Separate relocate_hdf

[samwaseda]

Separate the rewriting of the hdf address, in order for other jobs to be able to intervene (e.g. the new SPHInX implementation that requires an extra loading of lazy mode for the input)

[pmrv]

If I understand this correctly, the original code works because job.project_hdf5 points into the HDF file of the master job and this then resets it to point into its own file. Is this correct? If so, I would prefer to call this reset_hdf5 or something like this. Otherwise lgtm.

[samwaseda]

If I understand this correctly, the original code works because job.project_hdf5 points into the HDF file of the master job and this then resets it to point into its own file. Is this correct? If so, I would prefer to call this reset_hdf5 or something like this. Otherwise lgtm.

Well, I would rather say the original one works because the content of the HDF file is (supposed to be) stored in the job attributes (such as job.input) when the child job is loaded, and therefore the job doesn't care about the HDF file anymore. Of course this is not the case for the lazy loading.