[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Toon-members] pkgconfig support
From: |
Pekka Paalanen |
Subject: |
Re: [Toon-members] pkgconfig support |
Date: |
Wed, 23 Jun 2010 15:26:56 +0300 |
On Wed, 23 Jun 2010 11:04:20 +0100 (BST)
"E. Rosten" <address@hidden> wrote:
> I haven't had time to review the patch, but I'm happy about patching
> in pkg-config support when I get some time.
>
> By the way, TooN already tests for LAPACK in the configure script.
> The test is well debugged and has worked on very many platforms.
Yes, it tests for -llapack by linking a trivial program, but does this
test fail, if -llapack is not enough? Or is every LAPACK implementation
you have encountered self-contained in the sense, that the application
only ever needs to link to -llapack and nothing else?
Even TooN README instructs to add -lblas and for some cases also -lg2c.
I seem to recall that in Ubuntu 8.04 you need -llapack -lgfortran2
and/or fix a gfortran symlink by hand to get it to work. The failure
would occur at runtime when actually calling certain LAPACK functions.
Eh, I just noticed something. Both libcxcore.so.2.0 from OpenCV and
libblas.so from blas-atlas contain the symbol dscal_. If I link LAPACK
with only -llapack, TooN::SVD ends up calling into libcxcore and
segfaults. If I use ´pkg-config --libs lapack´ instead, my program
runs fine. Ugh.
OTOH, the TooN test suite links only with -llapack, and works just fine
on my Gentoo system with lapack-atlas.
I don't think the existing LAPACK test provides a complete list of
linker flags required to link to a LAPACK implementation. This whole
thing is a bit sad, since we are trying to automatically detect
something, that the LAPACK implementor or at least the packager should
have told us in the first place (via a pkg-config file, preferrably).
Thanks.
--
Pekka Paalanen, M.Sc. (Eng.) LUT/IT, room 6505
doctoral student http://www.it.lut.fi/project/rtmosaic/
address@hidden http://www.iki.fi/pq/