Skip to content

Commit 5097afa

Browse files
borisbatclaude
borisbat
and
claude
committed
docs/bench: refresh results.md + sqlite_linq_gaps.md + API_REWORK chunk N+1
results.md — full INTERP+JIT rerun across benchmarks/sql/ per the living-doc policy (this PR touches sqlite_linq.das). 4 new m1 SQL cells populated (distinct_by_order_take/distinct_by_order_to_array/ reverse_distinct_by/groupby_first); drift on existing cells mostly converges back toward pre-#2906 baselines. indexed_lookup m4 now populated (482.9 INTERP / 107.5 JIT — earlier `—` was stale). Notes section drops the 4 SQL window-function blocker bullets; reverse_distinct_by note trimmed to the (decs-only) remaining gap. sqlite_linq_gaps.md — entire "Window-function lowerings" group struck through. Closed in PR #2906 (2 cells) + PR #YYYY (4 cells via bare-aggregate composition + MAX(pk) + tuple projection). Note that 3 of the 4 cells in this PR did NOT need window functions — the gaps doc's original ROW_NUMBER prescription was conservative. API_REWORK.md — new "Shipped — chunk N+1" section covering all 3 arms: Arm A passthrough composition (closes silent-drop bug), Arm B reverse() MAX(pk), Arm C group-by + first() tuple projection. Strikes the 2 now-shipped items from chunk N's "Deferred to N+1" list. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
1 parent ff98a19 commit 5097afa

3 files changed

Lines changed: 231 additions & 182 deletions

File tree

‎benchmarks/sql/results.md‎

Lines changed: 138 additions & 150 deletions
Original file line numberDiff line numberDiff line change
@@ -1,6 +1,12 @@
11
# Benchmarks — SQL / Array / Decs comparison
22

3-
Generated 2026-05-27 from PR `bbatkin/sqlite-linq-distinct-by-aggregates` (sqlite_linq `_distinct_by` as chain operator + first-row aggregates). Two SQL `—` cells now populated: `order_distinct_take` m1 = 139.2 INTERP / 138.2 JIT via a new 1-column `Brand` table fixture (named-tuple projection lets `_sql` lower to clean `SELECT DISTINCT "value" FROM "Brands" ORDER BY "value" ASC LIMIT ?`); `distinct_count_pred` m1 = 253.1 INTERP / 252.8 JIT via the new bare-aggregate `_distinct_by(K) |> _count(P)` wrap (lowers to `SELECT COUNT(*) FROM (SELECT *, MIN("id") FROM "Cars" GROUP BY "brand") WHERE "year" > ?`). The remaining SQL `—` cells stay catalogued in [`sqlite_linq_gaps.md`](sqlite_linq_gaps.md). Other drift is bench-suite ordering noise: `distinct_by_order_take` m4 INTERP 23.4 → 30.9 (+32%), `groupby_select_order` m4 INTERP 18.7 → 24.2 (+29%), `select_where_order_take` m3f/m4 INTERP ±6-8% — same long-running pattern as prior PRs; sub-nanosecond JIT cells drift at measurement floor.
3+
Generated 2026-05-27 from PR `bbatkin/sqlite-linq-distinct-by-composition` (sqlite_linq `_distinct_by` composition + `reverse() |> _distinct_by` MAX(pk) + `_group_by |> first()` row projection). Four SQL `—` cells now populated, draining the entire "window-function lowerings" group from [`sqlite_linq_gaps.md`](sqlite_linq_gaps.md) — none of them actually needed window functions, all lower via SQLite's bare-aggregate optimization with outer ORDER BY / LIMIT or MAX(pk) variant:
4+
- `distinct_by_order_take` m1 = 240.5 INTERP / 253.6 JIT
5+
- `distinct_by_order_to_array` m1 = 240.6 INTERP / 239.3 JIT
6+
- `reverse_distinct_by` m1 = 312.5 INTERP / 297.1 JIT (MAX(pk) variant)
7+
- `groupby_first` m1 = 255.1 INTERP / 252.4 JIT (tuple-with-row projection)
8+
9+
Drift is minor and mostly converges back toward pre-#2906 baselines: `distinct_by_order_take` m4 INTERP 30.9 → 23.8 (-23%) and `groupby_select_order` m4 INTERP 24.2 → 18.8 (-22%) reverse the post-#2906 drift entries. `indexed_lookup` m4 now populated (482.9 INTERP / 107.5 JIT) — earlier `—` was stale bench-output parsing. Remaining JIT cells drift ±5% at measurement floor.
410
Fixture size: n = 100 000 (cars), 100 dealers, 5 brands. Each row is
511
one bench family in `benchmarks/sql/`; columns are nanoseconds per
612
logical operation. `—` marks an intentionally absent lane — see
@@ -27,157 +33,156 @@ before the timer resolution can measure them — they should be read as
2733

