Missing failure checks in HDF5

[Jan Blechta]

There is around one hundred checks of the type dolfin_assert(foo != HDF5_FAIL). That's not fair for users having asserts off not to be caught on an error condition.

I remind that asserts are exclusively for checking consistency in our library, i.e. a condition which should always be true. They should not be used for checking correctness of user input or error conditions which can happen.

[Chris Richardson]

We need a shorter one-line dolfin_error to avoid cluttering the code with error checks. Agree otherwise.

[Prof Garth Wells]

Agree on dolfin_error having too many arguments. It's tedious to use.

[Martin Sandve Alnæs]

Use a macro like this https://bitbucket.org/fenics-project/dolfin/src/0aa8af18ff46ddcaec22db1cc927c4f5fc88813c/dolfin/la/PETScVector.cpp?at=master&fileviewer=file-view-default#PETScVector.cpp-44 and you can get file and line automatically without having to type it out.

[Chris Richardson]

Macros are generally (and rightly) considered evil... but I could be persuaded.

[Prof Garth Wells]

It's often not important where an error is raised, e.g. if trying to open a file that doesn't exist the most helpful message is a short one that says the file does not exist.

[Chris Richardson]

@michalhabera will look into using HDF5's own error handler to produce error reports.