[L2Ork-dev] dynamic patching and undo/redo system

[Jonathan Wilkes]

Hi all,

It appears that dynamic patching nnd the undo system are irreconcilable.

For example-- Suppose you do this:

[namecanvas foo]

[obj 20 20 clip(
|
[send foo]

Whenever you click the message box, a new object will be created but
without any undo history.

That's probably desirable behavior, as we don't want a bunch of
dynamically created gumming up the undo system.

However, the undo system is based off object indices. So by inserting
"[clip]" into a glist we corrupt that system. You can see this if
dynamic patching in the middle of other editing actions. Doing undo
then redo will put the patch in a corrupt state because the indices
will be off by one due to the "[clip]" object that the undo system
didn't expect.

I don't know any clean way to deal with this. But it affects more than
dynamic patching:

[pointer(
|
[canvasinfo]
|
[20 20 $1(
|
[append some_scalar x y]

Same problem there, as well as:

[pointer(
|
[canvasinfo]
|
[next, delete(
|
[pointer some-scalar]

Each one breaks the assumptions of the undo system and undo/redo
actions can have wrong indices.

Even more pressing-- now that we've got pointer delete method, the
undo system can have an out-of-bounds index and crash when it tries to
deference a null pointer.

Any ideas on how to approach this?

-Jonathan