headers
Robert Kern | 2 Dec 02:20

Re: Installing ETS-3.0.3 in Sage

On Mon, Dec 1, 2008 at 18:51, Jaap Spies <j.spies@...> wrote:
> Robert Kern wrote:
>> On Mon, Dec 1, 2008 at 18:28, Jaap Spies <j.spies@...> wrote:
>>> Robert Kern wrote:
> /lib', 'search_static_first': '0'}
>>
>> Okay, that's just odd. Did the Sage packager of numpy modify
>> numpy/distutils/system_info.py at all?
>
> I asked around, but with no reslut yet.

I've looked at the numpy-1.2.0.p0.spkg, and there doesn't seem to be
anything suspicious. Is that the version of numpy you have installed?

>> Anyways, the workaround is to make a file ~/.numpy-site.cfg with the contents
>>
>> [x11]
>> library_dirs=/usr/lib
>> include_dirs=/usr/include
>>
>
> Don't know. How will this help installing ets with my install-pkg?

This file will tell numpy.distutils (and consequently Enable's setup
scripts) to look in those locations for the X11 libraries and headers
and ignore the defaults. The defaults appear to be broken in your
installation; I don't know why.

> Not only for me, but also for the person who downloads this package?
(Continue reading)

Permalink | Reply |
headers
Jaap Spies | 2 Dec 02:43
Favicon

Re: Installing ETS-3.0.3 in Sage

Robert Kern wrote:
> On Mon, Dec 1, 2008 at 18:51, Jaap Spies <j.spies@...> wrote:
>> Robert Kern wrote:
>>> On Mon, Dec 1, 2008 at 18:28, Jaap Spies <j.spies@...> wrote:
>>>> Robert Kern wrote:
>> /lib', 'search_static_first': '0'}
>>> Okay, that's just odd. Did the Sage packager of numpy modify
>>> numpy/distutils/system_info.py at all?
>> I asked around, but with no reslut yet.
> 
> I've looked at the numpy-1.2.0.p0.spkg, and there doesn't seem to be
> anything suspicious. Is that the version of numpy you have installed?
> 

Yes.

>>> Anyways, the workaround is to make a file ~/.numpy-site.cfg with the contents
>>>
>>> [x11]
>>> library_dirs=/usr/lib
>>> include_dirs=/usr/include
>>>
>> Don't know. How will this help installing ets with my install-pkg?
> 
> This file will tell numpy.distutils (and consequently Enable's setup
> scripts) to look in those locations for the X11 libraries and headers
> and ignore the defaults. The defaults appear to be broken in your
> installation; I don't know why.
> 
>> Not only for me, but also for the person who downloads this package?
(Continue reading)

Permalink | Reply |
headers
Robert Kern | 2 Dec 03:43

Re: Installing ETS-3.0.3 in Sage

On Mon, Dec 1, 2008 at 19:43, Jaap Spies <j.spies@...> wrote:
> Robert Kern wrote:

>> Let's check a couple of other things. At the shell:
>>
>
> Ok, just before I go to sleep (2.30 AM local time)
>
>> $ locate "*site.cfg"
>
> [jaap <at> paix sage-3.2.1.rc0]$ locate "*site.cfg"

> /home/jaap/work/downloads/sage-3.2.1.rc0/local/lib/python2.5/site-packages/numpy/distutils/site.cfg

Okay, I see the problem. The spkg-install creates this file with a
[DEFAULT] section. This overrides all of the defaults in
system_info.py. Unfortunately, there isn't a terribly convenient way
out of this. numpy's spkg-install could make individual sections for
those it is going to use to build scipy:

[lapack_opt]
library_dirs = ...
include_dirs = ...

[blas_opt]
library_dirs = ...
include_dirs = ...

[fftw]
library_dirs = ...
(Continue reading)

Permalink | Reply |
headers
Jaap Spies | 2 Dec 15:42
Favicon

Re: Installing ETS-3.0.3 in Sage

Robert Kern wrote:
> On Mon, Dec 1, 2008 at 19:43, Jaap Spies <j.spies@...> wrote:
>> Robert Kern wrote:
> 
>>> Let's check a couple of other things. At the shell:
>>>
>> Ok, just before I go to sleep (2.30 AM local time)
>>
>>> $ locate "*site.cfg"
>> [jaap <at> paix sage-3.2.1.rc0]$ locate "*site.cfg"
> 
>> /home/jaap/work/downloads/sage-3.2.1.rc0/local/lib/python2.5/site-packages/numpy/distutils/site.cfg
> 
> Okay, I see the problem. The spkg-install creates this file with a
> [DEFAULT] section. This overrides all of the defaults in
> system_info.py. Unfortunately, there isn't a terribly convenient way
> out of this. numpy's spkg-install could make individual sections for
> those it is going to use to build scipy:
> 
> [lapack_opt]
> library_dirs = ...
> include_dirs = ...
> 
> [blas_opt]
> library_dirs = ...
> include_dirs = ...
> 
> [fftw]
> library_dirs = ...
> include_dirs = ...
(Continue reading)

Permalink | Reply |
headers
Jaap Spies | 2 Dec 16:53
Favicon

Re: Installing ETS-3.0.3 in Sage

Jaap Spies wrote:

> 
> But in the mean time is there a possibility to get chaco running?
> 

To answer myself: I've got a working chaco by puting the [DEFAULT]
in numpy/distutils/site.cfg aside.

