Honza Macháček schrieb:
ccp4 depends on fftw-2. And the build has an error because of not finding the rfftw.h. It's not fatal, but there mide be some functions broken. It should be linked with and without "float". It looks like the ebuild doesn' t do it.Justin wrote:Does anyone has rfftw.h? It should be in /usr/include/ ...Does any package need it? I don't have that file and nothing ever complained, as far as I remember. After all, drfftw.h or srfftw.h can allways be included specifically.That said, the symlik belongs -- or rather should belong -- to fftw2, fftw3 does not structure its headers that way; I have installed both sci-libs/fftw-2.1.5-r3 and sci-libs/fftw-3.1.2, bud qdepends -Q says nothing depends on the 2 series version (that may be why nothing complained), and only sci-mathematics/octave-3.0.0 on the 3 series one.From fftw-2.1.5-r3.ebuild:if use float; then for f in "${D}"/usr/{include,$(get_libdir)}/*sfft*; do ln -s $(basename ${f}) ${f/sfft/fft} done else for f in "${D}"/usr/{include,$(get_libdir)}/*dfft*; do ln -s $(basename ${f}) ${f/dfft/fft} done fiThat code does not create the rfftw.h symlink, but looks like it actually was intended to do so and the omission is an ebuild error. Definitely if float is in USE, srfftw.h should be linked, otherwise drfftw.h. And of course it is you in the end who should decide if you want single or double precision variant to be used by default.With best regards Honza Macháček
signature.asc
Description: OpenPGP digital signature
