CMake soversion differs from Autotools

[opoplawski]

The autotools build creates a library with a soversion of 0 (soname of libdf.so.0), but the cmake build makes one with a soversion of 4 (soname of libdf.so.4). Why? Shouldn't the libraries be equivalent?

[byrnHDF]

Unfortunately, hdf4 never had versioning.

[opoplawski]

Interesting, I think I just realized that the library name itself changed as well from libdf to libhdf. Is that right? That's quite the change.

[derobins]

The name change was an oversight when CMake was added

[derobins]

I looked into this and the problem is that -version-info isn't specified for the Autotools, so you just get 0.0.0 for current.revision.age. If I try to add this flag to libdf and libmfdf, they don't build their shared libraries and I get linkage errors downstream. This appears to be due to the unwise mingling of Fortran 77 and C code in the library source directories, which forces usage of the Fortran linker. We have some hacks to work around this, but it's possible they are conflicting with per-target Automake linker options.

It's possible that this is a solvable problem, even with the weird C/Fortran scheme, but I'm going to kick this to 4.4.0, for release next February, when we'll be reworking the Fortran interface to behave like HDF5's. If I rip out all the Fortran code, I can use -version-info normally, so this problem will hopefully go away in at that time, when I can drag over HDF5's Autotools soversion scheme.