[jit] Bad error when calling ScriptModules with attributes/parameters

[driazati]

class Vocab(torch.jit.ScriptModule):
    def __init__(self):
        self.i = torch.jit.Attribute(2, int)

    @torch.jit.script_method
    def forward(self, x):
        return x + self.i

v = Vocab()

@torch.jit.script
def lookup(y):
    return v(y)

results in

for operator forward(Tensor x, int slot0) -> Tensor:
argument slot0 not provided.
@torch.jit.script
def lookup(y):
    return v(y)
           ~ <--- HERE
:
@torch.jit.script
def lookup(y):
    return v(y)

Script functions cannot call stateful modules, so this should produce an error but it should say something to that effect instead of a schema-mismatch.

[driazati]

Semi-related errors:

1. Explicitly calling the forward method

class Vocab(torch.jit.ScriptModule):
    def __init__(self):
        self.i = torch.jit.Attribute(2, int)

    @torch.jit.script_method
    def forward(self, x):
        return x + self.i

v = Vocab()

@torch.jit.script
def lookup(y):
    return v.forward(y)

RuntimeError: 
attribute lookup is not defined on method:
@torch.jit.script
def lookup(y):
    return v.forward(y)
           ~~~~~~~~~ <--- HERE

2. Calling a non-forward method

class Vocab(torch.jit.ScriptModule):
    def __init__(self):
        self.i = torch.jit.Attribute(2, int)

    @torch.jit.script_method
    def my_method(self, x):
        return x + self.i

v = Vocab()

def lookup(y):
    return v.my_method(y)

RuntimeError: 
 expected this module to have a forward function..