK).

Thanks in advance,
Andrey

Re: [gentoo-science] blas/lapack stuff

[Kacper Kowalik]

[-- Attachment #1: Type: text/plain, Size: 778 bytes --]

On 09/26/2012 05:09 AM, Andrey G. Grozin wrote:
> Hello *,
> 
> I don't understand how the blas stuff (from the science overlay) works.
> I have blas-reference and atlas (and also eigen) installed, atlas
> eselected. But /usr/lib/libblas.so is a file (not symlink) owned by
> blas-reference. So, a program demanding -lblas will be linked to this
> library from blas-reference, right? And what eselect blas then is doing?
> 
> This question is related to the new gsl-shell package. Its Makefile just
> says -lblas. It works on my computer (linking to /usr/lib/libblas.so
> from blas-reference, I suppose). But what should I write as the
> dependency? virtual/blas? (docs say that atlas should be OK).
Change -lblas into $(pkg-config --libs blas)
Cheers,
Kacper


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 900 bytes --]

Re: [gentoo-science] blas/lapack stuff

[Francois Bissey]

On 26/09/12 17:52, Kacper Kowalik wrote:
> On 09/26/2012 05:09 AM, Andrey G. Grozin wrote:
>> Hello *,
>>
>> I don't understand how the blas stuff (from the science overlay) works.
>> I have blas-reference and atlas (and also eigen) installed, atlas
>> eselected. But /usr/lib/libblas.so is a file (not symlink) owned by
>> blas-reference. So, a program demanding -lblas will be linked to this
>> library from blas-reference, right? And what eselect blas then is doing?
>>
>> This question is related to the new gsl-shell package. Its Makefile just
>> says -lblas. It works on my computer (linking to /usr/lib/libblas.so
>> from blas-reference, I suppose). But what should I write as the
>> dependency? virtual/blas? (docs say that atlas should be OK).
> Change -lblas into $(pkg-config --libs blas)
> Cheers,
> Kacper
> 
Just to add to Kacper's answer. eselect changes the link to the blas
pkg-config file. Any ebuild using blas/cblas/lapack should be configured
using the values from pkg-config.
If Sebastian had pushed it a little bit more we could have had
/usr/lib(64)/libblas-reference.so and no libblas whatsoever. Which may
be an idea just to catch packages that just do -lblas.

Francois

Re: [gentoo-science] blas/lapack stuff

[Sébastien Fabbro]

On Wed, Sep 26, 2012 at 4:20 AM, Francois Bissey
<fbissey@slingshot.co.nz> wrote:

> If Sebastian had pushed it a little bit more we could have had
> /usr/lib(64)/libblas-reference.so and no libblas whatsoever. Which may
> be an idea just to catch packages that just do -lblas.

Which I did for ealier versions of {blas,lapack}-reference, it use to
link to -lrefblas and -lreflapack.
I'll fix this again.

-- 
Sébastien

Re: [gentoo-science] blas/lapack stuff

[denis cohen]

On Wed, Sep 26, 2012 at 4:09 PM, Sébastien Fabbro <bicatali@gentoo.org> wrote:
> On Wed, Sep 26, 2012 at 4:20 AM, Francois Bissey
> <fbissey@slingshot.co.nz> wrote:
>
>> If Sebastian had pushed it a little bit more we could have had
>> /usr/lib(64)/libblas-reference.so and no libblas whatsoever. Which may
>> be an idea just to catch packages that just do -lblas.
>
> Which I did for ealier versions of {blas,lapack}-reference, it use to
> link to -lrefblas and -lreflapack.
> I'll fix this again.
>
> --
> Sébastien
>


Hello,

I have installed sci-libs/blas-cblas-lapack
and also sci-libs/atlas
as mentioned in an earlier email.
However, I can't eselect which flavor of blas-cblas-lapack to use
because eselect does not find these modules, i.e.