2834
| Benchmark | SQL (m1) | Array (m3f) | Decs (m4) | Decs vs Array |
2935
|---|---:|---:|---:|---:|
30-
| `aggregate_match` | 35.3 | 6.1 | 5.9 | 0.97× |
31-
| `all_match` | 28.1 | 3.5 | 3.5 | 1.00× |
36+
| `aggregate_match` | 34.8 | 6.0 | 5.9 | 0.98× |
37+
| `all_match` | 27.8 | 3.6 | 3.5 | 0.97× |
3238
| `any_match` | 0.00 | 0.00 | 0.00 | — |
33-
| `average_aggregate` | 30.5 | 5.9 | 8.8 | 1.49× |
34-
| `bare_order_where` | 279.7 | 116.5 | 124.8 | 1.07× |
35-
| `chained_select_collapse` | — | 17.9 | 17.7 | 0.99× |
36-
| `chained_where` | 37.0 | 6.7 | 7.1 | 1.06× |
39+
| `average_aggregate` | 30.2 | 5.9 | 8.8 | 1.49× |
40+
| `bare_order_where` | 276.7 | 117.6 | 123.5 | 1.05× |
41+
| `chained_select_collapse` | — | 17.9 | 17.6 | 0.98× |
42+
| `chained_where` | 36.9 | 6.7 | 7.1 | 1.06× |
3743
| `contains_match` | 0.00 | 2.2 | 1.4 | 0.64× |
38-
| `count_aggregate` | 30.0 | 4.2 | 4.1 | 0.98× |
44+
| `count_aggregate` | 29.8 | 4.2 | 4.1 | 0.98× |
3945
| `decs_count_bare_pred` | — | — | 4.2 | — |
40-
| `distinct_by_count` | 41.8 | 15.8 | 15.8 | 1.00× |
41-
| `distinct_by_order_take` | — | 21.8 | 30.9 | 1.42× |
42-
| `distinct_by_order_to_array` | — | 22.0 | 24.0 | 1.09× |
43-
| `distinct_count` | 41.7 | 16.0 | 15.9 | 0.99× |
44-
| `distinct_count_pred` | 253.1 | 15.9 | 16.0 | 1.01× |
46+
| `distinct_by_count` | 41.4 | 15.7 | 15.7 | 1.00× |
47+
| `distinct_by_order_take` | 240.5 | 21.9 | 23.8 | 1.09× |
48+
| `distinct_by_order_to_array` | 240.6 | 22.2 | 23.3 | 1.05× |
49+
| `distinct_count` | 42.0 | 16.1 | 15.9 | 0.99× |
50+
| `distinct_count_pred` | 253.2 | 15.8 | 16.1 | 1.02× |
4551
| `distinct_take` | 0.00 | 0.00 | 0.00 | — |
4652
| `element_at_match` | 0.00 | 0.00 | 0.00 | — |
4753
| `first_match` | 0.00 | 0.00 | 0.00 | — |
4854
| `first_or_default_match` | 0.00 | 0.00 | 0.00 | — |
49-
| `groupby_average` | 171.9 | 30.3 | 30.2 | 1.00× |
50-
| `groupby_count` | 142.9 | 19.4 | 19.2 | 0.99× |
51-
| `groupby_first` | — | 18.5 | 19.2 | 1.04× |
52-
| `groupby_having_count` | 142.4 | 19.3 | 19.2 | 0.99× |
53-
| `groupby_having_hidden_sum` | 175.2 | 24.1 | 24.4 | 1.01× |
54-
| `groupby_having_post_where` | 172.9 | 18.5 | 18.7 | 1.01× |
55-
| `groupby_max` | 174.2 | 25.0 | 25.6 | 1.02× |
56-
| `groupby_min` | 176.3 | 25.0 | 25.2 | 1.01× |
57-
| `groupby_multi_reducer` | 190.1 | 32.5 | 33.0 | 1.02× |
58-
| `groupby_select_order` | 171.4 | 18.6 | 24.2 | 1.30× |
59-
| `groupby_select_sum` | 200.9 | 36.5 | 35.9 | 0.98× |
60-
| `groupby_sum` | 172.6 | 18.8 | 18.8 | 1.00× |
61-
