On Wed, Sep 02, 2015 at 09:55:39PM -0400, Roland Haas wrote:
Would it already be sufficient to have a thorn IOHDF5Timeseries which provides another output method (along with IOASCII, IOHDF5 and IOScalar) that writes variables as time-series ie. extensible HDF5 datasets with the first dimension being time? If given an option like one_file_per_group it would combine all scalars in a group into different columns of the output dataset. It would not be able to write grid functions (or slices of them) since their extent can change during the simulation but it would be able to write SCALARs and ARRAYS as well as 0d data. The thorn would write one HDF5 dataset per Cactus variable. This would mean that the new output method behaves very much like the current ones and is more or less the equivalent CarpetIOASCII's compact output format which works decently well for ASCII timeseries output.
It would solve the "HDF5" issue, but not the one Ian mentioned: that sometimes data can be invalid. We don't have a good mechanism for thorns to mark data as currently invalid (e.g. AHFinder quantities).
Frank