# eselect lapack list
!!! Error: Can't load module lapack
exiting

If I try to emerge eselect-lapack for example I get:

# emerge -vp eselect-lapack

These are the packages that would be merged, in order:

Calculating dependencies... done!
[ebuild  N     ] app-admin/eselect-lapack-0.1  0 kB
[blocks B      ] app-admin/eselect-lapack ("app-admin/eselect-lapack"
is blocking sci-libs/gsl-1.15-r2, sci-libs/cblas-reference-20110218,
sci-libs/blas-reference-20110418, sci-libs/atlas-3.10.0,
sci-libs/lapack-reference-3.3.1-r1)

Total: 1 package (1 new), Size of downloads: 0 kB
Conflict: 1 block (1 unsatisfied)

 * Error: The above package list contains packages which cannot be
 * installed at the same time on the same system.

  (sci-libs/cblas-reference-20110218::science, installed) pulled in by
    sci-libs/cblas-reference required by @selected

  (app-admin/eselect-lapack-0.1::gentoo, ebuild scheduled for merge)
pulled in by
    eselect-lapack

  (sci-libs/atlas-3.10.0::science, installed) pulled in by
    sci-libs/atlas required by @selected


For more information about Blocked Packages, please refer to the following
section of the Gentoo Linux x86 Handbook (architecture is irrelevant):

http://www.gentoo.org/doc/en/handbook/handbook-x86.xml?full=1#blocked



I'd like to select the right packages. Any help appreciated.

Thanks

Denis

Re: [gentoo-science] blas/lapack stuff

[justin]

[-- Attachment #1: Type: text/plain, Size: 2486 bytes --]

On 10/10/12 14:28, denis cohen wrote:
> On Wed, Sep 26, 2012 at 4:09 PM, Sébastien Fabbro <bicatali@gentoo.org> wrote:
>> On Wed, Sep 26, 2012 at 4:20 AM, Francois Bissey
>> <fbissey@slingshot.co.nz> wrote:
>>
>>> If Sebastian had pushed it a little bit more we could have had
>>> /usr/lib(64)/libblas-reference.so and no libblas whatsoever. Which may
>>> be an idea just to catch packages that just do -lblas.
>>
>> Which I did for ealier versions of {blas,lapack}-reference, it use to
>> link to -lrefblas and -lreflapack.
>> I'll fix this again.
>>
>> --
>> Sébastien
>>
> 
> 
> Hello,
> 
> I have installed sci-libs/blas-cblas-lapack
> and also sci-libs/atlas
> as mentioned in an earlier email.
> However, I can't eselect which flavor of blas-cblas-lapack to use
> because eselect does not find these modules, i.e.
> 
> # eselect lapack list
> !!! Error: Can't load module lapack
> exiting
> 
> If I try to emerge eselect-lapack for example I get:
> 
> # emerge -vp eselect-lapack
> 
> These are the packages that would be merged, in order:
> 
> Calculating dependencies... done!
> [ebuild  N     ] app-admin/eselect-lapack-0.1  0 kB
> [blocks B      ] app-admin/eselect-lapack ("app-admin/eselect-lapack"
> is blocking sci-libs/gsl-1.15-r2, sci-libs/cblas-reference-20110218,
> sci-libs/blas-reference-20110418, sci-libs/atlas-3.10.0,
> sci-libs/lapack-reference-3.3.1-r1)
> 
> Total: 1 package (1 new), Size of downloads: 0 kB
> Conflict: 1 block (1 unsatisfied)
> 
>  * Error: The above package list contains packages which cannot be
>  * installed at the same time on the same system.
> 
>   (sci-libs/cblas-reference-20110218::science, installed) pulled in by
>     sci-libs/cblas-reference required by @selected
> 
>   (app-admin/eselect-lapack-0.1::gentoo, ebuild scheduled for merge)
> pulled in by
>     eselect-lapack
> 
>   (sci-libs/atlas-3.10.0::science, installed) pulled in by
>     sci-libs/atlas required by @selected
> 
> 
> For more information about Blocked Packages, please refer to the following
> section of the Gentoo Linux x86 Handbook (architecture is irrelevant):
> 
> http://www.gentoo.org/doc/en/handbook/handbook-x86.xml?full=1#blocked
> 
> 
> 
> I'd like to select the right packages. Any help appreciated.
> 
> Thanks
> 
> Denis
> 

