python-numpy prone to use host libraries
I have been resolving issues regarding python-numpy this whole day and I would like to share my findings. python-numpy is a package required by opencv_2.2, which is a dependency of gst-plugins-bad making it essential to quite a lot of images.
First of all, python-numpy requires a fortran compiler and gcc-4.5 specifically disables the fortran language per default. From gcc-4.5.inc
# Language Overrides
FORTRAN = ""
I am unsure what the best approach is, but I resolved this by changing to FORTRAN = ",fortran"
Looking at the compile log of python-numpy there were several instances of the following:
blas_info:
libraries blas not found in/home/bitbake/build/tmp_beagleboard/sysroots/x86_64-linux/usr/lib
libraries blas not found in /usr/local/lib64 libraries blas not found in /usr/local/lib
FOUND:
libraries = ['blas']
library_dirs = ['/usr/lib64']
language = f77
FOUND:
libraries = ['blas']
library_dirs = ['/usr/lib64']
define_macros = [('NO_ATLAS_INFO', 1)]
language = f77
It clearly tries to link against my host libraries and fails. My resort was simply to remove my host libraries (not needed at the moment). However, there are other ways to counter this problem.
Looking at the python-numpy sources, there is a file called site.cfg.example. In it, there is descriptions on how to supply python-numpy with configuration information about non-Python dependencies. It shouldn't be any problem generating site.cfg files for relevant parts of the python-numpy source. I am however uncertain on how to proceed with this and could use some assistance.
Signature · 13 lines
MvH / with best regards Patrik Dahlström Ordförande/president - http://www.studentfiket.com/ Penga-Mas (Kassör)/cashier - Dalarnas Nation "Alla djur, utom människan, vet att livets huvudsakliga mening är att njuta av det! --Samuel Butler "Every animal, except man, knows that the fundamental meaning of life is to enjoy it" --Samuel Butler () ascii ribbon campaign - against html e-mail /\ www.asciiribbon.org - against proprietary attachments
Dear Patrik,
Am Sonntag, den 10.04.2011, 21:45 +0200 schrieb Patrik Dahlström (student):
[…]
Quoted 2 lines
I am however uncertain on how to proceed with this and could use some assistance.
did you find a “generic” solution for this problem?
Thanks,
Paul
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Quoted 11 lines · Paul Menzel
Dear Patrik,
Am Sonntag, den 10.04.2011, 21:45 +0200 schrieb Patrik Dahlström (student):
[…]
Quoted 2 lines
I am however uncertain on how to proceed with this and could use some assistance.
did you find a “generic” solution for this problem?
It requires redoing the numpy buildsystem, it is broken by design. I tried to teach it about crosscompiling and sysroots last year, but gave up after a few days :(
regards,
Koen
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (Darwin)
iD8DBQFOCXo2MkyGM64RGpERAggYAJ0VMl1Ttg0QGqNCrmBhdlhnp9yPzgCgldQf d5EbEv+2sn1ssE10DGkigsM=
=Sx4L
-----END PGP SIGNATURE-----
Does that mean that numpy is a completely broken package, or is that a way to hack it to make it work? If it is broken and will stay that way, why don't we remove it from the recipe list? If that is the case then it's just misleading.
Thanks,
Bob
Quoted 38 lines · Koen Kooi
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1Quoted 4 lines · Paul Menzel
Dear Patrik,
Am Sonntag, den 10.04.2011, 21:45 +0200 schrieb Patrik Dahlström
(student):
Quoted 7 lines
[…]
Quoted 2 lines
I am however uncertain on how to proceed with this and could use some assistance.
did you find a “generic” solution for this problem?
It requires redoing the numpy buildsystem, it is broken by design. I tried to teach it about crosscompiling and sysroots last year, but gave up after a few days :(
regards,
Koen
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (Darwin)iD8DBQFOCXo2MkyGM64RGpERAggYAJ0VMl1Ttg0QGqNCrmBhdlhnp9yPzgCgldQf d5EbEv+2sn1ssE10DGkigsM=
=Sx4L
-----END PGP SIGNATURE-----_______________________________________________
Openembedded-devel mailing list
[email protected]
http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel
Signature · 5 lines
University of Illinois at Urbana-Champaign Electrical Engineering [email protected] (630) 306-4018
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Quoted 4 lines · Bob Cvengros
Does that mean that numpy is a completely broken package, or is that a way to hack it to make it work? If it is broken and will stay that way, why don't we remove it from the recipe list? If that is the case then it's just misleading.
It works if you don't have (from the top of my head) g99 and blas-dev installed on your host. I removed those and numpy builds for me.
regards,
Koen
Quoted 20 lines
Thanks,
BobOn Tue, Jun 28, 2011 at 1:52 AM, Koen Kooi <[email protected]>wrote:
Quoted 4 lines · Paul Menzel
Quoted 4 lines
Quoted 4 lines
Dear Patrik,
Am Sonntag, den 10.04.2011, 21:45 +0200 schrieb Patrik Dahlström
(student):
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
[&]
Quoted 2 lines
I am however uncertain on how to proceed with this and could use some assistance.
did you find a generic solution for this problem?
Quoted 9 lines
It requires redoing the numpy buildsystem, it is broken by design. I tried to teach it about crosscompiling and sysroots last year, but gave up after a few days :(
regards,
Koen
Quoted 2 lines
_______________________________________________
Openembedded-devel mailing list
[email protected]
http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel
Quoted 1 line
Quoted 1 line
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (Darwin)
iD8DBQFOCe6MMkyGM64RGpERAtm7AJ4otnJkPoVsupxC4c95EdfKL9s/CwCfYb5A +9ckzBIhVxGFCJ8Ict3fzug=
=Q31X
-----END PGP SIGNATURE-----