XDMFFile does not close correctly at end of python session about dolfinx HOT 7 CLOSED

As @garth-wells says, context manager and python's GC is recommended approach. However is not possible for the old XDMFFile::write unless you want to have your time loop wrapped and tabbed in context manager. That being said, once you destroy XDMFFile object the old write() method won't append it. On the other hand new write_checkpoint appends it and "local" context managers are possible for time loops.

There is also a parameter flush_output (https://github.com/FEniCS/dolfinx/blob/master/cpp/dolfin/io/XDMFFile.cpp#L58) which forces underlying HDF file to be closed after every XDMFFile::write method call. That should solve the issue you are reporting.
But I do not like these parameters, especially when they are set to an object as whole (e.g. XDMFFile) but they apply only to part of that object (like XDMFFile::write on functions). Very confusing.

If you'd like to do very quick fix of this I'd recommend putting flush_output = True by default.

from dolfinx.

Are there any objections to having XDMF::close back in the python interface? I agree with using with wherever possible, but the nested indentation from with and loops for complicated FE problems could quickly get out of hand...

from dolfinx.

with XDMFFile("out2.xdmf") as f:
   f.write(uu, float(100.0), XDMFFile.Encoding.HDF5)

is the recommended approach. Garbage is collection is complicated, especially when calling 3rd party libs, so we should change the i/o syntax in demos.

from dolfinx.

@michalhabera @nate-sime I would suggest using with where possible, and in other cases use close(). We should update demos to follow this pattern.

from dolfinx.

You can always use del xdmf2. But probably close is neater

from dolfinx.

@nate-sime is this now resolved?

from dolfinx.

Issues addressed and resolved in PR #77

from dolfinx.