[ebuild U ] sci-mathematics/octave-3.8.1-r1:0/3.8.1 [3.8.1:0/3.8.1] USE="X glpk imagemagick opengl qhull qrupdate readline sparse zlib -curl -doc -fftw -gnuplot -gui -hdf5 -java -jit -postscript -static-libs" Calculating dependencies... done! >>> Verifying ebuild manifests >>> Running pre-merge checks for sci-mathematics/octave-3.8.1-r1 * Checking linear algebra dependencies for libblas * /usr/lib64/libcholmod.so must be rebuilt with libblas * Try re-installing /usr/lib64/libcholmod.so * ERROR: sci-mathematics/octave-3.8.1-r1::gentoo failed (pretend phase): * check of blas/lapack consistency failed Reinstalling sci-libs/cholmod-2.1.2 does not help Installed versions: 2.1.2{tbz2}(23:27:00 04/16/14)(-cuda -doc -lapack -metis -minimal -static-libs) Reproducible: Always
$ eselect blas list Installed BLAS for library directory lib64 [1] reference *
[I] sci-libs/blas-reference Available versions: 20070226 (~)20070226-r2{tbz2} **20070226-r3 {doc} Installed versions: 20070226-r2{tbz2}(21:02:51 11/04/13)(-doc) Homepage: http://www.netlib.org/blas/ Description: Basic Linear Algebra Subprograms F77 reference implementations
Same problem here. query b /usr/lib/libblas.so * Searching for /usr/lib/libblas.so ... sci-libs/blas-reference-20070226 (/usr/lib64/blas/reference/libblas.so.0.0.0) equery b /usr/lib64/libcholmod.so * Searching for /usr/lib64/libcholmod.so ... sci-libs/cholmod-1.6.0-r1 (/usr/lib64/libcholmod.so -> libcholmod.so.0.0.0) sci-libs/cholmod-1.6.0-r1 (/usr/lib64/libcholmod.so.0.0.0) Rebuilding sci-libs/cholmod does not help.
As a note: enabling USE=blas on cholmod works around the issue. But the check is misguided somehow anyway... for lib in ${libs[@]}; do # linked with the current blas? if ! scanelf -n ${lib} | grep -q "${blaslib}"; then # linked with some blas or lapack library? if ! scanelf -n ${lib} | egrep -q "blas|lapack"; then failed_libs="${failed_libs} ${lib}" fi fi done Looks like the innermost conditional should not be negated like it is...
(In reply to Michał Górny from comment #4) > As a note: enabling USE=blas on cholmod works around the issue. My cholmod ebuilds do not seem to have a blas USE flag. [ebuild R ] sci-libs/cholmod-2.1.2 USE="-cuda -doc -lapack -metis -minimal -static-libs" 0 kB
(In reply to Kobboi from comment #5) > (In reply to Michał Górny from comment #4) > > As a note: enabling USE=blas on cholmod works around the issue. > > My cholmod ebuilds do not seem to have a blas USE flag. > > [ebuild R ] sci-libs/cholmod-2.1.2 USE="-cuda -doc -lapack -metis > -minimal -static-libs" 0 kB Sorry, I meant USE=lapack.
Is it naive to ask why this isn't handled with USE flag dependencies rather than a check for symbols in a library? It seems to me like the message indicates that cholmod must be built with the lapack flag and won't be acceptable any other way. Is there a reason that the check and message are obfuscated in the way they are? To verify Michał's diagnosis: I've tried re-building cholmod several times per the suggestion of this failure but it has no effect on the build. I do not have the lapack USE flag set on cholmod during these rebuilds. After rebuilding with USE=lapack, I also stop receiving the message from octave about blas/lapack checks as Michał found.
this check actually should not be part of an ebuild. this is more a missing feature of our eselect blas/lapack framework to keep the same soname. so i'm removing it. + 17 Apr 2014; Sébastien Fabbro <bicatali@gentoo.org> octave-3.8.1-r1.ebuild: + Removed checking of symbols in linear algebra libraries bug #507782 +
*** Bug 508066 has been marked as a duplicate of this bug. ***