hdf5 seems to require quad precision to compile, which was only added to gfortran via libquadmath in 4.6 (at least for common architectures; I wouldn't know if there was some machine that had quad support earlier). For me, emerging hdf5 (the first and only stable version right now) stopped on a line with "KIND=16", with gfortran noting that the kind was unsupported. Hopefully gcc 4.6 is stabilized soon. Even so, I think that the dependency should be added.
which versions of hdf5? I compile hdf-1.8.9 with gcc-4.3 on SLES11 box at work so I am bit surprised by this claim. Can you attach a build log of a failed install with gcc<4.6?
This is sci-libs/hdf5-1.8.8-r1. To be clear, compiling with the fortran USE flag is the problem, because this is an issue specific to gfortran. I would not be surprised if everything else works fine with gcc-4.3. Here's the error message: mpif90 -O2 -pipe -march=native -I../../src -I../../fortran/src -O2 -pipe -march=native -c -o H5fortran_detect.o H5fortran_detect.f90 H5fortran_detect.f90:175.16: REAL(KIND= 16) :: a 1 Error: Kind 16 not supported for type REAL at (1) H5fortran_detect.f90:178.21: a_size = SIZEOF(a) 1 Error: Symbol 'a' at (1) has no IMPLICIT type make[2]: *** [H5fortran_detect.o] Error 1 There are a prodigous number of warnings, so the log is too large to attach, but I think this is the only relevant part as I can reproduce the error in a short program: program test integer, parameter :: qk = 16 real(qk) :: a = 1._qk write(*,*) a end program test With gfortran 4.5.4 this yields: test.F90:5.9: real(qk) :: a = 1._qk 1 Error: Kind 16 not supported for type REAL at (1)
It's a dupe of bug 419895. Please use hdf5-1.8.9. I'll reuse this bug for stabilization.
Couple of bug fixes, notably the one described originally in this report. Arches please do your magic!
stable on ppc ppc64
amd64 stable
x86 stable
@alpha, @ia64, @sparc Guys you may consider skipping this version and go ahead straight to 1.8.10. I'll request stablereq for it later this week.
(In reply to comment #8) > @alpha, @ia64, @sparc > Guys you may consider skipping this version and go ahead straight to 1.8.10. > I'll request stablereq for it later this week. Don't forget to file the stablereq for the new autoconf before asking 1.8.10 stable
sparc stable
alpha stable. Last arch, closing