spg_engine/aggregate.rs
1//! Aggregate executor.
2//!
3//! Handles `SELECT … <aggs> … [GROUP BY …]` queries. The planning strategy
4//! is straightforward:
5//!
6//! 1. Walk the SELECT (and ORDER BY) expressions to find every aggregate
7//! function call. Dedupe by AST equality and assign each `__agg_<i>`.
8//! 2. Same for every `GROUP BY` expression: assign `__grp_<j>`.
9//! 3. Stream the WHERE-filtered rows, group by the tuple of GROUP BY
10//! values, and update per-group aggregate state.
11//! 4. Materialise a synthetic per-group row containing
12//! `[__grp_0..__grp_K, __agg_0..__agg_N]` and rewrite the user's
13//! SELECT / ORDER BY expressions to reference those synthetic columns
14//! instead of the originals.
15//! 5. Evaluate the rewritten expressions against the synthetic schema and
16//! emit results.
17//!
18//! v1.8 implements `count(*)`, `count(expr)`, `sum`, `min`, `max`, `avg`.
19//! NULL semantics follow PG: aggregates skip NULL inputs (except
20//! `count(*)`, which counts rows). `sum(int)` widens to `BigInt`;
21//! `avg(int|bigint)` returns `Float`.
22
23use alloc::borrow::Cow;
24use alloc::boxed::Box;
25use alloc::collections::BTreeSet;
26use alloc::format;
27use alloc::string::{String, ToString};
28use alloc::vec::Vec;
29
30use spg_sql::ast::{Expr, SelectItem, SelectStatement};
31use spg_storage::{ColumnSchema, DataType, Row, Value};
32
33use crate::eval::{self, EvalContext, EvalError};
34use crate::join::RowRef;
35
36/// True if this statement should go through the aggregate path.
37pub fn uses_aggregate(stmt: &SelectStatement) -> bool {
38 if stmt.group_by.is_some() || stmt.having.is_some() {
39 return true;
40 }
41 for item in &stmt.items {
42 if let SelectItem::Expr { expr, .. } = item
43 && contains_aggregate(expr)
44 {
45 return true;
46 }
47 }
48 for o in &stmt.order_by {
49 if contains_aggregate(&o.expr) {
50 return true;
51 }
52 }
53 if let Some(h) = &stmt.having
54 && contains_aggregate(h)
55 {
56 return true;
57 }
58 false
59}
60
61pub fn contains_aggregate(e: &Expr) -> bool {
62 match e {
63 Expr::FunctionCall { name, args } => {
64 is_aggregate_name(name) || args.iter().any(contains_aggregate)
65 }
66 Expr::AggregateOrdered { .. } => true,
67 Expr::Binary { lhs, rhs, .. } => contains_aggregate(lhs) || contains_aggregate(rhs),
68 Expr::Unary { expr, .. } | Expr::Cast { expr, .. } | Expr::IsNull { expr, .. } => {
69 contains_aggregate(expr)
70 }
71 Expr::Like { expr, pattern, .. } => contains_aggregate(expr) || contains_aggregate(pattern),
72 Expr::Extract { source, .. } => contains_aggregate(source),
73 // v4.10 subqueries + v4.12 window functions / Literal /
74 // Column — all non-aggregate leaves from the regular
75 // aggregate planner's POV. Window-bearing projections are
76 // routed to exec_select_with_window before this runs.
77 Expr::ScalarSubquery(_)
78 | Expr::Exists { .. }
79 | Expr::InSubquery { .. }
80 | Expr::WindowFunction { .. }
81 | Expr::Literal(_)
82 | Expr::Placeholder(_)
83 | Expr::Column(_) => false,
84 // v7.10.10 — recurse into array constructor / subscript /
85 // ANY/ALL children. Aggregates inside `ARRAY[SUM(x)]` are
86 // valid PG and must be detected here.
87 Expr::Array(items) => items.iter().any(contains_aggregate),
88 Expr::ArraySubscript { target, index } => {
89 contains_aggregate(target) || contains_aggregate(index)
90 }
91 Expr::AnyAll { expr, array, .. } => contains_aggregate(expr) || contains_aggregate(array),
92 Expr::InList { expr, list, .. } => {
93 contains_aggregate(expr) || list.iter().any(contains_aggregate)
94 }
95 // v7.13.0 — CASE WHEN … END. Recurse into operand,
96 // every (WHEN, THEN) pair, and the ELSE branch.
97 Expr::Case {
98 operand,
99 branches,
100 else_branch,
101 } => {
102 operand.as_deref().is_some_and(contains_aggregate)
103 || branches
104 .iter()
105 .any(|(w, t)| contains_aggregate(w) || contains_aggregate(t))
106 || else_branch.as_deref().is_some_and(contains_aggregate)
107 }
108 }
109}
110
111pub fn is_aggregate_name(name: &str) -> bool {
112 matches!(
113 name.to_ascii_lowercase().as_str(),
114 "count"
115 | "count_star"
116 | "sum"
117 | "min"
118 | "max"
119 | "avg"
120 // v7.17.0 — variadic / collection aggregates. ORM
121 // reports (Hibernate / Rails / Django) emit these in
122 // GROUP BY rollups; pre-7.17 SPG hit "unknown
123 // aggregate".
124 | "string_agg"
125 | "array_agg"
126 // v7.17.0 — boolean aggregates. `every` is SQL-standard
127 // alias for `bool_and`.
128 | "bool_and"
129 | "bool_or"
130 | "every"
131 // v7.32 (round-29) — statistical aggregates (every BI /
132 // dashboard emits these in rollups).
133 | "stddev" | "stddev_samp" | "stddev_pop"
134 | "variance" | "var_samp" | "var_pop"
135 // v7.32 (round-29) — bitwise aggregates.
136 | "bit_and" | "bit_or" | "bit_xor"
137 // v7.32 (round-29) — ordered-set aggregates (used with
138 // `WITHIN GROUP (ORDER BY …)`).
139 | "percentile_cont" | "percentile_disc" | "mode"
140 // v7.32 (round-29) — hypothetical-set aggregates (also
141 // `WITHIN GROUP`): the rank the direct args WOULD have.
142 | "rank" | "dense_rank" | "percent_rank" | "cume_dist"
143 // v7.32 (round-29) — two-argument regression family.
144 | "covar_pop" | "covar_samp" | "corr"
145 | "regr_count" | "regr_avgx" | "regr_avgy" | "regr_slope"
146 | "regr_intercept" | "regr_r2" | "regr_sxx" | "regr_syy" | "regr_sxy"
147 // v7.32 (round-29) — JSON aggregates.
148 | "json_agg" | "jsonb_agg" | "json_object_agg" | "jsonb_object_agg"
149 )
150}
151
152/// v7.32 (round-29) — two-argument regression aggregates `f(Y, X)`.
153fn is_regression_name(name: &str) -> bool {
154 matches!(
155 name,
156 "covar_pop"
157 | "covar_samp"
158 | "corr"
159 | "regr_count"
160 | "regr_avgx"
161 | "regr_avgy"
162 | "regr_slope"
163 | "regr_intercept"
164 | "regr_r2"
165 | "regr_sxx"
166 | "regr_syy"
167 | "regr_sxy"
168 )
169}
170
171/// v7.32 (round-29) — aggregates that consume a second positional
172/// argument: `string_agg(v, sep)`, the regression family `f(Y, X)`, and
173/// `json_object_agg(key, value)`.
174fn agg_uses_second_arg(name: &str) -> bool {
175 name == "string_agg"
176 || name == "json_object_agg"
177 || name == "jsonb_object_agg"
178 || is_regression_name(name)
179}
180
181/// v7.32 (round-29) — ordered-set aggregates: the value to aggregate
182/// comes from the `WITHIN GROUP (ORDER BY …)` sort spec, and any
183/// in-parens arguments are *direct* arguments (the percentile fraction).
184/// `mode()` takes no direct argument.
185pub fn is_ordered_set_name(name: &str) -> bool {
186 // v7.32 — `eq_ignore_ascii_case` instead of `to_ascii_lowercase()`:
187 // these classifiers run in the aggregate row/group loop, where the
188 // old per-call `String` allocation showed up as ~16% of the inbox's
189 // aggregate path in a sampled profile (the names are constant).
190 ["percentile_cont", "percentile_disc", "mode"]
191 .iter()
192 .any(|k| name.eq_ignore_ascii_case(k))
193}
194
195/// v7.32 (round-29) — hypothetical-set aggregates: `rank(args) WITHIN
196/// GROUP (ORDER BY …)` and friends compute the rank the hypothetical
197/// row would have. Like ordered-set, the value stream comes from the
198/// sort spec and the in-parens args are direct (the hypothetical row).
199pub fn is_hypothetical_set_name(name: &str) -> bool {
200 ["rank", "dense_rank", "percent_rank", "cume_dist"]
201 .iter()
202 .any(|k| name.eq_ignore_ascii_case(k))
203}
204
205/// v7.32 (round-29) — every aggregate that takes its value stream from
206/// a `WITHIN GROUP (ORDER BY …)` clause (ordered-set + hypothetical-set).
207pub fn is_within_group_name(name: &str) -> bool {
208 is_ordered_set_name(name) || is_hypothetical_set_name(name)
209}
210
211/// Per-aggregate running state.
212#[derive(Debug, Default, Clone)]
213struct AggState {
214 count: i64,
215 sum_int: i64,
216 sum_float: f64,
217 extreme: Option<Value>,
218 use_float: bool,
219 /// v7.17.0 — running collection for string_agg / array_agg.
220 /// Each entry is one row's contribution (NULL preserved as
221 /// `Value::Null`; string_agg's finalize step drops them, but
222 /// array_agg keeps them). Pushing in insertion order matches
223 /// PG behaviour when no `ORDER BY` is given inside the
224 /// aggregate call.
225 items: Vec<Value>,
226 /// v7.25 (round-17) — per-group dedupe set for DISTINCT
227 /// aggregates (encoded values; NULLs never reach it because
228 /// the caller's skip runs after the per-aggregate NULL rules).
229 seen: BTreeSet<String>,
230 /// v7.24 (round-16 A) — per-item ORDER BY key tuples, parallel
231 /// to `items` (pushed under the same skip/keep conditions).
232 /// Empty when the aggregate carries no internal ordering.
233 item_keys: Vec<Vec<Value>>,
234 /// v7.17.0 — captured separator for string_agg. PG accepts a
235 /// non-constant separator expression but in practice every
236 /// caller passes a literal; the engine snapshots the last
237 /// non-NULL text it sees, which matches PG's "use the latest
238 /// row's value" behaviour.
239 separator: Option<String>,
240 /// v7.17.0 — running boolean accumulator for bool_and /
241 /// bool_or / every. `None` until the first non-NULL input;
242 /// at finalize None → SQL NULL.
243 bool_acc: Option<bool>,
244 /// v7.32 (round-29) — sum of squares for the variance / stddev
245 /// family (`sum_float` carries the running sum; `count` the n).
246 sum_sq: f64,
247 /// v7.32 (round-29) — running accumulator for bit_and / bit_or /
248 /// bit_xor. `None` until the first non-NULL input → SQL NULL.
249 bit_acc: Option<i64>,
250 /// v7.32 (round-29) — two-argument regression family
251 /// (`covar_*` / `corr` / `regr_*`), PG arg order `f(Y, X)`. Only
252 /// rows where BOTH inputs are non-NULL contribute (`count` is the
253 /// paired n, independent of the single-arg `sum_*`).
254 reg_n: i64,
255 reg_sx: f64,
256 reg_sy: f64,
257 reg_sxx: f64,
258 reg_syy: f64,
259 reg_sxy: f64,
260 /// v7.32 (round-29) — second value stream for `json_object_agg`
261 /// (`items` holds the keys, `aux_items` the values).
262 aux_items: Vec<Value>,
263 /// v7.33 (array_agg argmax) — for a `first_ordered` spec
264 /// (`(array_agg(x ORDER BY y))[1]`), the running first-by-order
265 /// (sort-key tuple, value). Replaced only when a new row's key sorts
266 /// strictly before the current best (ties keep the earliest row, =
267 /// the stable-sort `[1]`). No items/item_keys array is built.
268 first_best: Option<(Vec<Value>, Value)>,
269}
270
271#[derive(Debug, Clone)]
272struct AggSpec {
273 name: String, // lowercased
274 /// First argument (value expression) for every aggregate
275 /// except `count(*)`. `None` for `count_star`.
276 arg: Option<Expr>,
277 /// v7.17.0 — second argument. Only `string_agg(value, sep)`
278 /// uses it today. `None` for every other aggregate (or for
279 /// `array_agg`, which is single-arg). Carried in the spec so
280 /// per-row evaluation can re-use the same separator
281 /// expression across calls.
282 arg2: Option<Expr>,
283 /// v7.25 (round-17) — `COUNT(DISTINCT x)` & friends: dedupe
284 /// the input stream per group before accumulation.
285 distinct: bool,
286 /// v7.24 (round-16 A) — aggregate-internal ORDER BY keys
287 /// (`array_agg(x ORDER BY y DESC NULLS LAST)`). Empty for the
288 /// plain form. Only the collection aggregates honour it;
289 /// other aggregates are order-insensitive and ignore it (PG
290 /// accepts the syntax everywhere too).
291 order_by: Vec<spg_sql::ast::OrderBy>,
292 /// v7.32 (round-29) — `FILTER (WHERE cond)`: a per-row predicate
293 /// evaluated against the source row before accumulation. A row
294 /// whose `cond` is not TRUE (false or NULL) is excluded from this
295 /// aggregate only. `None` for the unfiltered form.
296 filter: Option<Expr>,
297 /// v7.32 (round-29) — ordered-set aggregates only: the *direct*
298 /// argument (the percentile fraction for `percentile_cont/disc`).
299 /// PG requires it constant, so it is evaluated once. `None` for
300 /// `mode()` and for every non-ordered-set aggregate.
