Stance on having multiple versions of hdf5 library in the same process?

[bartoli]

Use case 1 :
A conda environment has h5py installed (which depends on a conda packages that installs hdf5 (hdf5 adhdf5_hl)), and fiona (that ships dependencies as renamed libraies, including hdf5 (hdf5 and hdf5_hl). I am building another module that uses it's own build of hdf5 (hdf5 and hdf5_cpp).
All seem to not conflict together and the process runs normally

Use case 2:
I have an integration test that temporarilly load 2 versions of hdf5 in memory (hdf5, hdf5_hl, and hdf5_cpp for both). Here, on linux, H5::DataSpace::getConstant() throws an exception because it does not like being called twice

So the question is, is there a subset of hdf5 that is considered to be able to work with multiple version of hdf5 in memory, or are h5py/fiona just lucky that they generally don't get conflicts that trigger crashes? Is the non c++ part of hdf5 considered multi-version -compatible as long as the most recent one is loaded first maybe?