Hi,

do you have the eselect version from the sci overlay installed? Only
that one is able to eslect blas for you correctly.

juistin


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 299 bytes --]

Re: [gentoo-science] blas/lapack stuff

[Thomas Kahle]

[-- Attachment #1: Type: text/plain, Size: 1699 bytes --]

On 17:10 Wed 10 Oct 2012, justin wrote:
[..]

> > Calculating dependencies... done!
> > [ebuild  N     ] app-admin/eselect-lapack-0.1  0 kB
> > [blocks B      ] app-admin/eselect-lapack ("app-admin/eselect-lapack"
> > is blocking sci-libs/gsl-1.15-r2, sci-libs/cblas-reference-20110218,
> > sci-libs/blas-reference-20110418, sci-libs/atlas-3.10.0,
> > sci-libs/lapack-reference-3.3.1-r1)
> > 
> > Total: 1 package (1 new), Size of downloads: 0 kB
> > Conflict: 1 block (1 unsatisfied)
> > 
> >  * Error: The above package list contains packages which cannot be
> >  * installed at the same time on the same system.
> > 
> >   (sci-libs/cblas-reference-20110218::science, installed) pulled in by
> >     sci-libs/cblas-reference required by @selected
> > 
> >   (app-admin/eselect-lapack-0.1::gentoo, ebuild scheduled for merge)
> > pulled in by
> >     eselect-lapack
> > 
> >   (sci-libs/atlas-3.10.0::science, installed) pulled in by
> >     sci-libs/atlas required by @selected
> > 
> > 

> do you have the eselect version from the sci overlay installed? Only
> that one is able to eslect blas for you correctly.

I'm using this eselect version and I have lapack reference-3.4.2
installed and eselected.  Still I get:

# pkg-config --libs lapack                                                        
Package lapack was not found in the pkg-config search path.
Perhaps you should add the directory containing `lapack.pc'
to the PKG_CONFIG_PATH environment variable
No package 'lapack' found

Which makes packages depending on lapack checks during configure fail.

Any ideas?

Cheers,
Thomas




-- 
Thomas Kahle
http://dev.gentoo.org/~tomka/

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 316 bytes --]

Re: [gentoo-science] blas/lapack stuff

[Francois Bissey]

On 27/10/12 10:53, Thomas Kahle wrote:
> On 17:10 Wed 10 Oct 2012, justin wrote:
> [..]
> 
>>> Calculating dependencies... done!
>>> [ebuild  N     ] app-admin/eselect-lapack-0.1  0 kB
>>> [blocks B      ] app-admin/eselect-lapack ("app-admin/eselect-lapack"
>>> is blocking sci-libs/gsl-1.15-r2, sci-libs/cblas-reference-20110218,
>>> sci-libs/blas-reference-20110418, sci-libs/atlas-3.10.0,
>>> sci-libs/lapack-reference-3.3.1-r1)
>>>
>>> Total: 1 package (1 new), Size of downloads: 0 kB
>>> Conflict: 1 block (1 unsatisfied)
>>>
>>>  * Error: The above package list contains packages which cannot be
>>>  * installed at the same time on the same system.
>>>
>>>   (sci-libs/cblas-reference-20110218::science, installed) pulled in by
>>>     sci-libs/cblas-reference required by @selected
>>>
>>>   (app-admin/eselect-lapack-0.1::gentoo, ebuild scheduled for merge)
>>> pulled in by
>>>     eselect-lapack
>>>
>>>   (sci-libs/atlas-3.10.0::science, installed) pulled in by
>>>     sci-libs/atlas required by @selected
>>>
>>>
> 
>> do you have the eselect version from the sci overlay installed? Only
>> that one is able to eslect blas for you correctly.
> 
> I'm using this eselect version and I have lapack reference-3.4.2
> installed and eselected.  Still I get:
> 
> # pkg-config --libs lapack                                                        
> Package lapack was not found in the pkg-config search path.
> Perhaps you should add the directory containing `lapack.pc'
> to the PKG_CONFIG_PATH environment variable
> No package 'lapack' found
> 
> Which makes packages depending on lapack checks during configure fail.
> 
> Any ideas
> 