| `groupby_where_count` | 75.9 | 14.5 | 15.0 | 1.03× |
62-
| `groupby_where_sum` | 86.9 | 14.2 | 14.6 | 1.03× |
63-
| `indexed_lookup` | 1468.7 | 204087.0 | 475.9 | 0.00× |
64-
| `join_count` | 38.5 | 121.4 | 64.1 | 0.53× |
65-
| `join_groupby_count` | — | 186.2 | 91.0 | 0.49× |
66-
| `join_groupby_to_array` | — | 216.7 | 90.7 | 0.42× |
67-
| `join_select` | — | 149.4 | 85.2 | 0.57× |
68-
| `join_where_count` | 39.4 | 148.9 | 74.4 | 0.50× |
69-
| `last_match` | 0.00 | 5.8 | 14.1 | 2.43× |
70-
| `long_count_aggregate` | 30.0 | 4.2 | 4.1 | 0.98× |
71-
| `max_aggregate` | 31.4 | 6.0 | 6.9 | 1.15× |
72-
| `min_aggregate` | 31.5 | 6.2 | 6.9 | 1.11× |
73-
| `order_distinct_take` | 139.2 | 15.8 | 95.5 | 6.04× |
74-
| `order_reverse_normalized` | 38.6 | 16.3 | 20.0 | 1.23× |
75-
| `order_take_desc` | 38.5 | 16.2 | 19.9 | 1.23× |
76-
| `reverse_distinct_by` | — | 21.5 | — | — |
77-
| `reverse_take` | 0.10 | 0.00 | 10.2 | — |
78-
| `reverse_take_select` | 0.00 | 34.3 | 48.4 | 1.41× |
79-
| `select_count` | 0.10 | 0.00 | 2.2 | — |
80-
| `select_where` | 195.7 | 11.3 | 19.9 | 1.76× |
81-
| `select_where_count` | 33.3 | 5.2 | 7.5 | 1.44× |
82-
| `select_where_order_take` | 37.0 | 12.3 | 14.9 | 1.21× |
83-
| `select_where_sum` | 37.3 | 7.6 | 7.6 | 1.00× |
84-
| `single_match` | 0.00 | 2.9 | 5.5 | 1.90× |
85-
| `skip_take` | 0.50 | 0.10 | 0.20 | 2.00× |
86-
| `skip_while_match` | 3.5 | 5.3 | 5.3 | 1.00× |
87-
| `sort_first` | 38.3 | 11.1 | 13.4 | 1.21× |
88-
| `sort_take` | 38.7 | 16.4 | 20.7 | 1.26× |
89-
| `sort_take_select` | 38.6 | 16.2 | 20.1 | 1.24× |
90-
| `sum_aggregate` | 30.1 | 2.2 | 2.1 | 0.95× |
91-
| `sum_where` | 33.2 | 4.3 | 4.2 | 0.98× |
92-
| `take_count` | 3.6 | 0.20 | 0.40 | 2.00× |
93-
| `take_count_filtered` | — | 0.20 | 0.20 | 1.00× |
94-
| `take_sum_aggregate` | — | 0.10 | 0.10 | 1.00× |
95-
| `take_where_count` | — | 0.10 | 0.10 | 1.00× |
96-
| `take_while_match` | 7.9 | 2.4 | 2.4 | 1.00× |
97-
| `to_array_filter` | 71.0 | 11.7 | 11.8 | 1.01× |
98-
| `zip_count_pred` | — | 15.1 | — | — |
99-
| `zip_dot_product` | — | 12.7 | 10.7 | 0.84× |
100-
| `zip_dot_product_3arg` | — | 12.8 | — | — |
101-
| `zip_reverse_to_array` | — | 31.1 | — | — |
55+
| `groupby_average` | 172.6 | 30.1 | 30.3 | 1.01× |
56+
| `groupby_count` | 142.5 | 19.4 | 19.5 | 1.01× |
57+
| `groupby_first` | 255.1 | 18.5 | 19.2 | 1.04× |
58+
| `groupby_having_count` | 143.0 | 19.3 | 19.2 | 0.99× |
59+
| `groupby_having_hidden_sum` | 176.0 | 23.8 | 24.4 | 1.03× |
60+
| `groupby_having_post_where` | 172.7 | 18.6 | 18.8 | 1.01× |
61+
