I’m trying to wrap my head around the use case this was created for. If a python service dumps a numpy array to bytes and sends that as a payload to a C++ service, is that blob of bytes not interpretable by C++ as e.g. a struct with some size fields and a pointer to the start of the array? (Cutting out the filesystem/npy file)
Maybe I’m overthinking it but I can’t quite envision a use case where a python service and C++ service would be sharing a filesystem
Many research labs generate GBs of data as npy files. Having readers in languages other than Python is useful when you have programs in other languages that need to process them.
We still are using a few Fortran programs mixed with python programs and who know how many other languages. Anyway, we are using text files in ascii, because when something get's wrong it's easier to debug.
I’m trying to wrap my head around the use case this was created for. If a python service dumps a numpy array to bytes and sends that as a payload to a C++ service, is that blob of bytes not interpretable by C++ as e.g. a struct with some size fields and a pointer to the start of the array? (Cutting out the filesystem/npy file)
Maybe I’m overthinking it but I can’t quite envision a use case where a python service and C++ service would be sharing a filesystem
Many research labs generate GBs of data as npy files. Having readers in languages other than Python is useful when you have programs in other languages that need to process them.
You spelled Fortran wrong :)
We still are using a few Fortran programs mixed with python programs and who know how many other languages. Anyway, we are using text files in ascii, because when something get's wrong it's easier to debug.
Since an .npy file is the standard binary format for saving a single NumPy array to disk I guess I can imagine a situation where people are doing so.