301 direct_arg: Option<Expr>,
302 /// v7.33 (array_agg argmax) — set when this spec came from
303 /// `(array_agg(x ORDER BY y))[1]`: accumulate only the first-by-order
304 /// element (a running argmax/argmin) and finalise to that scalar
305 /// value, instead of collecting + sorting + materialising the whole
306 /// per-group array just to take element 1. Returns the element type,
307 /// not the array type.
308 first_ordered: bool,
309}
310
311/// Output of running the aggregate path. Schema describes one row per
312/// group; rows are not yet ORDER BY-sorted (caller does it).
313#[derive(Debug)]
314pub struct AggResult {
315 pub columns: Vec<ColumnSchema>,
316 pub rows: Vec<Row>,
317 /// v7.31 (perf — PG lesson #1, post-LIMIT subquery projection):
318 /// select-list items whose rewritten expr carries a subquery and
319 /// is referenced by neither ORDER BY nor HAVING. Their output
320 /// cells hold NULL placeholders; the caller truncates to
321 /// LIMIT+OFFSET first and only then evaluates these for the
322 /// surviving rows (PG runs the same shape with SubPlan loops=50
323 /// instead of loops=24000). `(output_col, rewritten_expr)`.
324 pub deferred: Vec<(usize, Expr)>,
325 /// Synthetic group rows aligned 1:1 with `rows`; populated only
326 /// when `deferred` is non-empty.
327 pub synth_rows: Vec<Row>,
328 /// Schema the deferred exprs evaluate against.
329 pub synth_schema: Vec<ColumnSchema>,
330}
331
332/// Execute aggregate logic against an already-WHERE-filtered iterator of
333/// rows. `table_alias` is the alias accepted by column resolution.
334#[allow(clippy::too_many_lines)]
335/// v7.25.2 (round-19 A) — caller-injected evaluator for synth-row
336/// expressions that still carry subquery nodes after the rewrite
337/// (correlated subqueries in the select list / HAVING / aggregate
338/// ORDER BY of a GROUP BY query). The engine passes its
339/// correlated-aware evaluator; pure-library callers pass None and
340/// surviving subqueries keep erroring loudly.
341pub type CorrelatedEval<'a> = &'a dyn Fn(&Expr, &Row, &EvalContext<'_>) -> Result<Value, EvalError>;
342
343/// Output of the per-group projection stage (`project_groups`): the
344/// output schema, the projected rows, the synth rows kept alongside
345/// them for post-LIMIT deferred evaluation, the deferred subquery
346/// items, and the rewritten ORDER BY exprs (shared with the sort).
347struct Projection {
348 columns: Vec<ColumnSchema>,
349 out_rows: Vec<Row>,
350 kept_synth: Vec<Row>,
351 deferred: Vec<(usize, Expr)>,
352 order_rewritten: Vec<Expr>,
353}
354
355/// v7.35.0 — detect the `SELECT COUNT(*) FROM … [WHERE …]` shape
356/// (single item, no GROUP BY / HAVING / ORDER BY / DISTINCT /
357/// LIMIT WITH TIES / FILTER / window). For this shape the answer
358/// is exactly `rows.len()` as `BigInt`, no group state needed.
359/// Returns `None` for any deviation so the caller's full pipeline
360/// runs verbatim.
361///
362/// v7.35.2 — also short-circuit `COUNT(<literal>)` (e.g.
363/// `COUNT(1)`) and `COUNT(<column>)` when the column is declared
364/// NOT NULL on the input schema. PG handles both cases as
365/// `COUNT(*)` (the non-null filter is a no-op), so doing the same
366/// here keeps every `count this thing` shape on the same fast path
367/// instead of routing the literal / non-null-col variants through
368/// the four-stage aggregate pipeline.
369fn try_pure_count_star_short_circuit(
370 stmt: &SelectStatement,
371 rows: &[RowRef<'_>],
372 schema_cols: &[ColumnSchema],
373 table_alias: Option<&str>,
374) -> Option<AggResult> {
375 if stmt.distinct
376 || stmt.limit_with_ties
377 || stmt.group_by.is_some()
378 || stmt.having.is_some()
379 || !stmt.order_by.is_empty()
380 {
381 return None;
382 }
383 if stmt.items.len() != 1 {
384 return None;
385 }
386 let SelectItem::Expr { expr, alias } = &stmt.items[0] else {
387 return None;
388 };
389 let Expr::FunctionCall { name, args } = expr else {
390 return None;
391 };
392 if !name.eq_ignore_ascii_case("count") && !name.eq_ignore_ascii_case("count_star") {
393 return None;
394 }
395 let count_star_shape = match args.as_slice() {
396 // `COUNT(*)` parses to `count_star` with no args.
397 [] if name.eq_ignore_ascii_case("count_star") => true,
398 // `COUNT(<literal>)` — the per-row test is "is this literal
399 // non-null?" which is constant, so it's COUNT(*) when the
400 // literal is non-null.
401 [Expr::Literal(lit)] => !matches!(lit, spg_sql::ast::Literal::Null),
402 // `COUNT(<column>)` — same answer as COUNT(*) when the
403 // column is statically declared NOT NULL on the input
404 // schema. Resolve through the alias if one is set.
405 [Expr::Column(c)] => {
406 if let Some(q) = c.qualifier.as_deref()
407 && let Some(alias) = table_alias
408 && !q.eq_ignore_ascii_case(alias)
409 {
410 return None;
411 }
412 schema_cols
413 .iter()
414 .find(|s| s.name.eq_ignore_ascii_case(&c.name))
415 .is_some_and(|s| !s.nullable)
416 }
417 _ => return None,
418 };
419 if !count_star_shape {
420 return None;
421 }
422 let col_name = alias.clone().unwrap_or_else(|| "count".to_string());
423 let count = i64::try_from(rows.len()).unwrap_or(i64::MAX);
424 Some(AggResult {
425 columns: alloc::vec![ColumnSchema::new(col_name, DataType::BigInt, false)],
426 rows: alloc::vec![Row::new(alloc::vec![Value::BigInt(count)])],
427 deferred: Vec::new(),
428 synth_rows: Vec::new(),
429 synth_schema: Vec::new(),
430 })
431}
432
433pub(crate) fn run(
434 stmt: &SelectStatement,
435 rows: &[RowRef<'_>],
436 schema_cols: &[ColumnSchema],
437 table_alias: Option<&str>,
438 correlated_eval: Option<CorrelatedEval<'_>>,
439) -> Result<AggResult, EvalError> {
440 // v7.35.0 — pure `SELECT COUNT(*) FROM … WHERE …` short-circuit.
441 // The caller already filtered rows by WHERE (we run on the
442 // post-WHERE survivor set), so for the canonical pure-COUNT(*)
443 // shape (no GROUP BY / HAVING / ORDER BY / DISTINCT / FILTER /
444 // window) the answer is simply `rows.len()`. The four-stage
445 // aggregate pipeline below (accumulate_groups → build_synth_schema
446 // → finalize_synth_rows → project_groups) collapses to a single
447 // BigInt cell when there's a single group, but each stage still
448 // pays its own allocation tax — group state map, synth schema
449 // vec, finalize loop. `exists_in_60` (mailrs prod #4 baseline)
450 // is exactly this shape on a 25 k-row JOIN.
451 if let Some(short) = try_pure_count_star_short_circuit(stmt, rows, schema_cols, table_alias) {
452 return Ok(short);
453 }
454 let group_exprs: Vec<Expr> = stmt.group_by.clone().unwrap_or_default();
455
456 // Collect aggregate sub-expressions across items + order_by.
457 let mut agg_specs: Vec<AggSpec> = Vec::new();
458 for item in &stmt.items {
459 if let SelectItem::Expr { expr, .. } = item {
460 collect_aggregates(expr, &mut agg_specs);
461 }
462 }
463 for o in &stmt.order_by {
464 collect_aggregates(&o.expr, &mut agg_specs);
465 }
466 if let Some(h) = &stmt.having {
467 collect_aggregates(h, &mut agg_specs);
468 }
469 // v7.17.0 — arity validation. The collector tolerates an
470 // arbitrary positional-arg count; here we enforce the
471 // per-aggregate contract so a malformed call (e.g.
472 // `array_agg()` or `string_agg(x)`) surfaces as a SQL error
473 // rather than silently coercing to a degenerate aggregate.
474 validate_agg_arities(stmt, &agg_specs)?;
475 validate_within_group(&agg_specs)?;
476
477 // (1) Stream the WHERE-filtered rows into insertion-ordered group state.
478 let order = accumulate_groups(
479 rows,
480 &group_exprs,
481 &agg_specs,
482 schema_cols,
483 table_alias,
484 correlated_eval,
485 )?;
486
487 // (2) Build the synthetic per-group schema and finalise each group's row.
488 let synth_schema =
489 build_synth_schema(rows, &group_exprs, &agg_specs, schema_cols, table_alias)?;
490 let synth_rows = finalize_synth_rows(
491 &order,
492 &agg_specs,
493 &synth_schema,
494 rows,
495 schema_cols,
496 table_alias,
497 )?;
498
499 // (3) Rewrite the user's expressions, filter groups by HAVING and project.
500 let Projection {
501 columns,
502 mut out_rows,
503 mut kept_synth,
504 deferred,
505 order_rewritten,
506 } = project_groups(
507 synth_rows,
508 stmt,
509 &group_exprs,
510 &agg_specs,
511 &synth_schema,
512 correlated_eval,
513 )?;
514
515 // (4) ORDER BY on the aggregated output (the caller applies LIMIT).
516 //
517 // v7.37.3 (mailrs prod /api/contacts 3.21× regression — and the
518 // general inbox-listing-shape SPG-vs-PG gap) — top-K sink for
519 // `ORDER BY <agg> [DESC] LIMIT k`. Pre-7.37.3 this stage ran a
520 // full O(N log N) sort over every surviving group, then the
521 // caller truncated to `k`. With high-cardinality GROUP BY (a
522 // sender column with hundreds-thousands of distinct values) the
523 // truncated set is a tiny fraction of `N` — keep an O(k) top-K
524 // sink and never sort the discarded majority. Matches PG /
525 // MySQL / MariaDB's standard "LIMIT k under ORDER BY agg"
526 // optimisation; SPG previously implemented it only on the
527 // streamed inner-join path (`try_streamed_inner_join_topn`)
528 // and not on the aggregate output.
529 //
530 // Gate: needs a literal LIMIT (placeholder LIMIT we can't bound
531 // statically here), no DISTINCT (would need post-dedup, can't
532 // truncate during sort), no LIMIT WITH TIES (which extends past
533 // the literal k by run-time tie-key comparison).
534 let keep_n: Option<usize> = if !stmt.order_by.is_empty()
535 && !stmt.distinct
536 && !stmt.limit_with_ties
537 {
538 stmt.limit_literal().map(|l| {
539 let off = stmt.offset_literal().unwrap_or(0) as usize;
540 (l as usize).saturating_add(off)
541 })
542 } else {
543 None
544 };
545 if !stmt.order_by.is_empty() {
546 let (sorted_synth, sorted_out) = sort_synth_by_order_by(
547 &synth_schema,
548 &stmt.order_by,
549 &order_rewritten,
550 kept_synth,
551 out_rows,
552 correlated_eval,
553 keep_n,
554 )?;
555 kept_synth = sorted_synth;
556 out_rows = sorted_out;
557 }
558
559 let (synth_rows_out, synth_schema_out) = if deferred.is_empty() {
560 (Vec::new(), Vec::new())
561 } else {
562 (kept_synth, synth_schema.clone())
563 };
564 Ok(AggResult {
565 columns,
566 rows: out_rows,
567 deferred,
568 synth_rows: synth_rows_out,
569 synth_schema: synth_schema_out,
570 })
571}
572
573/// v7.32 (round-29) — validate the structural requirements of WITHIN
574/// GROUP (ordered-set / hypothetical-set) aggregates up front, so a
575/// malformed call surfaces as a SQL error rather than a silently
576/// degenerate aggregate.
577fn validate_within_group(agg_specs: &[AggSpec]) -> Result<(), EvalError> {
578 // v7.32 (round-29) — WITHIN GROUP aggregates require the clause (PG
579 // raises a hard error otherwise rather than silently degrading), and
580 // SPG supports the single-sort-key form only.
581 for spec in agg_specs {
582 if is_within_group_name(&spec.name) {
583 if spec.order_by.is_empty() {
584 return Err(EvalError::TypeMismatch {
585 detail: format!("{}() requires WITHIN GROUP (ORDER BY …)", spec.name),
586 });
587 }
588 // mode() is the only WITHIN GROUP aggregate with no direct
589 // argument; the rest carry one (percentile fraction /
590 // hypothetical value).
591 if spec.name != "mode" && spec.direct_arg.is_none() {
592 return Err(EvalError::TypeMismatch {
593 detail: format!("{}() requires a direct argument", spec.name),
594 });
595 }
596 // Multi-key WITHIN GROUP (multiple sort keys / hypothetical
597 // args) is not supported yet — error loudly instead of
598 // silently using only the first key.
599 if spec.order_by.len() > 1 {
600 return Err(EvalError::TypeMismatch {
601 detail: format!(
602 "{}() with multiple WITHIN GROUP sort keys is not supported yet",
603 spec.name
604 ),
605 });
606 }
607 }
608 }
609 Ok(())
610}
611
612/// (1) Stream the WHERE-filtered rows, group by the GROUP BY value
613/// tuple, and update per-group aggregate state. Returns the groups in
614/// insertion order. See `run` for the bind-once fast path rationale.
615#[allow(clippy::too_many_lines, clippy::type_complexity)]
616fn accumulate_groups(
617 rows: &[RowRef<'_>],
618 group_exprs: &[Expr],
619 agg_specs: &[AggSpec],
620 schema_cols: &[ColumnSchema],
621 table_alias: Option<&str>,
622 correlated_eval: Option<CorrelatedEval<'_>>,
623) -> Result<Vec<(Vec<Value>, Vec<AggState>)>, EvalError> {
624 let ctx = EvalContext::new(schema_cols, table_alias);
625 // Map group key (vec of values, encoded as canonical string) -> group state.
