#1994: PITTNUllCode/SphericalHarmonicRecon uses HDF5 1.8 functions ------------------------------------+--------------------------------------- Reporter: rhaas | Owner: Type: defect | Status: new Priority: minor | Milestone: Component: EinsteinToolkit thorn | Version: development version Keywords: SphericalHarmonicRecon | ------------------------------------+--------------------------------------- SphericalHarmonicRecon's Recon.cc file uses H5LTread_dataset_double (https://support.hdfgroup.org/HDF5/doc/HL/RM_H5LT.html#H5LTread_dataset_doubl...) which is only available in the hdf5 hl level library which is only provided in hdf5 1.8.7.
Nothing in the ET states that HDF5 1.8 is required.
This can be fixed my making sure that hdf5_hl is included in HDF5_LIBS which also will make compilation fail if the library in indeed missing.
We should decide whether we want to bump ExternalLibraries/HDF5's version number and have it itself check for HDF5 1.8 in which case thorns can rely on this version being present (if they require a new enough HDF5 ExternalLibraries) or provide workarounds in Reconn.cc.
#1994: PITTNUllCode/SphericalHarmonicRecon uses HDF5 1.8 functions ------------------------------------+--------------------------------------- Reporter: rhaas | Owner: Type: defect | Status: new Priority: minor | Milestone: Component: EinsteinToolkit thorn | Version: development version Resolution: | Keywords: SphericalHarmonicRecon ------------------------------------+---------------------------------------
Comment (by anonymous):
I think it is probably time to require the 1.8 API.
#1994: PITTNUllCode/SphericalHarmonicRecon uses HDF5 1.8 functions ------------------------------------+--------------------------------------- Reporter: rhaas | Owner: Type: defect | Status: new Priority: minor | Milestone: Component: EinsteinToolkit thorn | Version: development version Resolution: | Keywords: SphericalHarmonicRecon ------------------------------------+---------------------------------------
Comment (by anonymous):
Replying to [comment:1 anonymous]:
I think it is probably time to require the 1.8 API.
This was Frank...
#1994: PITTNUllCode/SphericalHarmonicRecon uses HDF5 1.8 functions ------------------------------------+--------------------------------------- Reporter: rhaas | Owner: Type: defect | Status: new Priority: minor | Milestone: Component: EinsteinToolkit thorn | Version: development version Resolution: | Keywords: SphericalHarmonicRecon ------------------------------------+---------------------------------------
Comment (by knarf):
On the ET users call on 9. Jan 2017 was decided that the best course of action is to make 1.8 a requirement for the toolkit.
#1994: PITTNUllCode/SphericalHarmonicRecon uses HDF5 1.8 functions ------------------------------------+--------------------------------------- Reporter: rhaas | Owner: Type: defect | Status: new Priority: minor | Milestone: Component: EinsteinToolkit thorn | Version: development version Resolution: | Keywords: SphericalHarmonicRecon ------------------------------------+---------------------------------------
Comment (by hinder):
I think this is good, but would like to know Erik's opinion. I think he joined the call after we discussed it.
#1994: PITTNUllCode/SphericalHarmonicRecon uses HDF5 1.8 functions ------------------------------------+--------------------------------------- Reporter: rhaas | Owner: Type: defect | Status: new Priority: minor | Milestone: Component: EinsteinToolkit thorn | Version: development version Resolution: | Keywords: SphericalHarmonicRecon ------------------------------------+---------------------------------------
Comment (by eschnett):
I was not there during the discussion. HDF5 is now at version 1.10, so requiring 1.8 is possible. I am not aware of an HPC system that only has HDF5 1.6, and HDF5 is easy to install. Plus, we started supporting HDF5 1.8 in Cactus more than ten years ago, so this is not a surprise.
Now, HDF5 1.10 does have some very interesting features that we should use (or explore), such as e.g. virtual datasets (no more recombining!), or the new file format that promises much higher performance. But that's for another time.
trac@lists.einsteintoolkit.org
-
Einstein Toolkit