First of, supposing you are on amd64, does
/usr/lib64/pkgconfig/lapack.pc
exist? It should be a link pointing to
../../../etc/env.d/alternatives/lapack/_current/usr/lib64/pkgconfig/lapack.pc
itself a link to /usr/lib64/pkgconfig/reflapack.pc
It is possible that some of these links have not been created properly.
There used to be bug about it on prefix (I don't know if anyone improved
my prefix bandaid). There is also the possibility that remnant from the
non-sci overlay mechanism interfere.

Francois

Re: [gentoo-science] blas/lapack stuff

[Thomas Kahle]

[-- Attachment #1: Type: text/plain, Size: 2996 bytes --]

On 21:50 Sat 27 Oct 2012, Francois Bissey wrote:
> On 27/10/12 10:53, Thomas Kahle wrote:
> > On 17:10 Wed 10 Oct 2012, justin wrote:
> > [..]
> > 
> >>> Calculating dependencies... done!
> >>> [ebuild  N     ] app-admin/eselect-lapack-0.1  0 kB
> >>> [blocks B      ] app-admin/eselect-lapack ("app-admin/eselect-lapack"
> >>> is blocking sci-libs/gsl-1.15-r2, sci-libs/cblas-reference-20110218,
> >>> sci-libs/blas-reference-20110418, sci-libs/atlas-3.10.0,
> >>> sci-libs/lapack-reference-3.3.1-r1)
> >>>
> >>> Total: 1 package (1 new), Size of downloads: 0 kB
> >>> Conflict: 1 block (1 unsatisfied)
> >>>
> >>>  * Error: The above package list contains packages which cannot be
> >>>  * installed at the same time on the same system.
> >>>
> >>>   (sci-libs/cblas-reference-20110218::science, installed) pulled in by
> >>>     sci-libs/cblas-reference required by @selected
> >>>
> >>>   (app-admin/eselect-lapack-0.1::gentoo, ebuild scheduled for merge)
> >>> pulled in by
> >>>     eselect-lapack
> >>>
> >>>   (sci-libs/atlas-3.10.0::science, installed) pulled in by
> >>>     sci-libs/atlas required by @selected
> >>>
> >>>
> > 
> >> do you have the eselect version from the sci overlay installed? Only
> >> that one is able to eslect blas for you correctly.
> > 
> > I'm using this eselect version and I have lapack reference-3.4.2
> > installed and eselected.  Still I get:
> > 
> > # pkg-config --libs lapack                                                        
> > Package lapack was not found in the pkg-config search path.
> > Perhaps you should add the directory containing `lapack.pc'
> > to the PKG_CONFIG_PATH environment variable
> > No package 'lapack' found
> > 
> > Which makes packages depending on lapack checks during configure fail.
> > 
> > Any ideas
> > 
> 
> First of, supposing you are on amd64, does
> /usr/lib64/pkgconfig/lapack.pc
> exist? It should be a link pointing to
> ../../../etc/env.d/alternatives/lapack/_current/usr/lib64/pkgconfig/lapack.pc
> itself a link to /usr/lib64/pkgconfig/reflapack.pc

Yes, it exists, but no it is a symlink to the wrong pc file:

$ ls -ahl /usr/lib64/pkgconfig/lapack.pc
lrwxrwxrwx 1 root root 77 Oct 26 15:19 /usr/lib64/pkgconfig/lapack.pc -> ../../../etc/env.d/alternatives/lapack/_current/usr/lib64/pkgconfig/lapack.pc

Which package is to provide this symlink?  It seems that it is a stale
symlink from before -llapack was renamed to -lreflapack recently?