| `groupby_max` | 175.5 | 25.2 | 25.2 | 1.00× |
62+
| `groupby_min` | 173.2 | 25.5 | 25.8 | 1.01× |
63+
| `groupby_multi_reducer` | 191.2 | 32.1 | 32.5 | 1.01× |
64+
| `groupby_select_order` | 175.4 | 18.6 | 18.8 | 1.01× |
65+
| `groupby_select_sum` | 202.2 | 36.5 | 35.8 | 0.98× |
66+
| `groupby_sum` | 175.9 | 18.7 | 18.6 | 0.99× |
67+
| `groupby_where_count` | 75.5 | 14.5 | 14.9 | 1.03× |
68+
| `groupby_where_sum` | 86.4 | 14.2 | 14.5 | 1.02× |
69+
| `indexed_lookup` | 1442.6 | 202815.2 | 482.9 | 0.00× |
70+
| `join_count` | 38.5 | 120.6 | 64.0 | 0.53× |
71+
| `join_groupby_count` | — | 186.6 | 91.7 | 0.49× |
72+
| `join_groupby_to_array` | — | 217.1 | 92.6 | 0.43× |
73+
| `join_select` | — | 148.7 | 85.8 | 0.58× |
74+
| `join_where_count` | 39.4 | 147.2 | 74.2 | 0.50× |
75+
| `last_match` | 0.00 | 5.8 | 14.0 | 2.41× |
76+
| `long_count_aggregate` | 29.7 | 4.2 | 4.1 | 0.98× |
77+
| `max_aggregate` | 31.3 | 6.0 | 6.9 | 1.15× |
78+
| `min_aggregate` | 31.2 | 6.1 | 6.9 | 1.13× |
79+
| `order_distinct_take` | 140.2 | 15.9 | 94.8 | 5.96× |
80+
| `order_reverse_normalized` | 38.4 | 16.2 | 20.1 | 1.24× |
81+
| `order_take_desc` | 38.4 | 16.0 | 20.0 | 1.25× |
82+
| `reverse_distinct_by` | 312.5 | 21.8 | — | — |
83+
| `reverse_take` | 0.1 | 0.00 | 10.1 | — |
84+
| `reverse_take_select` | 0.00 | 34.6 | 48.4 | 1.40× |
85+
| `select_count` | 0.1 | 0.00 | 2.2 | — |
86+
| `select_where` | 193.8 | 11.3 | 19.8 | 1.75× |
87+
| `select_where_count` | 33.1 | 5.2 | 7.4 | 1.42× |
88+
| `select_where_order_take` | 37.0 | 12.3 | 15.0 | 1.22× |
89+
| `select_where_sum` | 37.3 | 7.5 | 7.5 | 1.00× |
90+
| `single_match` | 0.00 | 2.9 | 5.6 | 1.93× |
91+
| `skip_take` | 0.5 | 0.1 | 0.2 | 2.00× |
92+
| `skip_while_match` | 3.5 | 5.2 | 5.3 | 1.02× |
93+
| `sort_first` | 38.1 | 11.0 | 13.5 | 1.23× |
94+
| `sort_take` | 38.7 | 16.5 | 20.7 | 1.25× |
95+
| `sort_take_select` | 38.4 | 16.1 | 20.0 | 1.24× |
96+
| `sum_aggregate` | 30.2 | 2.2 | 2.1 | 0.95× |
97+
| `sum_where` | 32.9 | 4.3 | 4.3 | 1.00× |
98+
| `take_count` | 3.6 | 0.2 | 0.5 | 2.50× |
99+
| `take_count_filtered` | — | 0.2 | 0.2 | 1.00× |
100+
| `take_sum_aggregate` | — | 0.1 | 0.1 | 1.00× |
101+
| `take_where_count` | — | 0.1 | 0.1 | 1.00× |
102+
| `take_while_match` | 7.8 | 2.4 | 2.4 | 1.00× |
103+
| `to_array_filter` | 70.2 | 11.7 | 11.8 | 1.01× |
104+
| `zip_count_pred` | — | 15.2 | — | — |
105+
| `zip_dot_product` | — | 12.6 | 10.6 | 0.84× |
106+
| `zip_dot_product_3arg` | — | 12.7 | — | — |
107+
| `zip_reverse_to_array` | — | 31.2 | — | — |
102108

