Created attachment 170958 [details] patch See http://lists.freebsd.org/pipermail/freebsd-ports/2016-May/103358.html The recent update from hdf5 1.8.16 to 1.10.0 has introduced an issue with ABI incompatibility. Thus, anyone that has code written to the 1.8.16 ABI may find that their code no longer functions with 1.10.0. This should have been noted in ports/UPDATING. with patch from Steve Kargl <sgk@troutmask.apl.washington.edu>, see attachment
Could you please explain how it has been fixed? Thanks
The two ports reported as broken are fixed, their tickets closed, so I assume the patch is no longer necessary. If it's still necessary, please reopen.
The 2 broken dependent ports have been fixed locally, but the general case (particularly for FORTRAN consumers) is not yet fixed. See the thread beginning at https://docs.freebsd.org/cgi/getmsg.cgi?fetch=65714+0+current/freebsd-ports
Assign to maintainer.
(In reply to Thierry Thomas from comment #3) Fixes were committed for matio and scilab. They now explicitly specify that they want to use the version 1.8 API. Portmon reports that they are now successfully building. There are a number of other consumers of hdf5. I don't know their status. Consumers written in C can specify the API version that they want with CPPFLAGS. Fortran consumers are more of a concern. Changing the default API version helps in the short term, but what happens when ports start wanting to use conflicting API versions of the fortran modules?
Unfortunately, this is often a problem with HDF: I remember that some years ago, Code_Aster's upstream had to deal with H5_USE_16_API with hdf5-18. We already have science/hdf and science/hdf5, and at some point we might need 2 un-conflicting ports for hdf5... Meanwhile, it would be better to respect the 1.8 API.
science/hdf5-18 has been resurrected.