I'm experimenting with the different implementations, it is entirely
possible that I'm in some undefined state.  Now, how to fix it?

Thanks!
Thomas

> It is possible that some of these links have not been created properly.
> There used to be bug about it on prefix (I don't know if anyone improved
> my prefix bandaid). There is also the possibility that remnant from the
> non-sci overlay mechanism interfere.
> 
> Francois
> 
> 

-- 
Thomas Kahle
http://dev.gentoo.org/~tomka/

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 316 bytes --]

Re: [gentoo-science] blas/lapack stuff

[Francois Bissey]

On 28/10/12 06:42, Thomas Kahle wrote:
> On 21:50 Sat 27 Oct 2012, Francois Bissey wrote:
>> On 27/10/12 10:53, Thomas Kahle wrote:
>>> On 17:10 Wed 10 Oct 2012, justin wrote:
>>> [..]
>>>
>>>>> Calculating dependencies... done!
>>>>> [ebuild  N     ] app-admin/eselect-lapack-0.1  0 kB
>>>>> [blocks B      ] app-admin/eselect-lapack ("app-admin/eselect-lapack"
>>>>> is blocking sci-libs/gsl-1.15-r2, sci-libs/cblas-reference-20110218,
>>>>> sci-libs/blas-reference-20110418, sci-libs/atlas-3.10.0,
>>>>> sci-libs/lapack-reference-3.3.1-r1)
>>>>>
>>>>> Total: 1 package (1 new), Size of downloads: 0 kB
>>>>> Conflict: 1 block (1 unsatisfied)
>>>>>
>>>>>  * Error: The above package list contains packages which cannot be
>>>>>  * installed at the same time on the same system.
>>>>>
>>>>>   (sci-libs/cblas-reference-20110218::science, installed) pulled in by
>>>>>     sci-libs/cblas-reference required by @selected
>>>>>
>>>>>   (app-admin/eselect-lapack-0.1::gentoo, ebuild scheduled for merge)
>>>>> pulled in by
>>>>>     eselect-lapack
>>>>>
>>>>>   (sci-libs/atlas-3.10.0::science, installed) pulled in by
>>>>>     sci-libs/atlas required by @selected
>>>>>
>>>>>
>>>
>>>> do you have the eselect version from the sci overlay installed? Only
>>>> that one is able to eslect blas for you correctly.
>>>
>>> I'm using this eselect version and I have lapack reference-3.4.2
>>> installed and eselected.  Still I get:
>>>
>>> # pkg-config --libs lapack                                                        
>>> Package lapack was not found in the pkg-config search path.
>>> Perhaps you should add the directory containing `lapack.pc'
>>> to the PKG_CONFIG_PATH environment variable
>>> No package 'lapack' found
>>>
>>> Which makes packages depending on lapack checks during configure fail.
>>>
>>> Any ideas
>>>
>>
>> First of, supposing you are on amd64, does
>> /usr/lib64/pkgconfig/lapack.pc
>> exist? It should be a link pointing to
>> ../../../etc/env.d/alternatives/lapack/_current/usr/lib64/pkgconfig/lapack.pc
>> itself a link to /usr/lib64/pkgconfig/reflapack.pc
> 
> Yes, it exists, but no it is a symlink to the wrong pc file:
> 
> $ ls -ahl /usr/lib64/pkgconfig/lapack.pc
> lrwxrwxrwx 1 root root 77 Oct 26 15:19 /usr/lib64/pkgconfig/lapack.pc -> ../../../etc/env.d/alternatives/lapack/_current/usr/lib64/pkgconfig/lapack.pc
> 

So
../../../etc/env.d/alternatives/lapack/_current/usr/lib64/pkgconfig/lapack.pc
is not a symlink to /usr/lib64/pkgconfig/reflapack.pc?
That should be provided by lapack-reference using the alternative2
eclass. /usr/lib64/pkgconfig/lapack.pc is created by eselect.