626 // v7.32 (architecture v2, P2b) — insertion-ordered group state in
627 // a Vec; the hash map only maps key → index. Removes the parallel
628 // `key_order: Vec<String>` (a second per-group key clone) and the
629 // per-group re-probe `groups[k]` at finalize (24k hash lookups for
630 // the inbox shape). The map owns its key once on vacant insert.
631 let mut order: Vec<(Vec<Value>, Vec<AggState>)> = Vec::new();
632 let mut groups: hashbrown::HashMap<String, usize> = hashbrown::HashMap::new();
633 // When there are no GROUP BY exprs *and* there is at least one aggregate,
634 // every row collapses into a single anonymous group keyed by "".
635 if rows.is_empty() && group_exprs.is_empty() {
636 // Single empty-aggregate group: count=0, sum=0, max=NULL, etc.
637 // No rows follow, so the map is never probed — seed `order` only.
638 let init: Vec<AggState> = (0..agg_specs.len()).map(|_| AggState::default()).collect();
639 order.push((Vec::new(), init));
640 }
641
642 // v7.30 (perf campaign) - hoist the per-row work that doesn't
643 // depend on the row: which group exprs need collation folding
644 // (none, for most queries - the old code cloned the whole
645 // group_vals vec per row just in case).
646 // v7.30 (perf campaign) - the no-tax row loop. When a group
647 // expr or an aggregate argument is a bare column reference
648 // (the overwhelmingly common shape), bind its position ONCE
649 // and read row cells by offset in the loop - no per-row tree
650 // walk, no owned-Value clone out of resolve_column. Anything
651 // more complex keeps the eval path.
652 let col_pos = |e: &Expr| -> Option<usize> {
653 // Qualified references only: the bare-name resolver carries
654 // alias/ambiguity logic the bind-once path must not fork.
655 if let Expr::Column(c) = e
656 && c.qualifier.is_some()
657 {
658 eval::find_column_pos(c, &ctx)
659 } else {
660 None
661 }
662 };
663 let group_pos: Vec<Option<usize>> = group_exprs.iter().map(col_pos).collect();
664 let all_groups_bound = group_pos.iter().all(Option::is_some);
665 let arg_pos: Vec<Option<usize>> = agg_specs
666 .iter()
667 .map(|spec| spec.arg.as_ref().and_then(|e| col_pos(e)))
668 .collect();
669 // v7.36 (perf — mailrs Ask 1 SUM(LENGTH(text_body)) 18ms → ?) —
670 // pre-compile every aggregate arg that's a `fully_compilable`
671 // PURE expression over bound columns. Without this, `LENGTH(col)`
672 // / `COALESCE(col, '')` / `CAST(col AS BIGINT)` etc. ALL fell
673 // through to the `(None, Some(e)) => eval_arg(e, mat, ...)` slow
674 // path that materialises a Cow<Row> per input row — for a 25k-row
675 // JOIN that's 25k full-row clones for one column read. The Step
676 // VM (`eval_compiled_ref`) reads columns by RowRef::get and runs
677 // the same `apply_function` dispatcher with zero materialisation.
678 let arg_compiled: Vec<Option<eval::CompiledExpr>> = agg_specs
679 .iter()
680 .enumerate()
681 .map(|(i, spec)| match (&arg_pos[i], &spec.arg) {
682 (Some(_), _) => None,
683 (None, Some(e)) if eval::fully_compilable(e) => Some(eval::compile_expr(e, &ctx)),
684 _ => None,
685 })
686 .collect();
687 // v7.33 (array_agg perf) — bound positions for each spec's internal
688 // ORDER BY keys, so an ordered aggregate (`array_agg(x ORDER BY y)`)
689 // reads the sort key by reference (RowRef::get) instead of
690 // materialising the whole combined join row per input row just to
691 // eval one bound column. Mirrors arg_pos. On the inbox shape this
692 // turned 24k full-row (~1 KB each) clones into 24k single-cell reads.
693 let order_pos: Vec<Vec<Option<usize>>> = agg_specs
694 .iter()
695 .map(|spec| spec.order_by.iter().map(|o| col_pos(&o.expr)).collect())
696 .collect();
697 // Does any spec need the fully-materialised row in the bound fast
698 // path — a FILTER, a non-bound value arg, a second arg, or a non-bound
699 // ORDER key? When false (every aggregate arg/key is a bound column —
700 // the inbox shape) the bound fast path never materialises a row.
701 let needs_mat = agg_specs.iter().enumerate().any(|(i, s)| {
702 s.filter.is_some()
703 || (s.arg.is_some() && arg_pos[i].is_none() && arg_compiled[i].is_none())
704 || s.arg2.is_some()
705 || order_pos[i].iter().any(Option::is_none)
706 });
707 let ci_positions: Vec<usize> = group_exprs
708 .iter()
709 .enumerate()
710 .filter(|(_, g)| {
711 matches!(
712 eval::column_collation(g, &ctx),
713 Some(spg_storage::Collation::CaseInsensitive)
714 )
715 })
716 .map(|(i, _)| i)
717 .collect();
718 // v7.31 (perf 3e) — per-row scratch buffers. The fast path used
719 // to allocate a key String (and a refs Vec) for EVERY row just
720 // to probe the group map; hits — the overwhelming case — now
721 // touch the allocator zero times.
722 let mut keybuf_s = String::new();
723 // v7.36 — reused Step VM eval stack for compiled aggregate args.
724 let mut eval_stack: Vec<Value> = Vec::new();
725 let mut dkeybuf = String::new();
726 let mut refs: Vec<&Value> = Vec::with_capacity(group_pos.len());
727 // v7.32 (round-31) — an aggregate's argument / FILTER / second arg /
728 // ORDER key may itself be a *correlated* subquery, e.g.
729 // `MAX((SELECT i.v FROM inner i WHERE i.fk = o.id))`. A non-correlated
730 // subquery is pre-resolved to a literal before this loop, but a
731 // correlated one survives as a subquery node and must be evaluated per
732 // outer row through the correlated evaluator — the same hook the
733 // select-list / HAVING / ORDER finalisers already use below. Plain
734 // `eval_expr` would hit "subquery reached row eval".
735 //
736 // The `any_agg_subquery` gate is computed once here so the common case
737 // (no subquery anywhere in the aggregate args — including every hot
738 // scan/group aggregate) short-circuits before the per-row
739 // `expr_has_subquery` walk: `eval_arg` is then exactly `eval_expr`.
740 let any_agg_subquery = correlated_eval.is_some()
741 && agg_specs.iter().any(|s| {
742 s.filter
743 .as_ref()
744 .is_some_and(|e| crate::expr_has_subquery(e))
745 || s.arg.as_ref().is_some_and(|e| crate::expr_has_subquery(e))
746 || s.arg2.as_ref().is_some_and(|e| crate::expr_has_subquery(e))
747 || s.order_by.iter().any(|o| crate::expr_has_subquery(&o.expr))
748 });
749 let eval_arg = |e: &Expr, r: &Row, c: &EvalContext<'_>| -> Result<Value, EvalError> {
750 match correlated_eval {
751 Some(f) if any_agg_subquery && crate::expr_has_subquery(e) => f(e, r, c),
752 _ => eval::eval_expr(e, r, c),
753 }
754 };
755 // v7.36 (perf — mailrs Phase 1, post u64-hash) — single
756 // anonymous group fast path. When the query has no GROUP BY
757 // (`SELECT SUM(LENGTH(col)) FROM ...`, COUNT, AVG, etc.) the
758 // whole input collapses into one group. The fast path below
759 // still pays one `groups.get("")` hash probe per row plus
760 // `entry = &mut order[0]` reindex even when the empty-key
761 // path encodes nothing — measured ~50 ns/row across 25 k rows
762 // = ~1.25 ms of pure bookkeeping on the user_storage_usage
763 // baseline.
764 //
765 // Bypass: lift `entry` outside the loop and feed every row
766 // straight into it. Same `update_state` machinery, zero
767 // per-row hash work, zero per-row index lookup.
768 let single_anon_group = group_exprs.is_empty() && !rows.is_empty();
769 if single_anon_group {
770 // Seed the single group at idx 0 once.
771 let init: Vec<AggState> = (0..agg_specs.len()).map(|_| AggState::default()).collect();
772 order.clear();
773 order.push((Vec::new(), init));
774 }
775 // v7.36 (perf — mailrs Phase 1, count_messages 2.58 → ?) —
776 // `COUNT(*)` short-circuit. For a single-anon-group `COUNT(*)`
777 // with no FILTER / DISTINCT, every survivor counts once — the
778 // answer IS `rows.len()`. Skips the 25 k iterations of
779 // `update_state("count_star", …)` on the mailrs count_messages
780 // shape; the JOIN already produced exactly the set of rows
781 // that must be counted.
782 if single_anon_group
783 && agg_specs.len() == 1
784 && agg_specs[0].name == "count_star"
785 && agg_specs[0].filter.is_none()
786 && agg_specs[0].arg.is_none()
787 && agg_specs[0].arg2.is_none()
788 && agg_specs[0].order_by.is_empty()
789 && !agg_specs[0].distinct
790 {
791 let state = &mut order[0].1[0];
792 state.count = rows.len() as i64;
793 return Ok(order);
794 }
795 // v7.36 (perf — mailrs Phase 1) — `COUNT(<bound col>)` (non-`*`)
796 // collapses to: read the cell, increment when not NULL. Skips
797 // the per-row spec dispatch + `update_state("count", …)`.
798 if single_anon_group
799 && agg_specs.len() == 1
800 && agg_specs[0].name == "count"
801 && agg_specs[0].filter.is_none()
802 && agg_specs[0].arg2.is_none()
803 && agg_specs[0].order_by.is_empty()
804 && !agg_specs[0].distinct
805 && arg_pos[0].is_some()
806 {
807 let p = arg_pos[0].unwrap();
808 let mut count: i64 = 0;
809 for row in rows {
810 if !matches!(row.get(p), Some(Value::Null) | None) {
811 count += 1;
812 }
813 }
814 let state = &mut order[0].1[0];
815 state.count = count;
816 return Ok(order);
817 }
818 // v7.36 (perf — mailrs Phase 1, user_storage_usage 7.5 → ?) —
819 // single-aggregate streaming accumulator. For
820 // `SUM(<compiled-expr>)` / `SUM(<bound col>)` with no GROUP BY,
821 // no FILTER, no arg2, no ORDER BY, no DISTINCT, the whole
822 // per-row work collapses to: eval the arg, match the Value
823 // variant, accumulate. Skips the spec-dispatch loop +
824 // `update_state` per-row name match. On a 25 k-row JOIN
825 // (user_storage_usage `SUM(LENGTH(text_body))`) that's
826 // ~50-100 ns/row of pure spec-dispatch overhead removed.
827 if single_anon_group
828 && agg_specs.len() == 1
829 && agg_specs[0].filter.is_none()
830 && agg_specs[0].arg2.is_none()
831 && agg_specs[0].order_by.is_empty()
832 && !agg_specs[0].distinct
833 && (agg_specs[0].name == "sum" || agg_specs[0].name == "avg")
834 && (arg_pos[0].is_some() || arg_compiled[0].is_some())
835 {
836 let arg_pos0 = arg_pos[0];
837 let arg_c0 = &arg_compiled[0];
838 let mut sum_int: i64 = 0;
839 let mut sum_float: f64 = 0.0;
840 let mut use_float = false;
841 let mut count: i64 = 0;
842 // Borrow-aware fast inner: avoid the per-row clone when arg
843 // is a bound column position.
844 if let Some(p) = arg_pos0 {
845 for row in rows {
846 let v_ref = row.get(p).unwrap_or(&Value::Null);
847 match v_ref {
848 Value::Null => continue,
849 Value::SmallInt(n) => {
850 sum_int += i64::from(*n);
851 count += 1;
852 }
853 Value::Int(n) => {
854 sum_int += i64::from(*n);
855 count += 1;
856 }
857 Value::BigInt(n) => {
858 sum_int += *n;
859 count += 1;
860 }
861 Value::Float(x) => {
862 sum_float += *x;
863 use_float = true;
864 count += 1;
865 }
866 other => {
867 return Err(EvalError::TypeMismatch {
868 detail: format!("sum/avg need numeric, got {:?}", other.data_type()),
869 });
870 }
871 }
872 }
873 } else if let Some(p) = arg_c0.as_ref().and_then(|c| c.as_single_column_length()) {
874 // v7.36 (perf — mailrs Phase 1, user_storage_usage hot
875 // inner) — `SUM(LENGTH(<text col>))` collapses to a
876 // straight scan: read the cell by ref, branch on the
877 // variant, do an ASCII probe + `len()` (or
878 // `chars().count()` on non-ASCII), accumulate. No Step
879 // VM, no stack push/pop, no `BigInt` boxing on the way
880 // out — pure i64 sum. The original Step VM path keeps
881 // running for everything outside this shape (`SUM(col)`,
882 // `SUM(expr)`, multi-step compiled args).
