Add bounded exact text item batch resolution

[MadeByDouglas]

Adds Yex.StickyIndex.resolve_text_items/2 for bounded exact item-ID runs against one XML text. It resolves all positions in one read transaction on a dirty CPU scheduler, returns live/collapsed/unavailable/wrong-type results, and checks the actual containing branch. Deleted original IDs stay collapsed when undo/redo recreates content under new IDs.

The binding uses public Yrs APIs only. Input is limited to 4,096 runs and 131,072 UTF-16 units; a separate linked-item traversal budget fails recoverably. Applications still own root reachability, plain-text validation, authorization and provenance.

This is a draft because it depends on the companion Yrs API in y-crdt/y-crdt#656. For reproducible integration, Cargo currently pins a public immutable backport of that change to Yrs 0.25. The fork pin should be replaced with a compatible released dependency before a normal package release. No machine-local paths are required.

The follow-up at a312e0c corrects two Elixir typespecs to match existing native behavior: Yex.Doc.Options.collection_id accepts nil, and Yex.Awareness.apply_update/3 can return {:error, term()} for malformed updates. It adds regressions for default collection options with UTF-16 editing and awareness decode errors preserving existing state. No Rust or runtime behavior changes are included in this follow-up.

Validation:

• Complete-suite baseline at b6fa2ba: 617 cases passed (63 doctests and 554 tests), using the actual remote Git dependency.
• Typespec follow-up at a312e0c: 70 focused checks passed (11 doctests and 59 tests) in the document and awareness test files, using the existing source-built native library. The complete suite has not been rerun at this commit.
• Shared Yjs fixtures cover emoji, multiple leaves, concurrent interior insertion, deletion and unavailable ancestors.
• Additional cases cover redo, wrong-type claims, active transactions, malformed/oversized runs and the maximum clock boundary.
• A 131,072-unit, 4,096-format-span benchmark improved from 13,222 ms before prefix caching to 36 ms afterward, checking every position. This is a local measurement, not an adversarial latency guarantee.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}