> Which package is to provide this symlink?  It seems that it is a stale
> symlink from before -llapack was renamed to -lreflapack recently?
> 
> I'm experimenting with the different implementations, it is entirely
> possible that I'm in some undefined state.  Now, how to fix it?
> 

yes, you may have to delete some left over files and re-emerge
lapack-reference.

Francois

Re: [gentoo-science] blas/lapack stuff

[My Th]

O , 2011-10-11 11:42 +0700, Andrey G. Grozin rakstīja:
> Hello *,
> 
> I use the science overlay, and for quite some time I have problems with 
> blas/lapack. Each attempt to emerge -uD @world produces something like
> 
> ================================================================================
> [blocks B      ] app-admin/eselect-blas ("app-admin/eselect-blas" is 
> blocking sci-libs/lapack-reference-3.3.1-r1, sci-libs/gsl-1.15-r1, 
> sci-libs/blas-reference-20110418)
> [blocks B      ] app-admin/eselect-cblas ("app-admin/eselect-cblas" is 
> blocking sci-libs/lapack-reference-3.3.1-r1, sci-libs/gsl-1.15-r1, 
> sci-libs/blas-reference-20110418)
> [blocks B      ] app-admin/eselect-lapack ("app-admin/eselect-lapack" is 
> blocking sci-libs/lapack-reference-3.3.1-r1, sci-libs/gsl-1.15-r1, 
> sci-libs/blas-reference-20110418)
> 
> Total: 21 packages (21 upgrades), Size of downloads: 19,795 kB
> Conflict: 4 blocks (3 unsatisfied)
> Portage tree and overlays:
>   [0] /usr/portage
>   [1] /var/lib/layman/science
> 
>   * Error: The above package list contains packages which cannot be
>   * installed at the same time on the same system.
> 
>    (sci-libs/lapack-reference-3.3.1-r1::science, ebuild scheduled for 
> merge) pulled in by
>      >=sci-libs/lapack-reference-3.3 required by 
> (virtual/lapack-3.3::science, ebuild scheduled for merge)
> 
>    (app-admin/eselect-cblas-0.1::gentoo, installed) pulled in by
>      app-admin/eselect-cblas required by 
> (sci-libs/blas-atlas-3.9.23-r4::gentoo, installed)
> 
>    (sci-libs/blas-reference-20110418::science, ebuild scheduled for merge) 
> pulled in by
>      >=sci-libs/blas-reference-20110417 required by 
> (virtual/blas-2.1::science, ebuild scheduled for merge)
> 
>    (app-admin/eselect-blas-0.1::gentoo, installed) pulled in by
>      app-admin/eselect-blas required by 
> (sci-libs/blas-atlas-3.9.23-r4::gentoo, installed)
> 
>    (sci-libs/gsl-1.15-r1::science, ebuild scheduled for merge) pulled in by
>      sci-libs/gsl required by (sci-mathematics/giac-0.9.3::grozin, 
> installed)
>      sci-libs/gsl required by (sci-libs/mathgl-1.11.2::gentoo, installed)
>      sci-libs/gsl required by (sci-libs/tamu_anova-0.2.1::gentoo, 
> installed)
>      sci-libs/gsl required by (sci-visualization/qtiplot-0.9.8.8::gentoo, 
> installed)
>      sci-libs/gsl required by (media-gfx/asymptote-2.13::grozin, installed)
>      sci-libs/gsl required by (media-gfx/inkscape-0.48.2::gentoo, 
> installed)
>      sci-libs/gsl required by (app-office/kspread-2.3.3::gentoo, installed)
>      >=sci-libs/gsl-1.10 required by 
> (sci-visualization/pyxplot-0.8.4::gentoo, installed)
>      >=sci-libs/gsl-1.14-r2 required by (virtual/cblas-2.0::science, ebuild 
> scheduled for merge)
> 
>    (app-admin/eselect-lapack-0.1::gentoo, installed) pulled in by
>      app-admin/eselect-lapack required by 
> (sci-libs/lapack-atlas-3.9.23-r3::gentoo, installed)
> ================================================================================
> Any ideas?
> 
> Andrey
> 

