Re: [PATCH v3 1/5] [gdb/python] Introduce py_{none, true, false, notimplemented} functions

[Tom Tromey <tom@tromey.com>]

>>>>> "Tom" == Tom de Vries <tdevries@suse.de> writes:

Tom> +/* The following four functions are refcount-safe wrappers around
Tom> +   Py_RETURN_{NONE,TRUE,FALSE,NOTIMPLEMENTED}.
Tom> +
Tom> +   Starting with python 3.12, None, True, False and Py_NotImplemented are
Tom> +   immortal, and increasing and decreasing refcount for such objects is a
Tom> +   no-op, and consequently for the limited C API 3.12 and newer,
Tom> +   Py_RETURN_NONE is simply "return Py_None".
Tom> +
Tom> +   So for the limited C API 3.12 and newer, we could just return a
Tom> +   "PyObject *" instead.

I don't think this sentence is correct.  I mean, it might work in some
places, but using this universally would severely uglify gdb because
then spots operating generically would have to check specifically for
such objects.  IMO object immortality is really best thought of as an
implementation detail of CPython and not an exposed feature.

Tom> +   For the limited C API 3.12 and newer, while returning gdbpy_ref<> there's
Tom> +   an imbalance (we do a Py_DECREF in gdbpy_ref_policy::decref, without
Tom> +   corresponding Py_INCREF in Py_RETURN_NONE), but this is not harmful because
Tom> +   Py_DECREF is no-op for immortal objects.  */

Tom> +static inline gdbpy_ref<>
Tom> +py_notimplemented ()
Tom> +{
Tom> +  auto f = [] { Py_RETURN_NOTIMPLEMENTED; };
Tom> +  return gdbpy_ref<> (f ());
Tom> +}

I don't think we use this and I think we shouldn't add it until it is needed.

Tom