Change default_datatype(df::HDF5.H5DataStore) to a new lazy wrapper type LH5OrderedDict <: AbstractDict{String,Any} or so.
Representing HDF5 groups without a "datatype" attribute as NamedTuples is often not the right thing to do semantically. It also places a high load on the compiler and results in a deep recursive metadata read.
Change
default_datatype(df::HDF5.H5DataStore)
to a new lazy wrapper typeLH5OrderedDict <: AbstractDict{String,Any}
or so.Representing HDF5 groups without a "datatype" attribute as
NamedTuple
s is often not the right thing to do semantically. It also places a high load on the compiler and results in a deep recursive metadata read.