Hi!

You have to resolve the blockers:
- Unmerge blas-atlas, lapack-atlas, eselect-{blas,cblas,lapack}
- Upgrade world
- Emerge sci-libs/atlas
- Eselect your blas, cblas, lapack implementations
- To be sure that all links properly, run revdep-rebuild

If eselect fails complaining that some file is not a symlink, run:
 emerge -1 atlas gsl lapack-reference blas-reference

Alter that eselecting should work without errors.

If you do not want atlas versions of blas, cblas and lapack, you can
skip the step where atlas is installed.


Reinis

Re: [gentoo-science] sci-mathematics/reduce on amd64 (fwd)

[Andrey G. Grozin]

On Wed, 21 Dec 2011, Kacper Kowalik wrote:
>> Could somebody with amd64 check if sci-mathematics/reduce-20110414
>> compiles on this arch? You'll have to keyword it. Upstream says it
>> should compile.
> 
> It does and passes test suite. Though it doesn't respect libdir:
> 
> /dev/shm/portage/sci-mathematics/reduce-20110414/image/usr/lib/reduce/reduce.fonts/README
> 
> Judging by the content it shouldn't be libdir in the first place, rather
> /usr/share...
Many thanks for checking.

csl-based reduce, and csl, are really strange. reduce is a compiled binary, and 
it needs to have reduce.img *in the same directory* (same for csl). Both of 
them also want to see reduce.doc and reduce.fonts also in the same directory, 
if they are started in the gui mode (in the command-line mode reduce.fonts and 
reduce.doc are not needed). So, all these files and directories should be 
together somewhere. And not in /usr/share, because reduce is a binary, and 
reduce.img is also processor-dependent, as far as I know. So, I choose 
/usr/lib/reduce; do you have better suggestions? Should it be 
/usr/lib64/reduce on amd64?

Moreover, symlinking /usr/lib/reduce -> /usr/bin/redcsl does not work: the 
program tries to find reduce.img in /usr/bin, and fails. The only way I've 
found is a 2-line script in /usr/bin which execs /usr/lib/reduce/reduce.

Thanks again,
Andrey

Re: [gentoo-science] sci-mathematics/reduce on amd64 (fwd)

[Kacper Kowalik]

On 21.12.2011 15:03, Andrey G. Grozin wrote:
> On Wed, 21 Dec 2011, Kacper Kowalik wrote:
>>> Could somebody with amd64 check if sci-mathematics/reduce-20110414
>>> compiles on this arch? You'll have to keyword it. Upstream says it
>>> should compile.
>>
>> It does and passes test suite. Though it doesn't respect libdir:
>>
>> /dev/shm/portage/sci-mathematics/reduce-20110414/image/usr/lib/reduce/reduce.fonts/README
>>
>>
>> Judging by the content it shouldn't be libdir in the first place, rather
>> /usr/share...
> Many thanks for checking.
>
> csl-based reduce, and csl, are really strange. reduce is a compiled
> binary, and it needs to have reduce.img *in the same directory* (same
> for csl). Both of them also want to see reduce.doc and reduce.fonts also
> in the same directory,