103109
## JIT
104110

105111
| Benchmark | SQL (m1) | Array (m3f) | Decs (m4) | Decs vs Array |
106112
|---|---:|---:|---:|---:|
107-
| `aggregate_match` | 35.2 | 0.40 | 0.70 | 1.75× |
108-
| `all_match` | 27.9 | 0.30 | 0.20 | 0.67× |
113+
| `aggregate_match` | 34.8 | 0.4 | 0.7 | 1.75× |
114+
| `all_match` | 27.7 | 0.3 | 0.2 | 0.67× |
109115
| `any_match` | 0.00 | 0.00 | 0.00 | — |
110-
| `average_aggregate` | 30.5 | 1.0 | 3.6 | 3.60× |
111-
| `bare_order_where` | 187.3 | 34.2 | 35.2 | 1.03× |
116+
| `average_aggregate` | 30.2 | 1.0 | 3.6 | 3.60× |
117+
| `bare_order_where` | 185.6 | 34.1 | 34.1 | 1.00× |
112118
| `chained_select_collapse` | — | 2.1 | 2.1 | 1.00× |
113-
| `chained_where` | 36.7 | 0.60 | 0.90 | 1.50× |
114-
| `contains_match` | 0.00 | 0.20 | 0.10 | 0.50× |
115-
| `count_aggregate` | 29.6 | 0.40 | 0.60 | 1.50× |
116-
| `decs_count_bare_pred` | — | — | 0.60 | — |
117-
| `distinct_by_count` | 41.4 | 2.1 | 2.1 | 1.00× |
118-
| `distinct_by_order_take` | — | 2.7 | 3.2 | 1.19× |
119-
| `distinct_by_order_to_array` | — | 2.7 | 3.3 | 1.22× |
120-
| `distinct_count` | 41.5 | 2.1 | 2.1 | 1.00× |
121-
| `distinct_count_pred` | 252.8 | 2.1 | 2.3 | 1.10× |
122-
| `distinct_take` | 0.00 | 0.00 | 0.00 | — |
119+
| `chained_where` | 36.9 | 0.6 | 0.9 | 1.50× |
120+
| `contains_match` | 0.00 | 0.2 | 0.1 | 0.50× |
121+
| `count_aggregate` | 29.7 | 0.4 | 0.6 | 1.50× |
122+
| `decs_count_bare_pred` | — | — | 0.6 | — |
123+
| `distinct_by_count` | 42.0 | 2.1 | 2.1 | 1.00× |
124+
| `distinct_by_order_take` | 253.6 | 2.6 | 3.2 | 1.23× |
125+
| `distinct_by_order_to_array` | 239.3 | 2.7 | 3.2 | 1.19× |
126+
| `distinct_count` | 42.2 | 2.1 | 2.1 | 1.00× |
127+
| `distinct_count_pred` | 253.0 | 2.1 | 2.3 | 1.10× |
128+
| `distinct_take` | 0.00 | 0.1 | 0.00 | 0.00× |
123129
| `element_at_match` | 0.00 | 0.00 | 0.00 | — |
124130
| `first_match` | 0.00 | 0.00 | 0.00 | — |
125131
| `first_or_default_match` | 0.00 | 0.00 | 0.00 | — |
126-
| `groupby_average` | 172.8 | 2.6 | 2.9 | 1.12× |
127-
| `groupby_count` | 142.0 | 2.4 | 2.5 | 1.04× |
128-
| `groupby_first` | — | 2.2 | 3.1 | 1.41× |
129-
| `groupby_having_count` | 142.4 | 2.3 | 2.5 | 1.09× |
130-
| `groupby_having_hidden_sum` | 176.4 | 2.5 | 2.8 | 1.12× |
131-
| `groupby_having_post_where` | 171.4 | 2.4 | 2.7 | 1.13× |
132-
| `groupby_max` | 175.4 | 2.4 | 2.7 | 1.13× |
133-
| `groupby_min` | 174.0 | 2.4 | 2.7 | 1.13× |
134-
| `groupby_multi_reducer` | 190.5 | 2.7 | 3.0 | 1.11× |
135-