883 for row in rows {
884 let Some(v_ref) = row.get(p) else {
885 continue;
886 };
887 let n = match v_ref {
888 Value::Null => continue,
889 Value::Text(s) => {
890 if s.is_ascii() {
891 s.len() as i64
892 } else {
893 s.chars().count() as i64
894 }
895 }
896 other => {
897 return Err(EvalError::TypeMismatch {
898 detail: format!("length() needs text, got {:?}", other.data_type()),
899 });
900 }
901 };
902 sum_int += n;
903 count += 1;
904 }
905 } else {
906 let c = arg_c0.as_ref().unwrap();
907 for row in rows {
908 let v = eval::eval_compiled_ref(c, row, &ctx, &mut eval_stack)?;
909 match v {
910 Value::Null => continue,
911 Value::SmallInt(n) => {
912 sum_int += i64::from(n);
913 count += 1;
914 }
915 Value::Int(n) => {
916 sum_int += i64::from(n);
917 count += 1;
918 }
919 Value::BigInt(n) => {
920 sum_int += n;
921 count += 1;
922 }
923 Value::Float(x) => {
924 sum_float += x;
925 use_float = true;
926 count += 1;
927 }
928 other => {
929 return Err(EvalError::TypeMismatch {
930 detail: format!("sum/avg need numeric, got {:?}", other.data_type()),
931 });
932 }
933 }
934 }
935 }
936 let state = &mut order[0].1[0];
937 state.count = count;
938 state.sum_int = sum_int;
939 state.sum_float = sum_float;
940 state.use_float = use_float;
941 return Ok(order);
942 }
943 for row in rows {
944 if single_anon_group {
945 let entry = &mut order[0];
946 let mat: Option<Cow<'_, Row>> = if needs_mat { Some(row.as_row()) } else { None };
947 for (i, spec) in agg_specs.iter().enumerate() {
948 if let Some(f) = &spec.filter
949 && !matches!(
950 eval_arg(f, mat.as_deref().expect("needs_mat for FILTER"), &ctx)?,
951 Value::Bool(true)
952 )
953 {
954 continue;
955 }
956 let arg_owned: Value;
957 let arg_ref: &Value = match (&arg_pos[i], &arg_compiled[i], &spec.arg) {
958 (Some(p), _, _) => row.get(*p).unwrap_or(&Value::Null),
959 (None, _, None) => {
960 arg_owned = Value::Bool(true);
961 &arg_owned
962 }
963 (None, Some(c), _) => {
964 arg_owned = eval::eval_compiled_ref(c, row, &ctx, &mut eval_stack)?;
965 &arg_owned
966 }
967 (None, None, Some(e)) => {
968 arg_owned = eval_arg(
969 e,
970 mat.as_deref().expect("needs_mat for non-bound arg"),
971 &ctx,
972 )?;
973 &arg_owned
974 }
975 };
976 let arg2_val = match &spec.arg2 {
977 None => None,
978 Some(e) => Some(eval_arg(
979 e,
980 mat.as_deref().expect("needs_mat for arg2"),
981 &ctx,
982 )?),
983 };
984 let order_keys = if spec.order_by.is_empty() {
985 None
986 } else {
987 let mut keys = Vec::with_capacity(spec.order_by.len());
988 for (k, o) in spec.order_by.iter().enumerate() {
989 let v = if let Some(p) = order_pos[i][k] {
990 row.get(p).cloned().unwrap_or(Value::Null)
991 } else {
992 eval_arg(
993 &o.expr,
994 mat.as_deref().expect("needs_mat for ORDER key"),
995 &ctx,
996 )?
997 };
998 keys.push(v);
999 }
1000 Some(keys)
1001 };
1002 // v7.36 (perf — bugfix v7.36.1 candidate) — first_ordered
1003 // was missing from the single_anon_group fast path,
1004 // sending `(array_agg(x ORDER BY y))[1]` values into
1005 // `update_state(array_agg, …)` whose finalize ignored
1006 // the absent `first_best` and returned `[]`. The slow
1007 // path below has the same branch — keep them aligned.
1008 if spec.first_ordered {
1009 if let Some(keys) = order_keys {
1010 let st = &mut entry.1[i];
1011 let better = match &st.first_best {
1012 None => true,
1013 Some((bk, _)) => {
1014 cmp_order_keys(&spec.order_by, &keys, bk)
1015 == core::cmp::Ordering::Less
1016 }
1017 };
1018 if better {
1019 st.first_best = Some((keys, arg_ref.clone()));
1020 }
1021 }
1022 continue;
1023 }
1024 if spec.distinct {
1025 encode_key_refs_into(core::slice::from_ref(&arg_ref), &mut dkeybuf);
1026 if entry.1[i].seen.contains(dkeybuf.as_str()) {
1027 continue;
1028 }
1029 entry.1[i].seen.insert(dkeybuf.clone());
1030 }
1031 update_state(
1032 &mut entry.1[i],
1033 &spec.name,
1034 arg_ref,
1035 arg2_val.as_ref(),
1036 order_keys,
1037 )?;
1038 }
1039 continue;
1040 }
1041 // Fast key: bound positions + no ci folding -> encode
1042 // straight from borrowed cells; group_vals materialise
1043 // only when the group is NEW.
1044 if all_groups_bound && ci_positions.is_empty() {
1045 refs.clear();
1046 refs.extend(
1047 group_pos
1048 .iter()
1049 .map(|p| row.get(p.unwrap()).unwrap_or(&Value::Null)),
1050 );
1051 encode_key_refs_into(&refs, &mut keybuf_s);
1052 let idx = match groups.get(keybuf_s.as_str()) {
1053 Some(&i) => i,
1054 None => {
1055 let i = order.len();
1056 let init: Vec<AggState> =
1057 (0..agg_specs.len()).map(|_| AggState::default()).collect();
1058 let owned: Vec<Value> = refs.iter().map(|v| (*v).clone()).collect();
1059 order.push((owned, init));
1060 groups.insert(keybuf_s.clone(), i);
1061 i
1062 }
1063 };
1064 let entry = &mut order[idx];
1065 // v7.33 (array_agg perf) — materialise the combined row AT
1066 // MOST once per input row, and only when a spec actually
1067 // needs the eval path (FILTER / non-bound arg / arg2 / non-
1068 // bound ORDER key). Bound args and bound ORDER keys read
1069 // cells by reference below, so the inbox shape (all bound)
1070 // never materialises — killing the per-row ~1 KB clone that
1071 // dominated the ordered-aggregate cost.
1072 let mat: Option<Cow<'_, Row>> = if needs_mat { Some(row.as_row()) } else { None };
1073 for (i, spec) in agg_specs.iter().enumerate() {
1074 // v7.32 (round-29) — FILTER (WHERE cond): exclude rows
1075 // where cond is not TRUE before they reach this
1076 // aggregate's accumulator (and before DISTINCT dedup).
1077 if let Some(f) = &spec.filter
1078 && !matches!(
1079 eval_arg(f, mat.as_deref().expect("needs_mat for FILTER"), &ctx)?,
1080 Value::Bool(true)
1081 )
1082 {
1083 continue;
1084 }
1085 let arg_owned: Value;
1086 let arg_ref: &Value = match (&arg_pos[i], &arg_compiled[i], &spec.arg) {
1087 (Some(p), _, _) => row.get(*p).unwrap_or(&Value::Null),
1088 (None, _, None) => {
1089 arg_owned = Value::Bool(true);
1090 &arg_owned
1091 }
1092 (None, Some(c), _) => {
1093 // v7.36 — compiled-arg fast path. `eval_stack`
1094 // is reused across rows; the Step VM never
1095 // materialises a row for column reads.
1096 arg_owned = eval::eval_compiled_ref(c, row, &ctx, &mut eval_stack)?;
1097 &arg_owned
1098 }
1099 (None, None, Some(e)) => {
1100 arg_owned = eval_arg(
1101 e,
1102 mat.as_deref().expect("needs_mat for non-bound arg"),
1103 &ctx,
1104 )?;
1105 &arg_owned
1106 }
1107 };
1108 let arg2_val = match &spec.arg2 {
1109 None => None,
1110 Some(e) => Some(eval_arg(
1111 e,
1112 mat.as_deref().expect("needs_mat for arg2"),
1113 &ctx,
1114 )?),
1115 };
1116 let order_keys = if spec.order_by.is_empty() {
1117 None
1118 } else {
1119 let mut keys = Vec::with_capacity(spec.order_by.len());
1120 for (k, o) in spec.order_by.iter().enumerate() {
1121 // Bound ORDER key → read the cell by reference; only
1122 // a non-bound key falls to the materialised eval path.
1123 keys.push(match order_pos[i][k] {
1124 Some(p) => row.get(p).cloned().unwrap_or(Value::Null),
1125 None => eval_arg(
1126 &o.expr,
1127 mat.as_deref().expect("needs_mat for non-bound ORDER key"),
1128 &ctx,
1129 )?,
1130 });
1131 }
1132 Some(keys)
1133 };
1134 // v7.33 (array_agg argmax) — first_ordered: keep only the
1135 // running first-by-order element (strict-less replacement
1136 // = ties keep the earliest row, matching the stable-sort
1137 // `[1]`), no array build.
1138 if spec.first_ordered {
1139 if let Some(keys) = order_keys {
1140 let st = &mut entry.1[i];
1141 let better = match &st.first_best {
1142 None => true,
1143 Some((bk, _)) => {
1144 cmp_order_keys(&spec.order_by, &keys, bk)
1145 == core::cmp::Ordering::Less
1146 }
1147 };
1148 if better {
1149 st.first_best = Some((keys, arg_ref.clone()));
1150 }
1151 }
1152 continue;
1153 }
1154 if spec.distinct {
1155 encode_key_refs_into(core::slice::from_ref(&arg_ref), &mut dkeybuf);
1156 if entry.1[i].seen.contains(dkeybuf.as_str()) {
1157 continue;
1158 }
1159 entry.1[i].seen.insert(dkeybuf.clone());
1160 }
1161 update_state(
1162 &mut entry.1[i],
1163 &spec.name,
1164 arg_ref,
1165 arg2_val.as_ref(),
1166 order_keys,
1167 )?;
1168 }
1169 continue;
1170 }
1171 // v7.32 (P4 increment 2) — eval (non-bound) path: present the
1172 // row as a borrowed Row once (Owned → zero-cost borrow; a join
1173 // tuple materialises here exactly once, never on the bound fast
1174 // path above), then the original eval loop runs unchanged.
1175 let row_materialised = row.as_row();
1176 let row: &Row = &row_materialised;
1177 let group_vals: Vec<Value> = group_exprs
1178 .iter()
1179 .map(|g| eval::eval_expr(g, row, &ctx))
1180 .collect::<Result<_, _>>()?;
1181 // v7.17.0 Phase 2.5b — case-insensitive group keying: fold
1182 // only the ci columns, and only when any exist. Display
1183 // value (`group_vals`) stays original — only the key folds.
1184 let key = if ci_positions.is_empty() {
1185 encode_key(&group_vals)
1186 } else {
1187 let mut key_vals = group_vals.clone();
1188 for &i in &ci_positions {
1189 if let Value::Text(s) = &key_vals[i] {
1190 key_vals[i] = Value::Text(s.to_ascii_lowercase());
1191 }
1192 }
1193 encode_key(&key_vals)
1194 };
1195 // Probe by index; the map owns the key once on vacant insert.
1196 let idx = match groups.get(key.as_str()) {
1197 Some(&i) => i,
1198 None => {
1199 let i = order.len();
1200 let init: Vec<AggState> =
1201 (0..agg_specs.len()).map(|_| AggState::default()).collect();
1202 order.push((group_vals.clone(), init));
1203 groups.insert(key, i);
1204 i
1205 }
1206 };
1207 let entry = &mut order[idx];
1208 for (i, spec) in agg_specs.iter().enumerate() {
1209 // v7.32 (round-29) — FILTER (WHERE cond): exclude rows where
1210 // cond is not TRUE before accumulation (and before DISTINCT).
1211 if let Some(f) = &spec.filter
1212 && !matches!(eval_arg(f, row, &ctx)?, Value::Bool(true))
1213 {
1214 continue;
1215 }
1216 let arg_val = match &spec.arg {
1217 None => Value::Bool(true), // count_star: sentinel non-null
1218 Some(e) => eval_arg(e, row, &ctx)?,
1219 };
1220 // v7.17.0 — `string_agg(value, separator)` evaluates the
1221 // separator per row but PG treats it as constant; we
1222 // pass the per-row value into update_state so a future
1223 // varying-separator caller still sees correct output,
1224 // even though SPG (like PG) only uses the most recent.
1225 let arg2_val = match &spec.arg2 {
1226 None => None,
1227 Some(e) => Some(eval_arg(e, row, &ctx)?),
1228 };
1229 // v7.24 (round-16 A) — aggregate-internal ORDER BY:
1230 // evaluate the key tuple against the source row.
1231 let order_keys = if spec.order_by.is_empty() {
1232 None
1233 } else {
1234 let mut keys = Vec::with_capacity(spec.order_by.len());
1235 for o in &spec.order_by {
1236 keys.push(eval_arg(&o.expr, row, &ctx)?);
1237 }
1238 Some(keys)
1239 };
1240 // v7.33 (array_agg argmax) — first_ordered: keep the running
1241 // first-by-order element only (mirrors the bound fast path).
1242 if spec.first_ordered {
1243 if let Some(keys) = order_keys {
1244 let st = &mut entry.1[i];
1245 let better = match &st.first_best {
1246 None => true,
1247 Some((bk, _)) => {
1248 cmp_order_keys(&spec.order_by, &keys, bk) == core::cmp::Ordering::Less
1249 }
1250 };
1251 if better {
1252 st.first_best = Some((keys, arg_val.clone()));
1253 }
1254 }
1255 continue;
1256 }
1257 // v7.25 (round-17) — DISTINCT: drop repeated inputs
1258 // before they reach the accumulator. NULLs flow through
1259 // (each aggregate's own NULL rule applies; PG also
1260 // treats NULL as a single distinct value for array_agg).
1261 if spec.distinct {
1262 let key = encode_key(core::slice::from_ref(&arg_val));
1263 if !entry.1[i].seen.insert(key) {
1264 continue;
1265 }
1266 }
1267 update_state(
1268 &mut entry.1[i],
1269 &spec.name,
1270 &arg_val,
1271 arg2_val.as_ref(),
1272 order_keys,
1273 )?;
1274 }
1275 }
1276 Ok(order)
1277}
1278
1279/// (2a) Build the synthetic per-group schema: `__grp_0..K` then
1280/// `__agg_0..N`. Group types are probed from the first row; aggregate
1281/// types from each spec.
