numpy and lapack on linux

I need to distribute matplotlib as part of a large and somewhat heterogeneous package of Python-based software. On Macs, lapack is installed by default, and I can rely on the linking working regardless of which OS version it was compiled on (we use 10.4.11, but it runs fine on 10.5). On Linux, it’s variable - the kernel 2.6-based OSes we use appear to have it, but not the older 2.4-based systems. For the older systems this is not a problem - it results in this:

nat@…2592…> ldd lapack_lite.so

linux-gate.so.1 => (0xffffe000)

libc.so.6 => /lib/libc.so.6 (0xf7d97000)

/lib/ld-linux.so.2 (0x56555000)

The lack of links to the full lapack module doesn’t appear to break anything that I do - I’m just plotting simple lines and I don’t use any of the numpy routines in my code.

This is what happens on the newer systems:

nat@…2593…> ldd lapack_lite.so

linux-gate.so.1 => (0xffffe000)

liblapack.so.3 => not found

libblas.so.3 => not found

libg2c.so.0 => /usr/lib/libg2c.so.0 (0xf7fb6000)

libm.so.6 => /lib/libm.so.6 (0xf7f8f000)

libgcc_s.so.1 => /lib/libgcc_s.so.1 (0xf7f82000)

libc.so.6 => /lib/libc.so.6 (0xf7e45000)

/lib/ld-linux.so.2 (0x56555000)

This results in an import error on systems that don’t have lapack3 installed. We would prefer to avoid having to tell our users to install more software in order to use what I’ve written. Is there any way to force numpy to not link to anything besides what it saw in the first example? I would prefer to avoid hacking into the numpy distutils scripts if possible.

Sorry this isn’t directly matplotlib-related, but I thought I’d see if someone here had encountered this problem before I subscribed to another list.

thanks,

Nat

Hi,

Just compile numpy on a system that doesn't have lapack3 installed.

Matthieu

2009/5/4 Nathaniel Echols <nathaniel.echols@...287...>:

···

I need to distribute matplotlib as part of a large and somewhat
heterogeneous package of Python-based software. On Macs, lapack is
installed by default, and I can rely on the linking working regardless of
which OS version it was compiled on (we use 10.4.11, but it runs fine on
10.5). On Linux, it's variable - the kernel 2.6-based OSes we use appear to
have it, but not the older 2.4-based systems. For the older systems this is
not a problem - it results in this:

nat@...2592...> ldd lapack_lite.so
linux-gate.so.1 => (0xffffe000)
libc.so.6 => /lib/libc.so.6 (0xf7d97000)
/lib/ld-linux.so.2 (0x56555000)
The lack of links to the full lapack module doesn't appear to break anything
that I do - I'm just plotting simple lines and I don't use any of the numpy
routines in my code.
This is what happens on the newer systems:
nat@...2593...> ldd lapack_lite.so
linux-gate.so.1 => (0xffffe000)
liblapack.so.3 => not found
libblas.so.3 => not found
libg2c.so.0 => /usr/lib/libg2c.so.0 (0xf7fb6000)
libm.so.6 => /lib/libm.so.6 (0xf7f8f000)
libgcc_s.so.1 => /lib/libgcc_s.so.1 (0xf7f82000)
libc.so.6 => /lib/libc.so.6 (0xf7e45000)
/lib/ld-linux.so.2 (0x56555000)
This results in an import error on systems that *don't* have lapack3
installed. We would prefer to avoid having to tell our users to install
more software in order to use what I've written. Is there any way to force
numpy to *not* link to anything besides what it saw in the first example? I
would prefer to avoid hacking into the numpy distutils scripts if possible.
Sorry this isn't directly matplotlib-related, but I thought I'd see if
someone here had encountered this problem before I subscribed to another
list.
thanks,
Nat
------------------------------------------------------------------------------
Register Now & Save for Velocity, the Web Performance & Operations
Conference from O'Reilly Media. Velocity features a full day of
expert-led, hands-on workshops and two days of sessions from industry
leaders in dedicated Performance & Operations tracks. Use code vel09scf
and Save an extra 15% before 5/3. http://p.sf.net/sfu/velocityconf
_______________________________________________
Matplotlib-users mailing list
Matplotlib-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/matplotlib-users

--
Information System Engineer, Ph.D.
Website: http://matthieu-brucher.developpez.com/
Blogs: http://matt.eifelle.com and http://blog.developpez.com/?blog=92
LinkedIn: http://www.linkedin.com/in/matthieubrucher

Unfortunately, this is going to be difficult - our build process is very automated (done every night) and we use these systems for other things too, so I’m very reluctant to tinker with the installed software, and I doubt we have spare boxes with suitable hardware. I will talk to our admin but my guess is he’ll say the same thing.

thanks,

Nat

···

On Mon, May 4, 2009 at 10:50 AM, Matthieu Brucher <matthieu.brucher@…287…> wrote:

Just compile numpy on a system that doesn’t have lapack3 installed.