Re: [PATCH] Wrap (and poison) PyObject_CallFunctionObjArgs

[Simon Marchi <simark@simark.ca>]

On 9/14/26 3:06 PM, Tom Tromey wrote:
> This patch adds a new wrapper for PyObject_CallFunctionObjArgs.  The
> wrapper function knows how to unwrap gdbpy_ref<> and
> gdbpy_borrowed_ref<>.  It rejects null arguments, and it automatically
> supplies the trailing NULL required by PyObject_CallFunctionObjArgs.
> Finally, PyObject_CallFunctionObjArgs is poisoned to avoid introducing
> new calls.
> 
> The idea behind this change is that passing anything other than
> PyObject* to this function will cause failures; this patch turns
> silent failures (like passing a gdbpy_borrowed_ref<>) into a
> compile-time failure.

LGTM, see minor comments below.

Approved-By: Simon Marchi <simon.marchi@efficios.com>

> +/* A wrapper for PyObject_CallFunctionObjArgs that takes various kinds
> +   of gdb wrappers, in addition to "PyObject *".  This variant does
> +   not allow NULL arguments.  While PyObject_CallFunctionObjArgs
> +   requires a trailing NULL, this function does not -- it supplies the
> +   required trailing NULL on its own.
> +
> +   As a safety measure, no argument may be NULL.  While this may be
> +   slightly inconvenient at times (you can't early-terminate the
> +   arguments, you have to add a special case at the call site), it
> +   avoids bugs where early termination was unintentional.  */

I think it's fine.  One example is in evpy_emit_event, and I find the
new code clearer, because it's more explicit about the two
possibilities.

> +template<typename Arg, typename... Args>
> +static inline gdbpy_ref<>
> +gdbpy_object_call_function_obj_args (Arg &&fn, Args && ...args)
> +{
> +  PyObject *result
> +    = PyObject_CallFunctionObjArgs (detail::unwrap_ref (fn),
> +				    detail::unwrap_ref (args)...,
> +				    nullptr);
> +  return gdbpy_ref<> (result);
> +}
> +
> +/* Poison PyObject_CallFunctionObjArgs.  The typesafe wrapper
> +   gdbpy_objects_call_function_obj_args should be used instead.  */

Typo, gdbpy_objects_call_function_obj_ar -> gdbpy_object_call_function_obj_args

> +#undef PyObject_CallFunctionObjArgs
> +#ifdef __GNUC__
> +# pragma GCC poison PyObject_CallFunctionObjArgs
> +#else
> +# define PyObject_CallFunctionObjArgs POISONED_PyObject_CallFunctionObjArgs
> +#endif

Clang appears to support "pragma GCC poison", could we use it
unconditionnaly?

Simon