Error in post-attach does not undo the operation

[kayjan]

Taking the read-only example and implementing checks on post_attach instead of pre_attach, it does not undo the operation that is supposed to be atomic.

Sample code to replicate the issue below,

from anytree import NodeMixin


class SampleNode(NodeMixin):
    def __init__(self, foo, parent=None):
        super(SampleNode, self).__init__()
        self.foo = foo
        self.readonly = False
        self.parent = parent
    def _post_attach(self, parent):
        if self.root.readonly:
            raise ValueError()
    def _post_detach(self, parent):
        if self.root.readonly:
            raise ValueError()

a = SampleNode("a")
a0 = SampleNode("a0", parent=a)
a1 = SampleNode("a1", parent=a)
a1a = SampleNode("a1a", parent=a1)
a2 = SampleNode("a2", parent=a)

assert a1.parent == a

a.readonly = True
with pytest.raises(ValueError):
    a1.parent = a2

a1.parent == a2 # returns True, it should undo the operation

[c0fec0de]

Only the _pre_* hookups prevent the tree manipulation (see Important note here https://anytree.readthedocs.io/en/latest/intro.html#custom-separator)

Will add another note, at the example.

[kayjan]

This is not an issue on pre- vs post- hookups, the example was just to simulate an error.

The issue is that any error should rollback to the previous state and undo the operation, but that is not the case.

[c0fec0de]

Got your point.

The rollback may be complicated. It is just another operation on tree, which cause another pre/post-hookup-processing. Which might cause errors as well. And how to handle them.

My decision was to keep-it-simple. pre-hookup errors prevent the tree manipulation. post-hookups not. This is what is also clearly stated in the documentation.

What is your exact use case for handling the post-hookup errors?

[kayjan]

Yeah understand that it is complicated to implement. I was thinking of using a try-except clause where the usual flow is within the try clause and rollback is implemented in the except clause, instead of implementing another pre/post-hookup-processing.

A possible use case would be to use pre-attach and post-attach to check the tree state. If there are errors in pre- or post-attach checks, then a rollback should happen such that the tree does not exist in an illegal state.

[c0fec0de]

I am aware of the try-except approach. Nevertheless, it gets quite complicated if the rollback fails too. Would like to skip for now.