Cannot patch __call__ on an object with vectorcall defined

It appears that when the type of an object has JAX_TPFLAGS_HAVE_VECTORCALL set on tp_flags, then overrides to call from within python are ignored.

• PyObject_Call checks if vectorcall is available (which is only a property of the type), and if so, then does the vectorcall: https://github.com/python/cpython/blob/3.11/Objects/call.c#L328
• The implementation of vectorcall does not check for tp_call (is this where call ends up?) https://github.com/python/cpython/blob/3.11/Include/internal/pycore_call.h#L39

I encountered this bug when using the latest version of JAX, which introduced vectorcalls in https://github.com/tensorflow/tensorflow/commit/bf3eb110f49fe9628a31d8e2052951cd6d0082f0 :

import jax
g = jax.jit(lambda x: x + 1)
print(type(g))
# > <class 'google3.third_party.tensorflow.compiler.xla.python.xla_extension.CompiledFunction'>
print(g(5))
# > 6
type(g).__call__ = lambda *args: 0
print(g(5))
# > 6

Commencez par reproduire l’exemple JAX, puis lisez dans Objects/call.c la partie autour de la vérification de vectorcall de PyObject_Call et dans Include/internal/pycore_call.h la partie autour de la distribution de vectorcall. Déterminez l’interaction attendue avec une redéfinition de call au niveau Python et ajoutez une couverture de régression montrant le comportement prévu.