That's a snippet from a build log:

gcc -DHAVE_CONFIG_H -I. 
-I/var/tmp/portage/sci-mathematics/reduce-20110414/work/reduce-20110414/csl/cslbase 
  -Dfontsdir=reduce.fonts -I/usr/include/freetype2 
-DBINDIR=/usr/local/bin -DPKGDATADIR=/usr/local/share/reduce 
-march=core2 -mtune=generic -O2 -pipe -I/usr/include/freetype2 
-fno-strict-aliasing -O2 -Wall   -MT arith01.o -MD -MP -MF 
.deps/arith01.Tpo -c -o arith01.o 
/var/tmp/portage/sci-mathematics/reduce-20110414/work/reduce-20110414/csl/cslbase/arith01.c

I haven't looked at the code but I have strong feeling that precompiler 
directives:
  fontsdir
  BINDIR
  PKGDATADIR
may help you ;) It looks that removing 'local' is all you need. Then 
place everything which is not a binary and not a lib in usr/share/reduce

> Should it be /usr/lib64/reduce on amd64?
In principle you should always use /usr/$(get_libdir) from mutlilib.eclass.
Cheers,
Kacper

Re: [gentoo-science] sci-mathematics/reduce on amd64

[Andrey G. Grozin]

Many thanks to everybody,

I've just committed reduce-20110414-r1 with ~amd64. Could somebody check 
that it actually compiles on amd64?

Andrey

Re: [gentoo-science] sci-mathematics/reduce on amd64

[Honza Macháček]

Dne 12/22/2011 01:26 PM, Andrey G. Grozin píše:
> I've just committed reduce-20110414-r1 with ~amd64. Could somebody check 
> that it actually compiles on amd64?

Compiles and seems to work for me.

Portage 2.1.10.41 (default/linux/amd64/10.0, gcc-4.5.3, glibc-2.14.1-r1,
3.0.4-gentoo-r1 x86_64)
System uname:
Linux-3.0.4-gentoo-r1-x86_64-Intel-R-_Core-TM-2_Duo_CPU_E6750_@_2.66GHz-with-gentoo-2.1
Timestamp of tree: Thu, 22 Dec 2011 14:15:01 +0000

Thanks for your work.

Honza Macháček

Re: [gentoo-science] sci-mathematics/reduce on amd64

[Miguel de Val-Borro]

On Thu, Dec 22, 2011 at 07:26:21PM +0700, Andrey G. Grozin wrote:
> Many thanks to everybody,
> 
> I've just committed reduce-20110414-r1 with ~amd64. Could somebody check 
> that it actually compiles on amd64?

I have upgraded from reduce-20110414 to reduce-20110414-r1 and got  a file
collision during merging:

 * sci-mathematics/reduce-20110414
 *      /usr/lib64/reduce/reduce.doc
 *      /usr/lib64/reduce/reduce.fonts
 * 
 * Package 'sci-mathematics/reduce-20110414-r1' NOT merged due to file
 * collisions. If necessary, refer to your elog messages for the whole
 * content of the above message.

This is the build log for reduce-20110414-r1:
http://paste.pocoo.org/show/524580/

If reduce-20110414 is unmerged first, then the new revision builds fine.

Thanks,
Miguel

Re: [gentoo-science] sci-mathematics/reduce on amd64

[Andrey G. Grozin]

On Thu, 22 Dec 2011, Miguel de Val-Borro wrote:
> I have upgraded from reduce-20110414 to reduce-20110414-r1 and got  a file
> collision during merging:
>
> * sci-mathematics/reduce-20110414
> *      /usr/lib64/reduce/reduce.doc
> *      /usr/lib64/reduce/reduce.fonts
> *
> * Package 'sci-mathematics/reduce-20110414-r1' NOT merged due to file
> * collisions. If necessary, refer to your elog messages for the whole
> * content of the above message.
Yes, I saw this behaviour. I think this must be a bug in portage. The 
directories /usr/lib64/reduce/reduce.doc and 
/usr/lib64/reduce/reduce.fonts belong to the old version of reduce, which 
is being removed during this install. Why are they triggering file 
collision??

Andrey