[nabobalis]

PR Description

Deep copying ExtraCoords, GlobalCoords or a plotter also deep copied the parent cube, data included. _new_instance deep copies the coords and the uncertainty, whose parent_nddata is the cube, so arithmetic, to and fill_masked copied the whole cube each time, which took RAM over 20 GB on some of my cubes.

• ExtraCoords, GlobalCoords and BasePlotter get a __deepcopy__ that doesn't copy the parent cube.
• _new_instance and fill_masked no longer copy the cube through the uncertainty's parent_nddata, nor copy attributes the operation replaces.

AI Assistance Disclosure

AI tools were used for:

• Code generation (e.g., when writing an implementation or fixing a bug)
• Test/benchmark generation
• Documentation (including examples)
• Research and understanding
• No AI tools were used

Regardless of AI use, the human contributor remains fully responsible for correctness, design choices, licensing compatibility, and long-term maintainability.

[alasdairwilson]

Once more I find myself in a nabil PR that is fixing the thing I came to ndcube to fix, though in my case the issue with this was not ram usage but thread safety because copying the extracoords means deep copying the entire parent cube via this stale _ndcube which means it iterates the parent's dict while another thread is writing to it.

I think this could be more generally, i.e. not just inside _new_instance, and in a more human readable and still have equivalent fix if we patched it in a less obtuse way by simply making the deepcopy not copy the parent cube at all _ndcube from the deepcopy method on both extracoords and globalcoords...

    def __deepcopy__(self, memo):
        new = copy.copy(self)
        for key, value in vars(self).items():
            if key != "_ndcube":
                setattr(new, key, copy.deepcopy(value, memo))
        return new

or i guess you are supposed to work on the memo or something with memo[id(self)] = new first?

then no matter who or what calls deepcopy its threading safe and not memory inefficient

[Cadair]

+1 for making sure any deepcopy of a NDCube works without copying the parent cube etc

[Cadair]

Thanks, I hate it.