1282fn build_synth_schema(
1283 rows: &[RowRef<'_>],
1284 group_exprs: &[Expr],
1285 agg_specs: &[AggSpec],
1286 schema_cols: &[ColumnSchema],
1287 table_alias: Option<&str>,
1288) -> Result<Vec<ColumnSchema>, EvalError> {
1289 let ctx = EvalContext::new(schema_cols, table_alias);
1290 // Build synthetic schema: __grp_0..K then __agg_0..N.
1291 let group_types: Vec<DataType> = if rows.is_empty() {
1292 // Use Text as a safe stand-in — empty result means schema isn't
1293 // observable. Avoids needing to evaluate group exprs on no row.
1294 group_exprs.iter().map(|_| DataType::Text).collect()
1295 } else {
1296 let probe_row = rows[0].as_row();
1297 let probe: &Row = &probe_row;
1298 group_exprs
1299 .iter()
1300 .map(|g| {
1301 eval::eval_expr(g, probe, &ctx).map(|v| v.data_type().unwrap_or(DataType::Text))
1302 })
1303 .collect::<Result<_, _>>()?
1304 };
1305 let agg_types: Vec<DataType> = agg_specs
1306 .iter()
1307 .map(|spec| infer_agg_type(spec, schema_cols))
1308 .collect();
1309 let mut synth_schema: Vec<ColumnSchema> = Vec::new();
1310 for (i, ty) in group_types.iter().enumerate() {
1311 synth_schema.push(ColumnSchema::new(format!("__grp_{i}"), *ty, true));
1312 }
1313 for (i, ty) in agg_types.iter().enumerate() {
1314 synth_schema.push(ColumnSchema::new(format!("__agg_{i}"), *ty, true));
1315 }
1316 Ok(synth_schema)
1317}
1318
1319/// (2b) Materialise one synthetic row per group (insertion order):
1320/// apply each aggregate's internal ORDER BY, then finalise the running
1321/// state into the group + aggregate cells.
1322/// v7.33 — compare two aggregate-internal ORDER BY key tuples under the
1323/// per-key DESC / NULLS directives. This is the exact comparator the
1324/// finalize sort uses, factored out so the `first_ordered` argmax
1325/// accumulator's "keep first" decision is provably identical to taking
1326/// element `[1]` of the fully-sorted array.
1327fn cmp_order_keys(
1328 order_by: &[spg_sql::ast::OrderBy],
1329 a: &[Value],
1330 b: &[Value],
1331) -> core::cmp::Ordering {
1332 for (k, o) in order_by.iter().enumerate() {
1333 let cmp = crate::order_by_value_cmp(o.desc, o.nulls_first, &a[k], &b[k]);
1334 if cmp != core::cmp::Ordering::Equal {
1335 return cmp;
1336 }
1337 }
1338 core::cmp::Ordering::Equal
1339}
1340
1341fn finalize_synth_rows(
1342 order: &[(Vec<Value>, Vec<AggState>)],
1343 agg_specs: &[AggSpec],
1344 synth_schema: &[ColumnSchema],
1345 rows: &[RowRef<'_>],
1346 schema_cols: &[ColumnSchema],
1347 table_alias: Option<&str>,
1348) -> Result<Vec<Row>, EvalError> {
1349 let ctx = EvalContext::new(schema_cols, table_alias);
1350 // v7.32 (round-29) — ordered-set direct arguments (the percentile
1351 // fraction) are constant per PG, so evaluate each once up front.
1352 let direct_arg_vals: Vec<Option<Value>> = agg_specs
1353 .iter()
1354 .map(|spec| match (&spec.direct_arg, rows.first()) {
1355 (Some(e), Some(r)) => eval::eval_expr(e, &r.as_row(), &ctx).map(Some),
1356 _ => Ok(None),
1357 })
1358 .collect::<Result<_, _>>()?;
1359
1360 // Materialise synthetic rows (insertion order = `order`).
1361 let mut synth_rows: Vec<Row> = Vec::new();
1362 for (gvals, states) in order {
1363 let mut values: Vec<Value> = Vec::with_capacity(synth_schema.len());
1364 values.extend(gvals.iter().cloned());
1365 for (i, st) in states.iter().enumerate() {
1366 // v7.33 (array_agg argmax) — first_ordered: the running
1367 // first-by-order value IS the result; no array build/sort.
1368 if agg_specs[i].first_ordered {
1369 values.push(
1370 st.first_best
1371 .as_ref()
1372 .map_or(Value::Null, |(_, v)| v.clone()),
1373 );
1374 continue;
1375 }
1376 // v7.24 (round-16 A) — order the collected items per the
1377 // aggregate-internal ORDER BY before finalize consumes
1378 // them.
1379 let st_sorted;
1380 let st_final: &AggState =
1381 if !agg_specs[i].order_by.is_empty() && st.item_keys.len() == st.items.len() {
1382 let mut idx: Vec<usize> = (0..st.items.len()).collect();
1383 let ob = &agg_specs[i].order_by;
1384 idx.sort_by(|&x, &y| cmp_order_keys(ob, &st.item_keys[x], &st.item_keys[y]));
1385 let mut sorted = st.clone();
1386 sorted.items = idx.iter().map(|&j| st.items[j].clone()).collect();
1387 st_sorted = sorted;
1388 &st_sorted
1389 } else {
1390 st
1391 };
1392 // Ordered-set aggregates compute from the sorted items + the
1393 // direct fraction; everything else uses the running state.
1394 let v = if is_within_group_name(&agg_specs[i].name) {
1395 finalize_ordered_set(
1396 &agg_specs[i].name,
1397 st_final,
1398 direct_arg_vals[i].as_ref(),
1399 agg_specs[i].order_by.first(),
1400 )
1401 } else {
1402 finalize(&agg_specs[i].name, st_final)
1403 };
1404 values.push(v);
1405 }
1406 synth_rows.push(Row::new(values));
1407 }
1408 Ok(synth_rows)
1409}
1410
1411/// (3) Rewrite the user's SELECT items + HAVING to reference the
1412/// synthetic columns, filter groups by HAVING, and project each
1413/// surviving group into an output row. The synth rows ride alongside
1414/// (`kept_synth`) so post-LIMIT deferred subqueries can evaluate later.
1415#[allow(clippy::too_many_lines)]
1416fn project_groups(
1417 synth_rows: Vec<Row>,
1418 stmt: &SelectStatement,
1419 group_exprs: &[Expr],
1420 agg_specs: &[AggSpec],
1421 synth_schema: &[ColumnSchema],
1422 correlated_eval: Option<CorrelatedEval<'_>>,
1423) -> Result<Projection, EvalError> {
1424 // Rewrite the user's SELECT items + ORDER BY to reference synthetic
1425 // columns. After rewriting, every remaining `Expr::Column` must
1426 // resolve against the synthetic schema (i.e. must have been a GROUP
1427 // BY expression).
1428 let columns: Vec<ColumnSchema> = stmt
1429 .items
1430 .iter()
1431 .map(|item| match item {
1432 SelectItem::Wildcard => Err(EvalError::TypeMismatch {
1433 detail: "SELECT * with aggregates is not supported".into(),
1434 }),
1435 SelectItem::Expr { expr, alias } => {
1436 let rewritten = rewrite_expr(expr, group_exprs, agg_specs);
1437 let name = alias.clone().unwrap_or_else(|| expr.to_string());
1438 Ok(ColumnSchema::new(
1439 name,
1440 agg_or_group_type(&rewritten, synth_schema),
1441 true,
1442 ))
1443 }
1444 })
1445 .collect::<Result<_, _>>()?;
1446
1447 // Project per synthetic row. HAVING filters out groups *before*
1448 // we keep the projected row — same semantics as PG: HAVING runs
1449 // against the aggregated row (so `HAVING count(*) > 1` works) and
1450 // sees only group-by'd columns plus aggregate values.
1451 let synth_ctx = EvalContext::new(synth_schema, None);
1452 let having_rewritten = stmt
1453 .having
1454 .as_ref()
1455 .map(|h| rewrite_expr(h, group_exprs, agg_specs));
1456 // v7.30 (phase 3e-1) - rewrite SELECT items ONCE. This ran per
1457 // GROUP (23.5k x 9 items of AST cloning = ~48% of the inbox
1458 // query in sampled stacks); the rewrite is group-independent.
1459 // Stable addresses also let the per-expression subquery plans
1460 // (v7.29 3c) hit across groups instead of rebuilding.
1461 let items_rewritten: alloc::vec::Vec<Option<Expr>> = stmt
1462 .items
1463 .iter()
1464 .map(|item| match item {
1465 SelectItem::Expr { expr, .. } => Some(rewrite_expr(expr, group_exprs, agg_specs)),
1466 SelectItem::Wildcard => None,
1467 })
1468 .collect();
1469 // v7.31 (perf — PG lesson #1): subquery-bearing select items
1470 // deferred to post-LIMIT, when no sort/filter key can observe
1471 // them. ORDER BY rewrites are hoisted here so the safety check
1472 // and the sort below share one rewrite pass.
1473 let order_rewritten: Vec<Expr> = stmt
1474 .order_by
1475 .iter()
1476 .map(|o| rewrite_expr(&o.expr, group_exprs, agg_specs))
1477 .collect();
1478 let defer_enabled = correlated_eval.is_some()
1479 && !stmt.distinct
1480 && !having_rewritten
1481 .as_ref()
1482 .is_some_and(crate::expr_has_subquery)
1483 && !order_rewritten.iter().any(crate::expr_has_subquery);
1484 let deferred: Vec<(usize, Expr)> = if defer_enabled {
1485 items_rewritten
1486 .iter()
1487 .enumerate()
1488 .filter_map(|(i, r)| {
1489 r.as_ref()
1490 .filter(|e| crate::expr_has_subquery(e))
1491 .map(|e| (i, e.clone()))
1492 })
1493 .collect()
1494 } else {
1495 Vec::new()
1496 };
1497 // v7.32 (architecture v2, P2) — compile the per-group synth-row
1498 // expressions ONCE. The projection / HAVING here run per GROUP
1499 // (24k for the inbox shape) × per item; the rewritten exprs are
1500 // mostly `Column(__agg_N)` / `Column(__grp_K)` against the synth
1501 // schema — flat step programs, no tree walk per group.
1502 let having_compiled = having_rewritten
1503 .as_ref()
1504 .filter(|h| eval::fully_compilable(h))
1505 .map(|h| eval::compile_expr(h, &synth_ctx));
1506 let items_compiled: Vec<Option<eval::CompiledExpr>> = items_rewritten
1507 .iter()
1508 .enumerate()
1509 .map(|(i, r)| {
1510 r.as_ref()
1511 .filter(|e| !deferred.iter().any(|(c, _)| *c == i) && eval::fully_compilable(e))
1512 .map(|e| eval::compile_expr(e, &synth_ctx))
1513 })
1514 .collect();
1515 let mut kept_synth: Vec<Row> = Vec::new();
1516 let mut out_rows: Vec<Row> = Vec::new();
1517 let mut stack: Vec<Value> = Vec::new();
1518 for srow in synth_rows {
1519 if let Some(hc) = &having_compiled {
1520 let cond = eval::eval_compiled(hc, &srow, &synth_ctx, &mut stack)?;
1521 if !matches!(cond, Value::Bool(true)) {
1522 continue;
1523 }
1524 } else if let Some(h) = &having_rewritten {
1525 let cond = match correlated_eval {
1526 Some(f) if crate::expr_has_subquery(h) => f(h, &srow, &synth_ctx)?,
1527 _ => eval::eval_expr(h, &srow, &synth_ctx)?,
1528 };
1529 if !matches!(cond, Value::Bool(true)) {
1530 continue;
1531 }
1532 }
1533 let mut values: Vec<Value> = Vec::with_capacity(columns.len());
1534 for (i, rewritten) in items_rewritten.iter().enumerate() {
1535 let Some(rewritten) = rewritten else { continue };
1536 if deferred.iter().any(|(c, _)| *c == i) {
1537 values.push(Value::Null);
1538 continue;
1539 }
1540 values.push(if let Some(cc) = &items_compiled[i] {
1541 eval::eval_compiled(cc, &srow, &synth_ctx, &mut stack)?
1542 } else {
1543 match correlated_eval {
1544 Some(f) if crate::expr_has_subquery(rewritten) => {
1545 f(rewritten, &srow, &synth_ctx)?
1546 }
1547 _ => eval::eval_expr(rewritten, &srow, &synth_ctx)?,
1548 }
1549 });
1550 }
1551 kept_synth.push(srow);
1552 out_rows.push(Row::new(values));
1553 }
1554 Ok(Projection {
1555 columns,
1556 out_rows,
1557 kept_synth,
1558 deferred,
1559 order_rewritten,
1560 })
1561}
1562
1563/// (4) Sort the projected output by the rewritten ORDER BY keys. The
1564/// synth rows ride through the sort so deferred subqueries evaluate
1565/// against the surviving groups after the caller's LIMIT truncation.
1566fn sort_synth_by_order_by(
1567 synth_schema: &[ColumnSchema],
1568 order_by: &[spg_sql::ast::OrderBy],
1569 order_rewritten: &[Expr],
1570 mut kept_synth: Vec<Row>,
1571 mut out_rows: Vec<Row>,
1572 correlated_eval: Option<CorrelatedEval<'_>>,
1573 keep_n: Option<usize>,
1574) -> Result<(Vec<Row>, Vec<Row>), EvalError> {
1575 let synth_ctx = EvalContext::new(synth_schema, None);
1576 // v6.4.0 — multi-key ORDER BY on aggregate output. Each key
1577 // gets its own rewrite + per-key DESC flag. (Rewrites hoisted
1578 // above as `order_rewritten` — shared with the deferral
1579 // safety check.)
1580 let keys_meta: Vec<(bool, Option<bool>)> =
1581 order_by.iter().map(|o| (o.desc, o.nulls_first)).collect();
1582 // P2: compile order-by keys once (per-group sort keys are
1583 // the same `__agg_N` / `__grp_K` shape as the projection).