| `groupby_select_order` | 171.9 | 2.4 | 2.7 | 1.13× |
136-
| `groupby_select_sum` | 200.4 | 3.2 | 3.7 | 1.16× |
137-
| `groupby_sum` | 172.1 | 2.4 | 2.7 | 1.13× |
138-
| `groupby_where_count` | 75.8 | 1.5 | 1.8 | 1.20× |
139-
| `groupby_where_sum` | 87.5 | 1.5 | 1.8 | 1.20× |
140-
| `indexed_lookup` | 1249.3 | 35189.7 | 105.7 | 0.00× |
141-
| `join_count` | 38.6 | 34.5 | 13.0 | 0.38× |
142-
| `join_groupby_count` | — | 47.5 | 21.9 | 0.46× |
143-
| `join_groupby_to_array` | — | 56.1 | 21.9 | 0.39× |
144-
| `join_select` | — | 41.7 | 22.7 | 0.54× |
145-
| `join_where_count` | 39.4 | 41.5 | 22.7 | 0.55× |
146-
| `last_match` | 0.00 | 0.50 | 1.4 | 2.80× |
147-
| `long_count_aggregate` | 29.9 | 0.40 | 0.60 | 1.50× |
148-
| `max_aggregate` | 31.3 | 0.60 | 0.50 | 0.83× |
149-
| `min_aggregate` | 31.2 | 0.60 | 0.50 | 0.83× |
150-
| `order_distinct_take` | 138.2 | 2.1 | 74.9 | 35.67× |
151-
| `order_reverse_normalized` | 38.3 | 0.70 | 1.3 | 1.86× |
152-
| `order_take_desc` | 38.4 | 0.70 | 1.3 | 1.86× |
153-
| `reverse_distinct_by` | — | 2.6 | — | — |
132+
| `groupby_average` | 171.9 | 2.6 | 2.9 | 1.12× |
133+
| `groupby_count` | 142.9 | 2.4 | 2.5 | 1.04× |
134+
| `groupby_first` | 252.4 | 2.2 | 3.1 | 1.41× |
135+
| `groupby_having_count` | 142.3 | 2.4 | 2.5 | 1.04× |
136+
| `groupby_having_hidden_sum` | 176.1 | 2.5 | 2.8 | 1.12× |
137+
| `groupby_having_post_where` | 172.4 | 2.4 | 2.7 | 1.13× |
138+
| `groupby_max` | 173.8 | 2.4 | 2.7 | 1.13× |
139+
| `groupby_min` | 181.6 | 2.4 | 2.7 | 1.13× |
140+
| `groupby_multi_reducer` | 190.9 | 2.7 | 3.0 | 1.11× |
141+
| `groupby_select_order` | 171.5 | 2.4 | 2.7 | 1.13× |
142+
| `groupby_select_sum` | 199.7 | 3.2 | 3.7 | 1.16× |
143+
| `groupby_sum` | 172.5 | 2.4 | 2.7 | 1.13× |
144+
| `groupby_where_count` | 76.2 | 1.5 | 1.8 | 1.20× |
145+
| `groupby_where_sum` | 86.9 | 1.5 | 1.8 | 1.20× |
146+
| `indexed_lookup` | 1265.6 | 35161.7 | 107.5 | 0.00× |
147+
| `join_count` | 38.4 | 34.5 | 12.8 | 0.37× |
148+
| `join_groupby_count` | — | 47.1 | 21.9 | 0.46× |
149+
| `join_groupby_to_array` | — | 54.7 | 21.9 | 0.40× |
150+
| `join_select` | — | 41.8 | 22.6 | 0.54× |
151+
| `join_where_count` | 39.5 | 41.5 | 22.9 | 0.55× |
152+
| `last_match` | 0.00 | 0.5 | 1.4 | 2.80× |
153+
| `long_count_aggregate` | 29.5 | 0.4 | 0.6 | 1.50× |
154+
| `max_aggregate` | 31.1 | 0.6 | 0.5 | 0.83× |
155+
| `min_aggregate` | 31.2 | 0.6 | 0.5 | 0.83× |
156+
| `order_distinct_take` | 138.7 | 2.1 | 75.1 | 35.76× |
157+
| `order_reverse_normalized` | 38.3 | 0.7 | 1.3 | 1.86× |
