Problem

Node.fragment(...) was missing entirely in slate-v2, even though the public node API still described it.

Porting it naively is easy to fuck up:

• mutate the live tree instead of a clone
• interpret the range against editor-global paths when the root is an element
• delete forward and invalidate later paths during the same pass

Root Cause

Node.fragment(...) is a pure node-utility helper, not an editor transaction.

That means it has to work on any root ancestor and any range expressed relative to that root, without leaning on editor-only fragment helpers or runtime ids.

Solution

Use a clone-and-prune algorithm in root-relative space:

1. deep-clone the ancestor root's children
2. walk the cloned tree with Node.nodes(...) in reverse order
3. prune nodes outside the range from the bottom up
4. trim the start and end text leaves in place

The reverse walk matters because child removals would otherwise invalidate later paths still waiting to be processed.

Why This Works

The helper stays:

• pure
• root-relative
• safe for both editor roots and element roots

It also composes cleanly with the rest of the recovered Node.* traversal surface, because the same path semantics drive both the walk and the pruning.

Prevention

• Treat Node.fragment(...) as a node-utility operation, not an editor transaction helper.
• Clone before pruning. Never mutate the live root while slicing a fragment.
• When path-based pruning is involved, walk in reverse unless you enjoy invalidating your own cursor mid-loop.