1584 let order_compiled: Vec<Option<eval::CompiledExpr>> = order_rewritten
1585 .iter()
1586 .map(|e| {
1587 Some(e)
1588 .filter(|e| eval::fully_compilable(e))
1589 .map(|e| eval::compile_expr(e, &synth_ctx))
1590 })
1591 .collect();
1592 // The synth row rides through the sort so deferred exprs can
1593 // evaluate against the surviving groups after the caller's
1594 // LIMIT truncation.
1595 let mut keystack: Vec<Value> = Vec::new();
1596 let mut tagged: Vec<(Vec<Value>, Row, Row)> = Vec::with_capacity(kept_synth.len());
1597 for (s, o) in kept_synth.into_iter().zip(out_rows) {
1598 let mut keys = Vec::with_capacity(order_rewritten.len());
1599 for (e, oc) in order_rewritten.iter().zip(&order_compiled) {
1600 keys.push(if let Some(oc) = oc {
1601 eval::eval_compiled(oc, &s, &synth_ctx, &mut keystack)?
1602 } else {
1603 match correlated_eval {
1604 Some(f) if crate::expr_has_subquery(e) => f(e, &s, &synth_ctx)?,
1605 _ => eval::eval_expr(e, &s, &synth_ctx)?,
1606 }
1607 });
1608 }
1609 tagged.push((keys, s, o));
1610 }
1611 let cmp = |a: &(Vec<Value>, Row, Row), b: &(Vec<Value>, Row, Row)| {
1612 use core::cmp::Ordering;
1613 for (i, (ka, kb)) in a.0.iter().zip(b.0.iter()).enumerate() {
1614 let (desc, nf) = keys_meta[i];
1615 let c = crate::order_by_value_cmp(desc, nf, ka, kb);
1616 if c != Ordering::Equal {
1617 return c;
1618 }
1619 }
1620 Ordering::Equal
1621 };
1622 // v7.37.3 — top-K partial sort when `keep_n` is small enough to
1623 // matter (`Some(k)` with `k < tagged.len()` and `k > 0`).
1624 // `select_nth_unstable_by` partitions in O(N), then we sort the
1625 // surviving prefix in O(K log K). Total = O(N + K log K) vs
1626 // O(N log N) the full sort would pay — matches the inbox-listing
1627 // shape PG uses.
1628 //
1629 match keep_n {
1630 Some(k) if k < tagged.len() && k > 0 => {
1631 let pivot = k - 1;
1632 tagged.select_nth_unstable_by(pivot, cmp);
1633 tagged[..k].sort_by(cmp);
1634 tagged.truncate(k);
1635 }
1636 _ => {
1637 tagged.sort_by(cmp);
1638 }
1639 }
1640 kept_synth = Vec::with_capacity(tagged.len());
1641 out_rows = Vec::with_capacity(tagged.len());
1642 for (_, s, o) in tagged {
1643 kept_synth.push(s);
1644 out_rows.push(o);
1645 }
1646 Ok((kept_synth, out_rows))
1647}
1648
1649/// v7.17.0 — walk the statement again to validate the positional
1650/// arity of every aggregate call site. Done after AST collection
1651/// rather than inside `collect_aggregates` so the collector stays
1652/// infallible; callers in `run()` can do a single early-error
1653/// exit before any per-row work.
1654fn validate_agg_arities(stmt: &SelectStatement, _specs: &[AggSpec]) -> Result<(), EvalError> {
1655 fn walk(e: &Expr) -> Result<(), EvalError> {
1656 if let Expr::FunctionCall { name, args } = e {
1657 let lower = name.to_ascii_lowercase();
1658 let expected: Option<usize> = match lower.as_str() {
1659 "count_star" => Some(0),
1660 "count" | "sum" | "avg" | "min" | "max" | "array_agg"
1661 // v7.17.0 — boolean aggregates also take exactly
1662 // one arg. `every` is an alias normalised inside
1663 // collect_aggregates / rewrite_expr.
1664 | "bool_and" | "bool_or" | "every"
1665 // v7.32 (round-29) — statistical + bitwise aggregates
1666 // + single-arg JSON aggregate.
1667 | "stddev" | "stddev_samp" | "stddev_pop"
1668 | "variance" | "var_samp" | "var_pop"
1669 | "bit_and" | "bit_or" | "bit_xor"
1670 | "json_agg" | "jsonb_agg" => Some(1),
1671 // v7.32 (round-29) — two-argument aggregates: string_agg,
1672 // the regression family f(Y, X), and json_object_agg.
1673 "string_agg"
1674 | "covar_pop" | "covar_samp" | "corr"
1675 | "regr_count" | "regr_avgx" | "regr_avgy" | "regr_slope"
1676 | "regr_intercept" | "regr_r2" | "regr_sxx" | "regr_syy" | "regr_sxy"
1677 | "json_object_agg" | "jsonb_object_agg" => Some(2),
1678 _ => None,
1679 };
1680 if let Some(want) = expected
1681 && args.len() != want
1682 {
1683 return Err(EvalError::TypeMismatch {
1684 detail: alloc::format!("{lower}() takes {want} arg(s), got {}", args.len()),
1685 });
1686 }
1687 for a in args {
1688 walk(a)?;
1689 }
1690 } else if let Expr::Binary { lhs, rhs, .. } = e {
1691 walk(lhs)?;
1692 walk(rhs)?;
1693 } else if let Expr::Unary { expr, .. }
1694 | Expr::Cast { expr, .. }
1695 | Expr::IsNull { expr, .. } = e
1696 {
1697 walk(expr)?;
1698 }
1699 Ok(())
1700 }
1701 for item in &stmt.items {
1702 if let SelectItem::Expr { expr, .. } = item {
1703 walk(expr)?;
1704 }
1705 }
1706 for o in &stmt.order_by {
1707 walk(&o.expr)?;
1708 }
1709 if let Some(h) = &stmt.having {
1710 walk(h)?;
1711 }
1712 Ok(())
1713}
1714
1715/// v7.33 (array_agg argmax) — recognise `(array_agg(x ORDER BY y))[1]`,
1716/// the argmax/argmin idiom: a non-DISTINCT ordered `array_agg`
1717/// subscripted by the constant 1. Returns `(value_arg, order_by,
1718/// filter)` on a match. When matched, the whole per-group array build +
1719/// sort + materialise is replaced by a running first-by-order scalar
1720/// accumulator and the subscript node is consumed (replaced by the
1721/// synthetic column). collect_aggregates and rewrite_expr share this one
1722/// matcher so their `__agg_<i>` assignment stays in lockstep.
1723fn first_ordered_array_agg(e: &Expr) -> Option<(&Expr, &[spg_sql::ast::OrderBy], Option<&Expr>)> {
1724 let Expr::ArraySubscript { target, index } = e else {
1725 return None;
1726 };
1727 if !matches!(
1728 index.as_ref(),
1729 Expr::Literal(spg_sql::ast::Literal::Integer(1))
1730 ) {
1731 return None;
1732 }
1733 let Expr::AggregateOrdered {
1734 call,
1735 order_by,
1736 distinct,
1737 filter,
1738 } = target.as_ref()
1739 else {
1740 return None;
1741 };
1742 if *distinct || order_by.is_empty() {
1743 return None;
1744 }
1745 let Expr::FunctionCall { name, args } = call.as_ref() else {
1746 return None;
1747 };
1748 if !name.eq_ignore_ascii_case("array_agg") || args.len() != 1 {
1749 return None;
1750 }
1751 Some((&args[0], order_by, filter.as_deref()))
1752}
1753
1754fn collect_aggregates(e: &Expr, out: &mut Vec<AggSpec>) {
1755 match e {
1756 // v7.24 (round-16 A) — ordered aggregate: register the inner
1757 // call's spec with the ordering attached.
1758 Expr::AggregateOrdered {
1759 call,
1760 order_by,
1761 distinct,
1762 filter,
1763 } => {
1764 if let Expr::FunctionCall { name, args } = call.as_ref() {
1765 let lower = name.to_ascii_lowercase();
1766 if is_aggregate_name(&lower) {
1767 let canonical = if lower == "every" {
1768 "bool_and".to_string()
1769 } else {
1770 lower
1771 };
1772 // Ordered-set aggregates (`percentile_cont(f)
1773 // WITHIN GROUP (ORDER BY x)`) take the value to
1774 // aggregate from the sort spec and the in-parens
1775 // arg as the direct (fraction) argument.
1776 let ordered_set = is_within_group_name(&canonical);
1777 let (arg, direct_arg) = if ordered_set {
1778 (
1779 order_by.first().map(|o| o.expr.clone()),
1780 args.first().cloned(),
1781 )
1782 } else {
1783 (args.first().cloned(), None)
1784 };
1785 let spec = AggSpec {
1786 name: canonical.clone(),
1787 arg,
1788 arg2: if agg_uses_second_arg(&canonical) {
1789 args.get(1).cloned()
1790 } else {
1791 None
1792 },
1793 distinct: *distinct,
1794 order_by: order_by.clone(),
1795 filter: filter.as_deref().cloned(),
1796 direct_arg,
1797 first_ordered: false,
1798 };
1799 if !out.iter().any(|s| {
1800 s.name == spec.name
1801 && s.arg == spec.arg
1802 && s.arg2 == spec.arg2
1803 && s.distinct == spec.distinct
1804 && s.order_by == spec.order_by
1805 && s.filter == spec.filter
1806 && s.direct_arg == spec.direct_arg
1807 && s.first_ordered == spec.first_ordered
1808 }) {
1809 out.push(spec);
1810 }
1811 return;
1812 }
1813 }
1814 collect_aggregates(call, out);
1815 for o in order_by {
1816 collect_aggregates(&o.expr, out);
1817 }
1818 }
1819 Expr::FunctionCall { name, args } => {
1820 let lower = name.to_ascii_lowercase();
1821 if is_aggregate_name(&lower) {
1822 let arg = if lower == "count_star" {
1823 None
1824 } else {
1825 args.first().cloned()
1826 };
1827 // v7.17.0 — second positional arg for
1828 // `string_agg(value, separator)`; v7.32 — also the
1829 // regression family `f(Y, X)` and `json_object_agg`.
1830 let arg2 = if agg_uses_second_arg(&lower) {
1831 args.get(1).cloned()
1832 } else {
1833 None
1834 };
1835 // v7.17.0 — `every` is the SQL-standard alias for
1836 // `bool_and`; collapse at collection time so
1837 // update_state / finalize need only one arm.
1838 let canonical = if lower == "every" {
1839 "bool_and".to_string()
1840 } else {
1841 lower
1842 };
1843 let spec = AggSpec {
1844 name: canonical,
1845 arg: arg.clone(),
1846 arg2: arg2.clone(),
1847 distinct: false,
1848 order_by: Vec::new(),
1849 filter: None,
1850 direct_arg: None,
1851 first_ordered: false,
1852 };
1853 if !out.iter().any(|s| {
1854 s.name == spec.name
1855 && s.arg == spec.arg
1856 && s.arg2 == spec.arg2
1857 && !s.distinct
1858 && s.order_by == spec.order_by
1859 && s.filter.is_none()
1860 && !s.first_ordered
1861 }) {
1862 out.push(spec);
1863 }
1864 // Don't recurse into the arg — nested aggregates are
1865 // illegal in standard SQL.
1866 } else {
1867 for a in args {
1868 collect_aggregates(a, out);
1869 }
1870 }
1871 }
1872 Expr::Binary { lhs, rhs, .. } => {
1873 collect_aggregates(lhs, out);
1874 collect_aggregates(rhs, out);
1875 }
1876 Expr::Unary { expr, .. } | Expr::Cast { expr, .. } | Expr::IsNull { expr, .. } => {
1877 collect_aggregates(expr, out);
1878 }
1879 Expr::Like { expr, pattern, .. } => {
1880 collect_aggregates(expr, out);
1881 collect_aggregates(pattern, out);
1882 }
1883 Expr::InList { expr, list, .. } => {
1884 collect_aggregates(expr, out);
1885 for item in list {
1886 collect_aggregates(item, out);
1887 }
1888 }
1889 Expr::Extract { source, .. } => collect_aggregates(source, out),
1890 // v4.10 subquery + v4.12 window / Literal / Column —
1891 // non-recursing leaves for the aggregate collector.
1892 Expr::ScalarSubquery(_)
1893 | Expr::Exists { .. }
1894 | Expr::InSubquery { .. }
1895 | Expr::WindowFunction { .. }
1896 | Expr::Literal(_)
1897 | Expr::Placeholder(_)
1898 | Expr::Column(_) => {}
1899 // v7.10.10 — recurse into array constructor children +
1900 // subscript / ANY/ALL operands.
1901 Expr::Array(items) => {
1902 for elem in items {
1903 collect_aggregates(elem, out);
1904 }
1905 }
1906 Expr::ArraySubscript { target, index } => {
1907 // v7.33 (array_agg argmax) — `(array_agg(x ORDER BY y))[1]`
1908 // collects as a first_ordered spec; the subscript is consumed
1909 // here (do NOT recurse into the array_agg, or it would also
1910 // register a plain full-array spec).