158+
| `order_take_desc` | 38.2 | 0.7 | 1.3 | 1.86× |
159+
| `reverse_distinct_by` | 297.1 | 2.7 | — | — |
154160
| `reverse_take` | 0.00 | 0.00 | 1.1 | — |
155-
| `reverse_take_select` | 0.00 | 8.5 | 10.5 | 1.24× |
156-
| `select_count` | 0.10 | 0.00 | 0.00 | — |
157-
| `select_where` | 108.0 | 4.2 | 5.6 | 1.33× |
158-
| `select_where_count` | 33.1 | 0.40 | 0.60 | 1.50× |
159-
| `select_where_order_take` | 36.7 | 0.70 | 1.3 | 1.86× |
160-
| `select_where_sum` | 37.2 | 0.50 | 0.60 | 1.20× |
161-
| `single_match` | 0.00 | 0.40 | 1.1 | 2.75× |
162-
| `skip_take` | 0.30 | 0.00 | 0.00 | — |
163-
| `skip_while_match` | 3.5 | 0.40 | 0.40 | 1.00× |
164-
| `sort_first` | 38.3 | 0.40 | 1.3 | 3.25× |
165-
| `sort_take` | 38.6 | 0.70 | 1.4 | 2.00× |
166-
| `sort_take_select` | 38.5 | 0.80 | 1.4 | 1.75× |
167-
| `sum_aggregate` | 30.4 | 0.40 | 0.30 | 0.75× |
168-
| `sum_where` | 34.7 | 0.40 | 0.60 | 1.50× |
169-
| `take_count` | 1.8 | 0.10 | 0.10 | 1.00× |
161+
| `reverse_take_select` | 0.00 | 8.7 | 11.2 | 1.29× |
162+
| `select_count` | 0.1 | 0.00 | 0.00 | — |
163+
| `select_where` | 106.1 | 4.3 | 5.7 | 1.33× |
164+
| `select_where_count` | 32.9 | 0.4 | 0.6 | 1.50× |
165+
| `select_where_order_take` | 36.8 | 0.7 | 1.4 | 2.00× |
166+
| `select_where_sum` | 37.6 | 0.5 | 0.6 | 1.20× |
167+
| `single_match` | 0.00 | 0.4 | 1.1 | 2.75× |
168+
| `skip_take` | 0.3 | 0.00 | 0.00 | — |
169+
| `skip_while_match` | 3.5 | 0.4 | 0.4 | 1.00× |
170+
| `sort_first` | 38.2 | 0.4 | 1.3 | 3.25× |
171+
| `sort_take` | 38.3 | 0.7 | 1.4 | 2.00× |
172+
| `sort_take_select` | 38.4 | 0.7 | 1.4 | 2.00× |
173+
| `sum_aggregate` | 30.4 | 0.4 | 0.3 | 0.75× |
174+
| `sum_where` | 32.9 | 0.4 | 0.6 | 1.50× |
175+
| `take_count` | 1.8 | 0.1 | 0.1 | 1.00× |
170176
| `take_count_filtered` | — | 0.00 | 0.00 | — |
171177
| `take_sum_aggregate` | — | 0.00 | 0.00 | — |
172178
| `take_where_count` | — | 0.00 | 0.00 | — |
173-
| `take_while_match` | 7.8 | 0.20 | 0.30 | 1.50× |
174-
| `to_array_filter` | 48.4 | 3.3 | 3.4 | 1.03× |
175-
| `zip_count_pred` | — | 0.70 | — | — |
176-
| `zip_dot_product` | — | 0.50 | 0.50 | 1.00× |
177-
| `zip_dot_product_3arg` | — | 0.50 | — | — |
179+
| `take_while_match` | 7.8 | 0.2 | 0.3 | 1.50× |
180+
| `to_array_filter` | 48.3 | 3.3 | 3.4 | 1.03× |
181+
| `zip_count_pred` | — | 0.7 | — | — |
182+
| `zip_dot_product` | — | 0.5 | 0.5 | 1.00× |
183+
| `zip_dot_product_3arg` | — | 0.5 | — | — |
178184
| `zip_reverse_to_array` | — | 4.5 | — | — |
179185

