I hadn't considered using internal versions of Blas and lapack. If this fixes the graphing races I am seeing using in-ports versions (hacked to build using gfortran), then I would actually favor this. These libs are unlikely to need regular patching.
What do other people think? On Nov 26, 2011 7:46 AM, <j...@bitminer.ca> wrote: > > >and also using gnuplot from octave failed completely > > I have a port that builds Octave 3.2.4 on OpenBSD 4.9, and runs gnuplot, > or at least sombrero works... > > Email me if you want a copy. > > It "features" > > MODGCC4_LANGS += c++ fortran > ... > CONFIGURE_ARGS= ${CONFIGURE_SHARED} \ > --with-fftw \ > --without-blas --without-lapack \ > --without-hdf5 > > that is, it builds with gfortan and uses the internal copies of blas and > lapack, not the packages. (The task of validating blas and lapack with > gfortran for all uses in ports is beyond my abilities. ) > > I also patched the regressions to avoid specific test cases with infinite > loops caused by bugs in precision routines as mentioned in the upstream > bug report. The bug doesn't seem to affect most of the test cases. > > The key advantage is that octave 3.2.x is vastly quicker than anything > earlier. 10x easily on some array stuff (FDTD calculations). > > > --John > > > >