Hi. This is an ABI stability and portability question. Suppose I write a C program that uses the numerical values returned by numpy.dtype.num to identify types, can these values be relied upon to be consistent in future releases of numpy? I'm only talking about base types like np.int32, np.float64 and such. What if this was about "int", which doesn't have an explicit size defined? Along the same lines, what about different architectures? What if I'm cross-compiling: building for arm64 on an amd64 box, for example. On the build machine (amd64) I would run the amd64 python to get values of numpy.dtype.num, and use those to build arm64 code that would then talk to the arm64 python. Can the numpy.dtype.num values be relied upon to be consistent across architectures? I'd like the answer to both questions to be "yes" :) Thanks!