Robert, thanks again!

Jaap
Permalink | Reply |
headers
Jaap Spies | 2 Dec 22:31
Favicon

Re: Installing ETS-3.0.3 in Sage

Robert Kern wrote:
> On Mon, Dec 1, 2008 at 19:43, Jaap Spies <j.spies@...> wrote:
>> Robert Kern wrote:
> 
>>> Let's check a couple of other things. At the shell:
>>>
>> Ok, just before I go to sleep (2.30 AM local time)
>>
>>> $ locate "*site.cfg"
>> [jaap <at> paix sage-3.2.1.rc0]$ locate "*site.cfg"
> 
>> /home/jaap/work/downloads/sage-3.2.1.rc0/local/lib/python2.5/site-packages/numpy/distutils/site.cfg
> 
> Okay, I see the problem. The spkg-install creates this file with a
> [DEFAULT] section. This overrides all of the defaults in
> system_info.py.

I think this is a real design issue. Why use the crappy numpy distutils to get system info.
I think there are better (more portable) ways to get this info.

Jaap
Permalink | Reply |
headers
Gael Varoquaux | 2 Dec 22:33
Favicon

Re: Installing ETS-3.0.3 in Sage

On Tue, Dec 02, 2008 at 10:31:00PM +0100, Jaap Spies wrote:
> I think this is a real design issue. Why use the crappy numpy distutils to get system info.
> I think there are better (more portable) ways to get this info.

Which one? I'd really like to know, because this is a problem I have seen
everywhere, with all kind of configuration systems. Quite often people
try to fix the problem by themselves by rewritten a configuration system,
and thus introducing different problems.

Gaël
Permalink | Reply |
headers
Jaap Spies | 2 Dec 23:49
Favicon

Re: Installing ETS-3.0.3 in Sage

Gael Varoquaux wrote:
> On Tue, Dec 02, 2008 at 10:31:00PM +0100, Jaap Spies wrote:
>> I think this is a real design issue. Why use the crappy numpy distutils to get system info.
>> I think there are better (more portable) ways to get this info.
> 
> Which one? I'd really like to know, because this is a problem I have seen
> everywhere, with all kind of configuration systems. Quite often people
> try to fix the problem by themselves by rewritten a configuration system,
> and thus introducing different problems.

For now I think I made my point by referring to the failing numpy install.

Alternatives may have there own problems, I agree. There are true believers
in many systems. I don't believe in the approach of Chaco -> Enable -> kiva
using an extern software numpy to get critical system information.

Jaap

> 
> Gaël
> 
> _______________________________________________
> Enthought-dev mailing list
> Enthought-dev@...
> https://mail.enthought.com/mailman/listinfo/enthought-dev
> 
> 
>
Permalink | Reply |
headers
Robert Kern | 3 Dec 00:08

Re: Installing ETS-3.0.3 in Sage

On Tue, Dec 2, 2008 at 16:49, Jaap Spies <j.spies@...> wrote:
> Gael Varoquaux wrote:
>> On Tue, Dec 02, 2008 at 10:31:00PM +0100, Jaap Spies wrote:
>>> I think this is a real design issue. Why use the crappy numpy distutils to get system info.
>>> I think there are better (more portable) ways to get this info.
>>
>> Which one? I'd really like to know, because this is a problem I have seen
>> everywhere, with all kind of configuration systems. Quite often people
>> try to fix the problem by themselves by rewritten a configuration system,
>> and thus introducing different problems.
>
> For now I think I made my point by referring to the failing numpy install.

No, you haven't. If you present an alternative that works better, I am
happy to concede the point. Build configuration across many disparate
platforms is simply a hard problem. Demonstrating that one approach
has issues does not mean that there are other approaches which are
free of such issues.

> Alternatives may have there own problems, I agree. There are true believers
> in many systems. I don't believe in the approach of Chaco -> Enable -> kiva
> using an extern software numpy to get critical system information.

<shrug> We're using numpy.distutils to build Enable for several other
reasons (not least that the extension modules use numpy). That
configuration mechanism happened to be there, so we are using it.
Ultimately, it gives the person installing complete control via
~/.numpy-site.cfg, but it does sacrifice some automatic assumptions
that might be made.
(Continue reading)

Permalink | Reply |
headers
Jaap Spies | 3 Dec 14:36
Favicon

Re: Installing ETS-3.0.3 in Sage

Robert Kern wrote:
> On Tue, Dec 2, 2008 at 16:49, Jaap Spies <j.spies@...> wrote:
[...]
>> For now I think I made my point by referring to the failing numpy install.
> 
> No, you haven't. If you present an alternative that works better, I am
> happy to concede the point. Build configuration across many disparate
> platforms is simply a hard problem. Demonstrating that one approach
> has issues does not mean that there are other approaches which are
> free of such issues.
> 

Ok, you have your point. I was speaking too early (before I studied
the sources!).

[...]
> <shrug> We're using numpy.distutils to build Enable for several other
> reasons (not least that the extension modules use numpy). That
> configuration mechanism happened to be there, so we are using it.
> Ultimately, it gives the person installing complete control via
> ~/.numpy-site.cfg, but it does sacrifice some automatic assumptions
> that might be made.
> 

On second thoughts I think the maintainers of the numpy.spkg are to
blame here by the introduction the [DEFAULT] in the site.cfg.

Thanks for making this clear.

Jaap
(Continue reading)

Permalink | Reply |

Gmane