[AdaWorldAPI]

Lab crates only. Implements the no-boundaries ruling for joins. Values stay in the relation's lanes, and the only thing preparation builds is the global-to-local index map $i.

What changed

mork-pathmap-probe

• RelationView holds two lanes (arrow_buffer::ScalarBuffer<u32>, refcounted, shared with Lance column buffers). The row fixture is from_rows.
• Side is an index map. It holds a presence plane, per-slot offsets, and row positions into the lanes. Run reads lane[pos[k]] in place.
• JoinExec::answers is gone. Answers are folds:
  • bag_count reads offsets only.
  • matching(pattern) with a pinned center inverts the center to one slot.
  • for_each_answer yields a JoinRow that is read in place.
• JoinDomain handles dense id sets as a presence mask plus a running count per word: slot = rank + popcount, in value order, with no sort or search. Sparse sets keep the sorted list.
• A side's keep is a mask over row positions. prepare_where and prepare_star_where walk only the set bits.
• The reference is a counting oracle. MORK/PathMap row sets stay as outside comparators, checked in both directions.

crepe-plan-probe

• Store holds RelationViews. Lance reads hand over the column buffers with no to_vec, and restricted reads are no longer trimmed with retain.
• A restriction is one ndarray::simd::eq_u32_to_mask over the lane. Semijoin centers are a domain probed per row.
• Derivations and slot views read Run with a one-slot prefetch lookahead (prefetch_read_t0).

Verification

• cargo +1.99.0 fmt, clippy --release --all-targets -D warnings, and test --release --all-targets are green in both crates.
• Disable runs, each red and then restored:
  • Run::get and iter skip positions.
  • for_each_set_ahead drops the last slot.
  • prepare_star_where ignores keep.
  • The dense rank drops the popcount.
  • The restriction compare uses the wrong constant.
  • Semijoin centers admit everything.
  • The kept-row iterator drops the last row or yields bits past n.
• examples/measure.rs: with pushdown and semijoin, prepare is µs at every n up to 50 000, the mask and terminal steps stay ns to low µs, and every oracle agrees.

Open: one failed gate

The ablation arms without pruning (no-semijoin, none) at n = 50 000 index 56k and 119k rows and go over the µs budget. Building $i from unordered lanes is a counting sort, and its count and scatter passes cost about 10 ns per row on this host. A rank-free dense counting sort was tried and did not help, because the passes are bound by the random writes.

The candidate fix is lanes sealed in key order at write time, using the OrderedLaneWitness path that addressed.rs already attests. Then $i is just run offsets. That is a write-path decision and is not taken here. Recorded in .claude/board/entries/2026-10-21-zero-copy-join-index-map.md.

Known remaining tuple sites, not touched: crepe reference::eval (the oracle) and the external-engine timings in mork-pathmap-probe/examples/cost.rs.

---

Generated by Claude Code

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

• Configuration used: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: 60a3410a-098e-4b64-b48d-9402c8def6f1

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Warning

Billing warning: we have not been able to collect payment for this subscription for more than 72 hours. Please update the payment method or pay any pending invoices in Billing to avoid service interruption.

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.}

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-21T13:16:40.957057Z	cd499c4	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.