1911 if let Some((arg, order_by, filter)) = first_ordered_array_agg(e) {
1912 let spec = AggSpec {
1913 name: "array_agg".to_string(),
1914 arg: Some(arg.clone()),
1915 arg2: None,
1916 distinct: false,
1917 order_by: order_by.to_vec(),
1918 filter: filter.cloned(),
1919 direct_arg: None,
1920 first_ordered: true,
1921 };
1922 if !out.iter().any(|s| {
1923 s.name == spec.name
1924 && s.arg == spec.arg
1925 && s.order_by == spec.order_by
1926 && s.filter == spec.filter
1927 && s.first_ordered
1928 }) {
1929 out.push(spec);
1930 }
1931 return;
1932 }
1933 collect_aggregates(target, out);
1934 collect_aggregates(index, out);
1935 }
1936 Expr::AnyAll { expr, array, .. } => {
1937 collect_aggregates(expr, out);
1938 collect_aggregates(array, out);
1939 }
1940 Expr::Case {
1941 operand,
1942 branches,
1943 else_branch,
1944 } => {
1945 if let Some(o) = operand {
1946 collect_aggregates(o, out);
1947 }
1948 for (w, t) in branches {
1949 collect_aggregates(w, out);
1950 collect_aggregates(t, out);
1951 }
1952 if let Some(e) = else_branch {
1953 collect_aggregates(e, out);
1954 }
1955 }
1956 }
1957}
1958
1959fn update_state(
1960 st: &mut AggState,
1961 name: &str,
1962 v: &Value,
1963 arg2: Option<&Value>,
1964 order_keys: Option<Vec<Value>>,
1965) -> Result<(), EvalError> {
1966 let is_null = matches!(v, Value::Null);
1967 match name {
1968 "count_star" => st.count += 1,
1969 "count" => {
1970 if !is_null {
1971 st.count += 1;
1972 }
1973 }
1974 "sum" | "avg" => {
1975 if is_null {
1976 return Ok(());
1977 }
1978 st.count += 1;
1979 match v {
1980 Value::Int(n) => st.sum_int += i64::from(*n),
1981 Value::BigInt(n) => st.sum_int += *n,
1982 Value::Float(x) => {
1983 st.use_float = true;
1984 st.sum_float += *x;
1985 }
1986 other => {
1987 return Err(EvalError::TypeMismatch {
1988 detail: format!("sum/avg need numeric, got {:?}", other.data_type()),
1989 });
1990 }
1991 }
1992 }
1993 "min" => {
1994 if is_null {
1995 return Ok(());
1996 }
1997 match &st.extreme {
1998 None => st.extreme = Some(v.clone()),
1999 Some(cur) => {
2000 if value_cmp(v, cur) == core::cmp::Ordering::Less {
2001 st.extreme = Some(v.clone());
2002 }
2003 }
2004 }
2005 }
2006 "max" => {
2007 if is_null {
2008 return Ok(());
2009 }
2010 match &st.extreme {
2011 None => st.extreme = Some(v.clone()),
2012 Some(cur) => {
2013 if value_cmp(v, cur) == core::cmp::Ordering::Greater {
2014 st.extreme = Some(v.clone());
2015 }
2016 }
2017 }
2018 }
2019 // v7.17.0 — string_agg(value, separator). NULL value is
2020 // skipped (PG aggregate-skip-null). Separator captured
2021 // from the latest row that flows through; matches PG's
2022 // semantics of evaluating the separator per row but using
2023 // the last value at finalize time (in practice it's
2024 // constant). count is bumped so we can distinguish "empty
2025 // group → NULL" from "all-NULL group → NULL".
2026 "string_agg" => {
2027 if let Some(sep) = arg2
2028 && let Value::Text(s) = sep
2029 {
2030 st.separator = Some(s.clone());
2031 }
2032 if is_null {
2033 return Ok(());
2034 }
2035 if let Value::Text(s) = v {
2036 st.items.push(Value::Text(s.clone()));
2037 if let Some(k) = order_keys {
2038 st.item_keys.push(k);
2039 }
2040 st.count += 1;
2041 } else {
2042 return Err(EvalError::TypeMismatch {
2043 detail: format!("string_agg requires text value, got {:?}", v.data_type()),
2044 });
2045 }
2046 }
2047 // v7.17.0 — array_agg(value). Unlike string_agg, NULL
2048 // elements are KEPT in the array (PG behaviour); the
2049 // result is NULL only when ZERO rows fed in. Element type
2050 // is locked from the first row's value type; subsequent
2051 // rows must match (PG also rejects mixed-type array_agg).
2052 "array_agg" => {
2053 st.items.push(v.clone());
2054 if let Some(k) = order_keys {
2055 st.item_keys.push(k);
2056 }
2057 st.count += 1;
2058 }
2059 // v7.17.0 — bool_and(p): TRUE iff every non-NULL input is
2060 // TRUE. NULL skipped; running accumulator stays at TRUE
2061 // until the first non-NULL FALSE.
2062 "bool_and" => {
2063 if is_null {
2064 return Ok(());
2065 }
2066 let b = match v {
2067 Value::Bool(b) => *b,
2068 other => {
2069 return Err(EvalError::TypeMismatch {
2070 detail: format!("bool_and requires bool, got {:?}", other.data_type()),
2071 });
2072 }
2073 };
2074 st.bool_acc = Some(st.bool_acc.map_or(b, |acc| acc && b));
2075 }
2076 // v7.17.0 — bool_or(p): TRUE iff any non-NULL input is
2077 // TRUE. NULL skipped.
2078 "bool_or" => {
2079 if is_null {
2080 return Ok(());
2081 }
2082 let b = match v {
2083 Value::Bool(b) => *b,
2084 other => {
2085 return Err(EvalError::TypeMismatch {
2086 detail: format!("bool_or requires bool, got {:?}", other.data_type()),
2087 });
2088 }
2089 };
2090 st.bool_acc = Some(st.bool_acc.map_or(b, |acc| acc || b));
2091 }
2092 // v7.32 (round-29) — variance / stddev family. Accumulate the
2093 // running sum (sum_float) and sum of squares (sum_sq) over the
2094 // non-NULL numeric inputs; finalize divides by n or n-1.
2095 "stddev" | "stddev_samp" | "stddev_pop" | "variance" | "var_samp" | "var_pop" => {
2096 if is_null {
2097 return Ok(());
2098 }
2099 let x = match v {
2100 Value::Int(n) => f64::from(*n),
2101 Value::SmallInt(n) => f64::from(*n),
2102 Value::BigInt(n) => *n as f64,
2103 Value::Float(x) => *x,
2104 other => {
2105 return Err(EvalError::TypeMismatch {
2106 detail: format!("{name} needs numeric, got {:?}", other.data_type()),
2107 });
2108 }
2109 };
2110 st.count += 1;
2111 st.sum_float += x;
2112 st.sum_sq += x * x;
2113 }
2114 // v7.32 (round-29) — bitwise aggregates over integer inputs.
2115 "bit_and" | "bit_or" | "bit_xor" => {
2116 if is_null {
2117 return Ok(());
2118 }
2119 let n = match v {
2120 Value::Int(n) => i64::from(*n),
2121 Value::SmallInt(n) => i64::from(*n),
2122 Value::BigInt(n) => *n,
2123 other => {
2124 return Err(EvalError::TypeMismatch {
2125 detail: format!("{name} needs integer, got {:?}", other.data_type()),
2126 });
2127 }
2128 };
2129 st.bit_acc = Some(match (st.bit_acc, name) {
2130 (None, _) => n,
2131 (Some(acc), "bit_and") => acc & n,
2132 (Some(acc), "bit_or") => acc | n,
2133 (Some(acc), _) => acc ^ n, // bit_xor
2134 });
2135 }
2136 // v7.32 (round-29) — WITHIN GROUP aggregates (ordered-set +
2137 // hypothetical-set) collect the sort value (NULLs ignored, per
2138 // PG) into `items`, sorted at finalize by the parallel
2139 // `item_keys`.
2140 n if is_within_group_name(n) => {
2141 if is_null {
2142 return Ok(());
2143 }
2144 st.items.push(v.clone());
2145 if let Some(k) = order_keys {
2146 st.item_keys.push(k);
2147 }
2148 st.count += 1;
2149 }
2150 // v7.32 (round-29) — regression family f(Y, X). Only rows with
2151 // BOTH inputs non-NULL contribute (PG semantics). `v` is Y,
2152 // `arg2` is X.
2153 n if is_regression_name(n) => {
2154 let (Some(y), Some(x)) = (agg_value_to_f64(v), arg2.and_then(agg_value_to_f64)) else {
2155 return Ok(()); // NULL (or non-numeric) in either input
2156 };
2157 st.reg_n += 1;
2158 st.reg_sx += x;
2159 st.reg_sy += y;
2160 st.reg_sxx += x * x;
2161 st.reg_syy += y * y;
2162 st.reg_sxy += x * y;
2163 }
2164 // v7.32 (round-29) — json_agg / jsonb_agg collect every input
2165 // (NULL becomes JSON null, per PG) in row order.
2166 "json_agg" | "jsonb_agg" => {
2167 st.items.push(v.clone());
2168 st.count += 1;
2169 }
2170 // v7.32 (round-29) — json_object_agg(key, value): keys in
2171 // `items`, values in `aux_items`. A NULL key is skipped (PG
2172 // raises; we drop it rather than abort the whole query).
2173 "json_object_agg" | "jsonb_object_agg" => {
2174 if is_null {
2175 return Ok(());
2176 }
2177 st.items.push(v.clone());
2178 st.aux_items.push(arg2.cloned().unwrap_or(Value::Null));
2179 st.count += 1;
2180 }
2181 _ => unreachable!("non-aggregate {name} in update_state"),
2182 }
2183 Ok(())
2184}
2185
2186#[allow(clippy::cast_precision_loss)]
2187fn finalize(name: &str, st: &AggState) -> Value {
2188 match name {
2189 "count" | "count_star" => Value::BigInt(st.count),
2190 "sum" => {
2191 if st.count == 0 {
2192 Value::Null
2193 } else if st.use_float {
2194 Value::Float(st.sum_float + (st.sum_int as f64))
2195 } else {
2196 Value::BigInt(st.sum_int)
2197 }
2198 }
2199 "avg" => {
2200 if st.count == 0 {
2201 Value::Null
2202 } else {
2203 let total = if st.use_float {
2204 st.sum_float + (st.sum_int as f64)
2205 } else {
2206 st.sum_int as f64
2207 };
2208 Value::Float(total / (st.count as f64))
2209 }
2210 }
2211 "min" | "max" => st.extreme.clone().unwrap_or(Value::Null),
2212 // v7.17.0 — string_agg: join all collected text items with
2213 // the captured separator. Empty / all-NULL group → NULL
2214 // (PG semantics).
2215 "string_agg" => {
2216 if st.items.is_empty() {
2217 return Value::Null;
2218 }
2219 let sep = st.separator.clone().unwrap_or_default();
2220 let mut out = String::new();
2221 for (i, item) in st.items.iter().enumerate() {
2222 if i > 0 {
2223 out.push_str(&sep);
2224 }
2225 if let Value::Text(s) = item {
2226 out.push_str(s);
2227 }
2228 }
2229 Value::Text(out)
2230 }
2231 // v7.17.0 — array_agg: collect into a typed array. NULL
2232 // elements are preserved per PG. Result type is decided
2233 // by the first non-NULL element seen (or Text fallback
2234 // when the whole group is NULL — PG would surface the
2235 // declared input type, but SPG hasn't yet wired the
2236 // aggregate's static input-type from `describe`).
2237 "array_agg" => {
2238 if st.items.is_empty() {
2239 return Value::Null;
2240 }
2241 let probe = st.items.iter().find(|v| !v.is_null());
2242 match probe.and_then(spg_storage::Value::data_type) {
2243 Some(DataType::Int) | Some(DataType::SmallInt) => {
2244 let items: Vec<Option<i32>> = st
2245 .items
2246 .iter()
2247 .map(|v| match v {
2248 Value::Int(n) => Some(*n),
2249 Value::SmallInt(n) => Some(i32::from(*n)),
2250 _ => None,
2251 })
2252 .collect();
2253 Value::IntArray(items)
2254 }
2255 Some(DataType::BigInt) => {
2256 let items: Vec<Option<i64>> = st
2257 .items
2258 .iter()
2259 .map(|v| match v {
2260 Value::BigInt(n) => Some(*n),
2261 _ => None,
2262 })
2263 .collect();
2264 Value::BigIntArray(items)
2265 }
2266 _ => {
2267 let items: Vec<Option<String>> = st
2268 .items
2269 .iter()
2270 .map(|v| match v {
2271 Value::Text(s) => Some(s.clone()),
2272 Value::Null => None,
2273 other => Some(format!("{other:?}")),
2274 })
2275 .collect();
2276 Value::TextArray(items)
2277 }
2278 }
2279 }
2280 // v7.17.0 — bool_and / bool_or finalize: lazy-init pattern
2281 // means `None` is exactly "empty group or all-NULL", which
2282 // PG surfaces as SQL NULL.
2283 "bool_and" | "bool_or" => st.bool_acc.map_or(Value::Null, Value::Bool),
2284 // v7.32 (round-29) — variance / stddev. PG: `variance` ==
2285 // `var_samp`, `stddev` == `stddev_samp`. samp needs n >= 2
2286 // (n < 2 → NULL); pop needs n >= 1 (n == 1 → 0).
2287 "variance" | "var_samp" | "var_pop" | "stddev" | "stddev_samp" | "stddev_pop" => {
2288 let n = st.count;
2289 if n == 0 {
2290 return Value::Null;
2291 }
2292 let nf = n as f64;
2293 // Sum of squared deviations from the mean.
2294 let ss = st.sum_sq - (st.sum_float * st.sum_float) / nf;
2295 let pop = name.ends_with("_pop");
2296 let denom = if pop { nf } else { nf - 1.0 };
2297 if denom <= 0.0 {
2298 // var_samp / stddev (samp) with n == 1 → NULL.
2299 return Value::Null;
2300 }
2301 let var = (ss / denom).max(0.0); // clamp fp noise below 0
2302 if name.starts_with("stddev") {
2303 Value::Float(crate::eval::f64_sqrt(var))
2304 } else {
2305 Value::Float(var)
2306 }
2307 }
2308 // v7.32 (round-29) — bitwise aggregates: None (empty / all-NULL)
2309 // → SQL NULL.
2310 "bit_and" | "bit_or" | "bit_xor" => st.bit_acc.map_or(Value::Null, Value::BigInt),
2311 // v7.32 (round-29) — regression family. `regr_count` is the
2312 // paired n; everything else is NULL over an empty set. Terms
2313 // are the mean-centred sums of squares / cross-products.
