two_ways_to_connect.py

# Technically even in the "easy case" of t._base.requires_grad == t.requires_grad
# I need to perform two views to recreate that view authentically. why?

# There are actually two things I need to recreate, (1) the autograd
# graph relationship and (2) the view relationship.

# The reason we don't handle this today is because this autograd connectivity information
# is not accessible during tracing and hence not relevant to compile in part because dynam
# doesn't support grad_fn access. 

for requires_grad_1, requires_grad_2 in [(True, False), (False, True), (True, True), (False, False)]:
    a = torch.tensor([1.], requires_grad=requires_grad_1) # leaf
    b = a[:]  # view
    c = a.clone() # non-leaf, non-view

    for x in (a, b, c):
        with torch.no_grad():
            x_view = x[:]

        assert x_view._is_view()
        assert x_view.is_leaf
        assert x_view.requires_grad == requires_grad_1

        rg_before = x_view.requires_grad
        # !! Doing requires_grad_ here has very strange behavior
        x_view.requires_grad_(requires_grad_2)

        if rg_before:
            assert x_view.requires_grad == rg_before
        else:
            assert x_view.requires_grad == requires_grad_2

        x_view_view = x_view[:]

        if requires_grad_2:
            # (1) autograd info: points to the intermediate view
            assert x_view_view.grad_fn.next_functions[0][0].variable is x_view
        else:
            if requires_grad_1:
                assert x_view_view.grad_fn is not None
                assert x_view_view.grad_fn.next_functions[0][0] is None
            else:
                assert x_view_view.grad_fn is None

        # (2) view info: points to the original base
        if x._is_view():
            assert x_view_view._base is a
        else:
            assert x_view_view._base is x