#625: CarpetIOHDF5 sliced output does not output symmetry points unless it also outputs buffer points ---------------------+------------------------------------------------------ Reporter: hinder | Owner: eschnett Type: defect | Status: reviewed_ok Priority: major | Milestone: Component: Carpet | Version: Resolution: | Keywords: CarpetIOHDF5 ---------------------+------------------------------------------------------
Comment (by rhaas):
It produces readable output files in at least one case where the old one did not. I have not tested how many cases where the old code used to work it breaks. The dangerous point seems to me having a cctk_nghostzones attribute with 0 value.
There is no possibility for HDF5 output in test suites as far as I know. A possible solution could use the h5diff utility that comes with HDF5. There is already some discussion about using HDF5 in test suites in #808 (though it is mostly concerned with moving the test suites over to HDF5 to possibly save space).
If out3D_XXX are decprecated then the code should be ok right now, the place where I did not use out3d_ghosts was when deciding what value to write into the ghostzones attributes.
The code unfortunately duplicates code in the buffers and no-buffers branches of an if statement. This is not nice (and the reason for the TODO).
I had to fix a bug in the no-buffers routine and added a test case (though it cannot really be used yet, it passes with 0 of 0 files being identical).
There is likely still room for improvement so I am leaving this ticket open.
Applied as hashes a352f39386df and 1bc87fa7222a of CarpetIOHDF5.