While looking up background info for #36347, I noticed that the docs for PyMemberDef ( https://docs.python.org/3/extending/newtypes.html#generic-attribute-management ) define the following flags:

* READONLY
* READ_RESTRICTED
* WRITE_RESTRICTED
* RESTRICTED

Of those, WRITE_RESTRICTED has been documented incorrectly since Python 3.0a2 (it got renamed to PY_WRITE_RESTRICTED in structmember.h, apparently due to a name clash with VS 2008 on Windows).

Luckily, the constants are useless; AFAICT they were there solely to support restricted mode (via the rexec and Bastion modules). The modules were apparently disabled in Python 2.3 "due to various known and not readily fixable security holes" ( https://docs.python.org/2/library/restricted.html ), and the modules themselves officially deprecated in 2.6, and removed completely in 3.0.

Rather than fixing up the docs, maybe it's time to kill these constants for good? Probably shouldn't remove the definitions in structmember.h (if someone wrote code using them, we shouldn't needlessly break compilation), but the interpreter and standard library should stop setting them, and the documentation for them should be removed.

To be clear, only the constants that include the substring RESTRICTED are useless; READONLY remains useful and should not be removed (though per my suggestion on #36347, it might be good to add an alias for it named PY_READONLY to match the Python C API naming conventions).

Josh, I don't know for the uselessness or not of these macros, but these macros are used.

RESTRICTED is used in Objects/classobject.c and Objects/funcobject.c

PY_WRITE_RESTRICTED is used in Objects/funcobject.c and Objects/methodobject.c

Yes, they're set. They're never *read* anywhere. My suggestion was to stop setting them (because it gives the impression they do something when reading the docs, when in fact they do nothing), and remove them from the docs.

Sorry, that should have been "it gives the impression they do something when reading the *source code*"