After thinking about #5 and #3 for a while, we possible need to come up with a way better high-level testing framework. This idea is somewhat inspired by the TOML community and one of their testing frameworks: If we really want to test HSD implementations, we have to store the data in a programming language independent format, the tree representation in DFTB+ or a Python dict is not such a format, but the XML format which inspired HSD would certainly be suitable.
Implementing a serializer to write HSD data as XML would allow to reread the XML and compare to a reference XML by a trusted decoder implementation, this would ensure the decoding of HSD is working correctly. A similar scheme could test the encoding, but requires a trusted HSD implementation (e.g. DFTB+).
After thinking about #5 and #3 for a while, we possible need to come up with a way better high-level testing framework. This idea is somewhat inspired by the TOML community and one of their testing frameworks: If we really want to test HSD implementations, we have to store the data in a programming language independent format, the tree representation in DFTB+ or a Python dict is not such a format, but the XML format which inspired HSD would certainly be suitable.
Implementing a serializer to write HSD data as XML would allow to reread the XML and compare to a reference XML by a trusted decoder implementation, this would ensure the decoding of HSD is working correctly. A similar scheme could test the encoding, but requires a trusted HSD implementation (e.g. DFTB+).