180-
181186
## Notes on missing lanes (the `—` cells)
182187

183188
The reasons each cell is empty are also recorded as a comment in the
@@ -196,20 +201,6 @@ and which gaps could land in a single PR — see
196201
`forExpr.iteratorVariables` was unpopulated). Array-side bare `count(P)`
197202
was always reachable; SQL `count(P)` is covered by `count_aggregate.das`
198203
with a where shape. By design.
199-
- **`distinct_by_order_take` SQL** — no faithful SQL form. "Pick one row
200-
per dealer, sort by price, take N" needs window functions
201-
(`ROW_NUMBER() OVER (PARTITION BY dealer_id ORDER BY price)` + outer
202-
`WHERE rn=1` + `LIMIT N`); sqlite_linq does not currently lower window
203-
functions. Follow-up TODO 2026-05-25 — see `sqlite_linq_gaps.md`
204-
"Window-function lowerings".
205-
- **`distinct_by_order_to_array` SQL** — same window-function blocker as
206-
`distinct_by_order_take`, drop the `LIMIT N`. See
207-
`sqlite_linq_gaps.md` "Window-function lowerings".
208-
- **`groupby_first` SQL** — no direct SQL aggregator for "first
209-
source-order row per group". Window functions (`ROW_NUMBER() OVER
210-
(PARTITION BY brand ORDER BY id)` + outer `WHERE rn=1`) would be the
211-
SQL equivalent; sqlite_linq does not currently lower window
212-
functions. Follow-up TODO 2026-05-23.
213204
- **`indexed_lookup` Decs vs Array ratio** — the ratio is reported as
214205
`0.00×` because array's lane measures the unspliced linear scan
215206
(~204k ns/op), while decs uses `query(eid)` for O(1) lookup. The
@@ -223,13 +214,10 @@ and which gaps could land in a single PR — see
223214
bare `_` (only the `into` lambda's parameter names are valid
224215
receivers; computed projections would need to be inlined into the
225216
`into` lambda). By design — sqlite_linq surface limitation.
226-
- **`reverse_distinct_by` SQL / Decs** — SQL: same window-function
227-
blocker as `distinct_by_order_take`, with descending key
228-
(`ROW_NUMBER() OVER (PARTITION BY brand ORDER BY id DESC) WHERE rn=1`
229-
picks "last-per-group"); see `sqlite_linq_gaps.md`. Decs: splice arm
230-
is array-only (`array_source` predicate in
231-
`daslib/linq_fold.das:3221`) because backward-index walk has no
232-
archetype analog.
217+
- **`reverse_distinct_by` Decs** — splice arm is array-only
218+
(`array_source` predicate in `daslib/linq_fold.das:3221`) because
219+
backward-index walk has no archetype analog. SQL lane closed by
220+
Arm B of this PR via MAX(pk) bare-aggregate.
233221
- **`order_distinct_take` Decs vs Array ratio** — m4 (95.5 INTERP /
234222
74.9 JIT) vs m3f (15.8 INTERP / 2.1 JIT) is NOT apples-to-apples.
235223
`unique_key` (`daslib/linq.das:614`) short-circuits to a direct

0 commit comments

Comments
 (0)