> I see the one very dangerous change in the libhdf5 binary interface between
> 1.8.5 and 1.8.6 versions [2], but the soname was not bumped: the value of
> H5T_ORDER_NONE member has been changed from 3 to 4 in the H5T_order_t
> enumeration type (i.e. old apps execute H5T_ORDER_MIXED instead of NONE).

> Applications shouldn't use numbers...

Even when the application is coded using "H5T_ORDER_NONE" the compiler
ultimately "uses numbers" in the binary.  The situation described is
problematic when an application compiled against 1.8.5 headers uses
1.8.6 at runtime.  This will happen when the shared library is
upgraded but the application is not recompiled.

Granted, HD5's big scary banner about using the wrong shared library
minor version provides some measure of safety.  Still, pesky folks
like me who set HDF5_DISABLE_VERSION_CHECK are at risk.

- Rhys

_______________________________________________
Hdf-forum is for HDF software users discussion.
[email protected]
http://mail.hdfgroup.org/mailman/listinfo/hdf-forum_hdfgroup.org

Reply via email to