Join forces with anytree

[Laurent Verweijen]

Would it be an idea to join forces with anytree?
I see a lot of overlap between both projects and perhaps a merger can combine the best of both worlds.
Since both have a MIT license, it might not even be that difficult.

One difference is that bigtree depends on pandas, whereas anytree is python-pure.
Maybe pandas can be made optional or even dropped if there is no performance difference.

Projects:

• anytree
• bigtree

[Abdur-Rahmaan Janhangeer]

@lverweijen Bigtree is really cool and maintained. Feature-wise it outshines anytree. At this point, it's a bit too late maybe as i see anytree as a subset of bigtree.

[Laurent Verweijen]

@Abdur-rahmaanJ bigtree has more features (I think), but anytree's code / design looks more fleshed out.
I was thinking of maybe adding to anytree what's missing there, but already present in bigtree.

Something I would like anytree to have is more options for importing / exporting to different formats.
Since the apis are similar enough, it shouldn't be that hard to port features back and forth, but I would hope for a collaboration.

edit: I originally wrote "anytree has more features". I meant "bigtree" has more. Although, they both have features lacking from the other.

[Abdur-Rahmaan Janhangeer]

@lverweijen Can you make a list of features / imports needed / todos?

• x
• y
• z

Thanks ^^.

[Laurent Verweijen]

Here are some differences:

• anytree has NodeMixin, bigtree has BaseNode. NodeMixin is slightly more flexible, because it doesn't require a name attribute.
• anytree has SymlinkNode. It's like a shallow copy of a Node.
• anytree has separator as class attribute. bigtree has sep on root. I prefer class attribute, because of O(1) lookup time.
• anytree is python only, bigtree has a dependency on pandas. Having it as an optional dependency is preferable.
• anytree has resolver. bigtree.finds_paths comes close, but doesn't support wildcards.
• anytree has ZigZagGroupIter. Not sure if it's needed.
• bigtree has type annotations in the codebase. These are missing from anytree.
• bigtree has BinaryNode. Basically a node with at most 2 children.
• bigtree has DAGNode. A node with multiple parents.
• bigtree has import / export to / from list, dict, nested dict, dataframe and list. anytree only supports json and dicts.
• bigtree has bulk modification functions (shift_nodes/copy_nodes). Not sure what benefits they have over modifying nodes directly.
• bigtree has workflows. They seem a bit too specific to include in the library itself. Maybe include them in documentation/examples instead.

There are a few ways to continue from:

1. Add one project to the other. So either anytree should consume bigtree or bigtree should consume anytree.
2. Start a new project that is the successor to both, to which both source packages can contribute. About individual differences a discussion can be started.
3. The projects should gradually grow towards each other, copying features until they are exactly the same and maybe merge over time.

[Kay Jan W.]

Hello, thanks for your comprehensive comparison! To address your points,

• BaseNode in bigtree does not require name attribute, which I would think it is similar to NodeMixin as it is easily extendable. Similarly, I don't see a need for SymlinkNode because in any case, users can just copy or extend Node for their usecase, unless I have understood the usage and purpose of SymlinkNode wrongly. Examples of how to extend Node can be found in the documentation.
• separator/sep should be consistent for the whole tree, which should not be implemented as a class attribute for each node. If you do this on anytree, you will notice the issue,

from anytree import Node
a = Node("a")
b = Node("b", parent=a)
b.separator = "-"
b
# Node('-a-b', separator='-')
a
# Node('/a')

• Making pandas an optional library has been raised as an issue previously, you can refer to the issue here.
• Resolver is interesting! This can be a future enhancement 👍
• I didn't see a need for ZigZagGroupIter, but this can be a possible future enhancement as well!

Moving forward, I'd be happy to continuously enhance and fix bigtree, do continue to raise issues as well. Thanks for your support on bigtree!

[Laurent Verweijen]

separator/sep should be consistent for the whole tree, which should not be implemented as a class attribute for each node. If you do this on anytree, you will notice the issue

You are right about that. If using a class attribute, it can perhaps be prevented by using type(self).separator instead.

[Kay Jan W.]

If you use type(self).separator, you are using the separator of anytree.node.node.Node class which is always /. You solve the issue on separator discrepancy, but you sacrifice on customizability; users cannot choose their own separator since it is referencing the default Node class.

Given these concerns, I would advise against using class attribute and have the separator/sep synced and consistent yet customizable for the whole tree, which is why I chose to implement it as a class property referencing root node's sep.

[Kay Jan W.]

The following enhancements has been made available on bigtree v0.10.0

• find_relative_path: Similar to Resolver from anytree, able to find relative path with ./../* notations
• zigzaggroup_iter: Similar to ZigZagGroupIter from anytree
• zigzag_iter: Zig Zag iteration, not present in anytree

Do upgrade bigtree with pip install --upgrade bigtree to get the latest changes! 😄

I will be closing this ticket as well, if there are any new features, enhancements, or bugfix, do raise another issue.

[Kay Jan W.]

Update: Pandas is now an optional library in v0.12.0!

Thanks for all your support and suggestions in making bigtree better 😄