[Python-Dev] enum in the stable ABI (Was: PEP 558: Defined semantics for locals)

23 Jul 2021

      On 22. 07. 21 12:41, Nick Coghlan wrote:
...
On Thu, 22 Jul 2021, 6:01 pm Petr Viktorin, mailto:encukou@gmail.com> wrote:
On 21. 07. 21 14:18, Nick Coghlan wrote:
     > On Mon, 19 Jul 2021 at 21:32, Petr Viktorin mailto:encukou@gmail.com> wrote:
     >> The proposal assumes that in the future, ``PyLocals_Get``, and thus
     >> ``locals()``, will never gain another kind of return value, however
     >> unlikely that is.
     >> AFAICS, code that uses this will usually check for a single
    special case
     >> and fall back (or error) for the other(s), so I think it'd be
    reasonable
     >> to make this an "enum" with two values. e.g.:
     >>
     >> int PyLocals_GetReturnBehavior();  # better name?
     >> #define PyLocals_DIRECT_REFERENCE 0
     >> #define PyLocals_SHALLOW_COPY 1
     >
     > After looking at PyUnicode_Kind, PySendResult, and other already
     > public enums for inspiration, my proposed spelling is as follows:
     >
     > ====================
     > typedef enum {
     >      PyLocals_UNDEFINED = -1;
     >      PyLocals_DIRECT_REFERENCE = 0,
     >      PyLocals_SHALLOW_COPY = 1
     > } PyLocals_Kind;
     >
     > PyLocals_Kind PyLocals_GetKind(void);
     > PyLocals_Kind PyFrame_GetLocalsKind(PyFrameObject *);
     > ====================
     >
     > The PyLocals_UNDEFINED case comes from PyLocals_GetKind() needing an
     > error value to return when the query API is called with no active
     > thread state.
     >
     > I've updated the draft reference implementation to use this API, and
     > added the associated PEP changes to the review PR at
     > https://github.com/python/peps/pull/2038/files
    https://github.com/python/peps/pull/2038/files
Please don't put the enum in the stable ABI. If we would add another
    value and then an older extension would receive it, we'd get undefined
    behavior.
Hmm, I was copying an example that is already in the stable ABI 
(PySendResult).
I think it's new in 3.10, though, so it should still be possible to fix 
that.
After researching a bit more, I see that casting unknown values to enum 
is only undefined/unspecified behavior in C++. But we do support C++ 
extensions, and so I'll try to get enums out of the stable ABI.

(In both C & C++, the size of an `enum` is implementation-defined. 
That's unlikely to be a problem in practice, but one more point against 
enum.)

NB. I don't have access to the actual standards; feel free to check this 
if you do!