[AdaWorldAPI]

This PR implements D-QSP-0 remainder slice 1 (track C in EXO-50 /033, claimed as session_013z2e1YXkm6o6HNrDn5BPWA/026). It closes gap G7 from entries/2026-10-21-pg-classview-polyfill-map.md.

Defect. Scalar Agg::SumI32 lowers to MaskedSumI32, which starts from 0. A selection with no non-NULL x therefore answered 0, where PostgreSQL answers NULL. Values that really sum to zero also answer 0, so the two cases could not be told apart.

Change (crates/lance-graph-quack/src/lib.rs, next to AvgPlan):

• lower_sum(filter, val, valid) lowers the existing MaskedSumI32 and Count terminals, both over filter ∧ valid(val). Because the function takes the validity plane itself, the NULL gate cannot end up on one half and not the other.
• sum_finish(sum, count) -> Option<i64> returns NULL when count == 0.
• SumPlan is a type alias of AvgPlan.
• There is no new terminal, operator, rule or registry. set_algebra, set_rules, relalg and mask-risc are untouched, and so is every Semijoin path (ADMISSION B).

Supported SQL subset, as pinned. One table with two nullable int4 columns.

• WHERE: any tree, depth ≤ 3, of comparisons (= <> < <= > >=) against an integer constant, IS [NOT] NULL, AND, OR and NOT, read through sql_where.
• Aggregates: count(*), count(x), sum(x), min(x), max(x) and avg(x), ungrouped.
• P-EMPTYSUM: an empty WHERE gives one NULL ungrouped and zero groups grouped (GroupReduce { SumI32 } after normalize_group_sink).

Differential results

• PostgreSQL 16.15 (examples/qsp_agg_pg_parity.rs, 400 tables, 3600 queries): Quack, PostgreSQL and the Kleene oracle agree on all six aggregates, with 0 mismatches.
  • The corpus contains 1455 empty selections, 507 all-NULL selections and 64 real zero sums.
  • The naive scalar SUM differs on exactly 1962 = 1455 + 507 queries, and the example asserts this.
  • AVG is compared with a relative tolerance of 1e-12, because PostgreSQL's avg(int) is numeric.
  • Without a server the example prints SKIP and exits 0.
• tests/sql_aggregate_null.rs compares the same six aggregates against a column-lane Kleene oracle (folds over positions, no row objects) on 1200 random WHERE clauses. It asserts that the corpus contains empty, all-NULL and real-zero cases. It also has targeted NULL-vs-zero cases and the P-EMPTYSUM pin.

Disable runs (each red at an assertion line):

• D1 (sum_finish always returns Some): both sum tests fail. The PostgreSQL example reports 1962 sum(x) mismatches and exits 101.
• D2 (lower_sum ignores valid): both sum tests fail.
• D3 (only the sum half is gated): both sum tests fail.
• D4 (normalize_group_sink marks every slot present): only the P-EMPTYSUM test fails.

Unsupported, listed in the entry:

• grouped Agg::GroupSumI32, which still coalesces an unreached group to 0 (documented);
• a reached group whose x are all NULL (T4), and NULL group keys (G2), which belong to the GROUP BY slice;
• NULL literals and NOT IN (…, NULL) (G1);
• COALESCE and NULLIF;
• nullable arithmetic;
• int4 sum overflow, and numeric AVG.

There is no performance claim. SUM uses two passes, the same as AVG.

Board: entry entries/2026-10-21-d-qsp-0-scalar-aggregates-null.md, the D-QSP STATUS_BOARD row, and the regenerated entries index. The quack crate passes every suite, and clippy -D warnings is clean (Rust 1.99.0).

---

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: e19edb55-904e-409d-b1e2-8992b1330516

• 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 2026-10-21T19:50:56.147925Z	0d2d5b1	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.