How can non-built-in type instances be immutable in python?

Question

If I try to run

import numpy as np

type(np.dtype).moo = 7

then I get

TypeError: cannot set 'moo' attribute of immutable type 'numpy._DTypeMeta'

which I can't find an explanation for.

numpy._DTypeMeta is a type instance, so this should be roughly equivalent to

type.__setattr__(numpy._DTypeMeta, "moo", 7)

So what is happening in type.__setattr__ that detects numpy._DTypeMeta should be immutable?

I could imagine that immutable built-ins such as float, might be stored in a specific place in logical memory, and that type.__setattr__ could check for that, but numpy isn't built-in.

Answer 1

This is an interesting question, and I'm curious to see an answer from someone with a better understanding of the C API.

Here is my quick incomplete (and probably incorrect) take:

• numpy._DTypeMeta is defined here https://github.com/numpy/numpy/blob/a115ed3a3c92d8607c5eaa38a98825ac8a5e1bfc/numpy/core/src/multiarray/dtypemeta.c#L964
• The immutable flag (i.e. Py_TPFLAGS_IMMUTABLETYPE) is not explicitly applied, but...
• PyType.Ready() automatically applies the Py_TPFLAGS_IMMUTABLETYPE - https://docs.python.org/3/c-api/typeobj.html#c.Py_TPFLAGS_IMMUTABLETYPE
• numpy._DTypeMetadata type is thus immutable.