2314 "regr_count" => Value::BigInt(st.reg_n),
2315 "covar_pop" | "covar_samp" | "corr" | "regr_avgx" | "regr_avgy" | "regr_slope"
2316 | "regr_intercept" | "regr_r2" | "regr_sxx" | "regr_syy" | "regr_sxy" => {
2317 let n = st.reg_n;
2318 if n == 0 {
2319 return Value::Null;
2320 }
2321 let nf = n as f64;
2322 let sxx = st.reg_sxx - st.reg_sx * st.reg_sx / nf;
2323 let syy = st.reg_syy - st.reg_sy * st.reg_sy / nf;
2324 let sxy = st.reg_sxy - st.reg_sx * st.reg_sy / nf;
2325 let avgx = st.reg_sx / nf;
2326 let avgy = st.reg_sy / nf;
2327 let out = match name {
2328 "regr_avgx" => Some(avgx),
2329 "regr_avgy" => Some(avgy),
2330 "regr_sxx" => Some(sxx),
2331 "regr_syy" => Some(syy),
2332 "regr_sxy" => Some(sxy),
2333 "covar_pop" => Some(sxy / nf),
2334 "covar_samp" => (n >= 2).then(|| sxy / (nf - 1.0)),
2335 "regr_slope" => (sxx != 0.0).then(|| sxy / sxx),
2336 "regr_intercept" => (sxx != 0.0).then(|| avgy - (sxy / sxx) * avgx),
2337 "corr" => {
2338 let d = sxx * syy;
2339 (d > 0.0).then(|| sxy / crate::eval::f64_sqrt(d))
2340 }
2341 // PG: NULL when sxx==0; 1 when syy==0 (and sxx>0).
2342 "regr_r2" => {
2343 if sxx == 0.0 {
2344 None
2345 } else if syy == 0.0 {
2346 Some(1.0)
2347 } else {
2348 Some((sxy * sxy) / (sxx * syy))
2349 }
2350 }
2351 _ => None,
2352 };
2353 out.map_or(Value::Null, Value::Float)
2354 }
2355 // v7.32 (round-29) — json_agg / jsonb_agg: a JSON array of every
2356 // collected element in row order; empty set → SQL NULL.
2357 "json_agg" | "jsonb_agg" => {
2358 if st.items.is_empty() {
2359 return Value::Null;
2360 }
2361 let mut out = String::from("[");
2362 for (i, item) in st.items.iter().enumerate() {
2363 if i > 0 {
2364 out.push_str(", ");
2365 }
2366 out.push_str(&crate::json::value_to_json_text(item));
2367 }
2368 out.push(']');
2369 Value::Json(out)
2370 }
2371 // v7.32 (round-29) — json_object_agg: a JSON object built from
2372 // the parallel key (`items`) / value (`aux_items`) streams.
2373 "json_object_agg" | "jsonb_object_agg" => {
2374 if st.items.is_empty() {
2375 return Value::Null;
2376 }
2377 let mut out = String::from("{");
2378 for (i, key) in st.items.iter().enumerate() {
2379 if i > 0 {
2380 out.push_str(", ");
2381 }
2382 // Object keys are always JSON strings (PG coerces).
2383 let key_text = match key {
2384 Value::Text(s) | Value::Json(s) => s.clone(),
2385 other => crate::json::value_to_json_text(other),
2386 };
2387 out.push_str(&crate::json::value_to_json_text(&Value::Text(key_text)));
2388 out.push_str(": ");
2389 let val = st.aux_items.get(i).unwrap_or(&Value::Null);
2390 out.push_str(&crate::json::value_to_json_text(val));
2391 }
2392 out.push('}');
2393 Value::Json(out)
2394 }
2395 // Ordered-set aggregates are finalized in `run` (they need the
2396 // sorted items + the direct fraction argument), never here.
2397 _ => unreachable!(),
2398 }
2399}
2400
2401/// v7.32 (round-29) — numeric coercion for the percentile interpolation.
2402fn agg_value_to_f64(v: &Value) -> Option<f64> {
2403 match v {
2404 Value::Int(n) => Some(f64::from(*n)),
2405 Value::SmallInt(n) => Some(f64::from(*n)),
2406 Value::BigInt(n) => Some(*n as f64),
2407 Value::Float(x) => Some(*x),
2408 _ => None,
2409 }
2410}
2411
2412/// v7.32 (round-29) — finalize a WITHIN GROUP aggregate. `st.items` is
2413/// already sorted by the `WITHIN GROUP (ORDER BY …)` spec. `direct` is
2414/// the evaluated direct argument: the fraction for `percentile_*`, the
2415/// hypothetical value for the hypothetical-set family (`rank` etc.),
2416/// and unused by `mode`. `order` is the (single) sort key, needed by
2417/// the hypothetical-set family to compare in the sort direction.
2418#[allow(
2419 clippy::cast_precision_loss,
2420 clippy::cast_possible_truncation,
2421 clippy::cast_sign_loss
2422)]
2423fn finalize_ordered_set(
2424 name: &str,
2425 st: &AggState,
2426 direct: Option<&Value>,
2427 order: Option<&spg_sql::ast::OrderBy>,
2428) -> Value {
2429 let fraction = direct;
2430 let items = &st.items;
2431 if items.is_empty() {
2432 // A hypothetical row ranks first over an empty group; the
2433 // distribution functions are 0 / divide-by-(n+1).
2434 return match name {
2435 "rank" | "dense_rank" => Value::BigInt(1),
2436 "percent_rank" => Value::Float(0.0),
2437 "cume_dist" => Value::Float(1.0),
2438 _ => Value::Null,
2439 };
2440 }
2441 let n = items.len();
2442 match name {
2443 // v7.32 (round-29) — hypothetical-set: the rank the direct value
2444 // would have if inserted into the group, in the sort direction.
2445 "rank" | "dense_rank" | "percent_rank" | "cume_dist" => {
2446 let Some(h) = fraction else {
2447 return Value::Null;
2448 };
2449 let (desc, nulls_first) = order.map_or((false, None), |o| (o.desc, o.nulls_first));
2450 let mut before = 0usize; // sort strictly before h
2451 let mut before_or_eq = 0usize; // sort before-or-peer with h
2452 let mut distinct_before = 0usize;
2453 let mut last_before: Option<&Value> = None;
2454 for it in items {
2455 match crate::order_by_value_cmp(desc, nulls_first, it, h) {
2456 core::cmp::Ordering::Less => {
2457 before += 1;
2458 before_or_eq += 1;
2459 if last_before
2460 .is_none_or(|p| value_cmp(p, it) != core::cmp::Ordering::Equal)
2461 {
2462 distinct_before += 1;
2463 last_before = Some(it);
2464 }
2465 }
2466 core::cmp::Ordering::Equal => before_or_eq += 1,
2467 core::cmp::Ordering::Greater => {}
2468 }
2469 }
2470 let nn = n as f64;
2471 match name {
2472 "rank" => Value::BigInt((before + 1) as i64),
2473 "dense_rank" => Value::BigInt((distinct_before + 1) as i64),
2474 "percent_rank" => Value::Float(before as f64 / nn),
2475 "cume_dist" => Value::Float((before_or_eq as f64 + 1.0) / (nn + 1.0)),
2476 _ => unreachable!(),
2477 }
2478 }
2479 // Most frequent value; equal values are adjacent in the sorted
2480 // run, and a frequency tie resolves to the earliest run (the
2481 // smallest value under an ascending sort), matching PG.
2482 "mode" => {
2483 let (mut best_i, mut best_cnt) = (0usize, 1usize);
2484 let (mut run_i, mut run_cnt) = (0usize, 1usize);
2485 for i in 1..n {
2486 if value_cmp(&items[i], &items[run_i]) == core::cmp::Ordering::Equal {
2487 run_cnt += 1;
2488 } else {
2489 run_i = i;
2490 run_cnt = 1;
2491 }
2492 if run_cnt > best_cnt {
2493 best_cnt = run_cnt;
2494 best_i = run_i;
2495 }
2496 }
2497 items[best_i].clone()
2498 }
2499 // The first value whose cumulative fraction reaches `f`.
2500 "percentile_disc" => {
2501 let f = fraction
2502 .and_then(agg_value_to_f64)
2503 .unwrap_or(0.0)
2504 .clamp(0.0, 1.0);
2505 let idx = if f <= 0.0 {
2506 0
2507 } else {
2508 (crate::eval::f64_ceil(f * n as f64) as usize)
2509 .saturating_sub(1)
2510 .min(n - 1)
2511 };
2512 items[idx].clone()
2513 }
2514 // Linear interpolation between the two bracketing values.
2515 "percentile_cont" => {
2516 let f = fraction
2517 .and_then(agg_value_to_f64)
2518 .unwrap_or(0.0)
2519 .clamp(0.0, 1.0);
2520 let Some(nums) = items
2521 .iter()
2522 .map(agg_value_to_f64)
2523 .collect::<Option<Vec<f64>>>()
2524 else {
2525 return Value::Null; // non-numeric ordered set
2526 };
2527 if n == 1 {
2528 return Value::Float(nums[0]);
2529 }
2530 let rank = f * (n as f64 - 1.0);
2531 let lo = crate::eval::f64_floor(rank) as usize;
2532 let hi = crate::eval::f64_ceil(rank) as usize;
2533 let frac = rank - lo as f64;
2534 Value::Float(nums[lo] + (nums[hi] - nums[lo]) * frac)
2535 }
2536 _ => unreachable!(),
2537 }
2538}
2539
2540fn infer_agg_type(spec: &AggSpec, schema_cols: &[ColumnSchema]) -> DataType {
2541 // v7.26 (round-20 C) — the argument's statically-derived shape
2542 // types MIN/MAX/SUM/array_agg properly; RowDescription used to
2543 // report TEXT for these, breaking every sqlx typed decode.
2544 let arg_ty = spec
2545 .arg
2546 .as_ref()
2547 .and_then(|a| crate::describe::describe_expr(a, schema_cols))
2548 .map(|shape| shape.ty);
2549 // v7.33 (array_agg argmax) — `(array_agg(x ORDER BY y))[1]` yields the
2550 // ELEMENT type (x), not the array type.
2551 if spec.first_ordered {
2552 return arg_ty.unwrap_or(DataType::Text);
2553 }
2554 match spec.name.as_str() {
2555 "count" | "count_star" => DataType::BigInt,
2556 "sum" => match arg_ty {
2557 Some(DataType::Float) => DataType::Float,
2558 _ => DataType::BigInt,
2559 },
2560 "avg" => DataType::Float,
2561 // v7.17.0 — string_agg always returns TEXT.
2562 "string_agg" => DataType::Text,
2563 "array_agg" => match arg_ty {
2564 Some(DataType::Int | DataType::SmallInt) => DataType::IntArray,
2565 Some(DataType::BigInt) => DataType::BigIntArray,
2566 _ => DataType::TextArray,
2567 },
2568 // v7.17.0 — boolean aggregates always return BOOL (nullable
2569 // — empty / all-NULL group → NULL).
2570 "bool_and" | "bool_or" => DataType::Bool,
2571 // v7.32 (round-29) — variance / stddev are floating point;
2572 // percentile_cont interpolates to float; the regression family
2573 // (except regr_count) is floating point.
2574 "stddev" | "stddev_samp" | "stddev_pop" | "variance" | "var_samp" | "var_pop"
2575 | "percentile_cont" | "covar_pop" | "covar_samp" | "corr" | "regr_avgx" | "regr_avgy"
2576 | "regr_slope" | "regr_intercept" | "regr_r2" | "regr_sxx" | "regr_syy" | "regr_sxy" => {
2577 DataType::Float
2578 }
2579 // v7.32 (round-29) — bitwise aggregates, regr_count, and the
2580 // integer hypothetical-set ranks return an integer.
2581 "bit_and" | "bit_or" | "bit_xor" | "regr_count" | "rank" | "dense_rank" => DataType::BigInt,
2582 // v7.32 (round-29) — hypothetical-set distribution functions.
2583 "percent_rank" | "cume_dist" => DataType::Float,
2584 // v7.32 (round-29) — JSON aggregates return JSON.
2585 "json_agg" | "jsonb_agg" | "json_object_agg" | "jsonb_object_agg" => DataType::Json,
2586 // min/max, percentile_disc, mode, and anything pass-through:
2587 // the argument's shape (for ordered-set aggs `spec.arg` is the
2588 // WITHIN GROUP value expression).
2589 _ => arg_ty.unwrap_or(DataType::Text),
2590 }
2591}
2592
2593fn agg_or_group_type(e: &Expr, synth: &[ColumnSchema]) -> DataType {
2594 if let Expr::Column(c) = e
2595 && let Some(s) = synth.iter().find(|s| s.name == c.name)
2596 {
2597 return s.ty;
2598 }
2599 // v7.26 (round-20 C) — compound expressions over aggregates
2600 // (COALESCE(BOOL_OR(…), false), (array_agg(…))[1], CASE …)
2601 // derive their shape statically against the synth schema; the
2602 // old Text fallback broke sqlx typed decodes of exactly these
2603 // columns.
2604 crate::describe::describe_expr(e, synth)
2605 .map(|shape| shape.ty)
2606 .unwrap_or(DataType::Text)
2607}
2608
2609fn rewrite_expr(e: &Expr, group_exprs: &[Expr], aggs: &[AggSpec]) -> Expr {
2610 // v7.33 (array_agg argmax) — `(array_agg(x ORDER BY y))[1]` rewrites
2611 // to its first_ordered synth column, consuming the subscript. Checked
2612 // before the AggregateOrdered/recursion arms (which would otherwise
2613 // rewrite the inner array_agg and leave the subscript). Same matcher
2614 // as collect_aggregates, so the spec it finds is the one collected.
2615 if let Some((arg, order_by, filter)) = first_ordered_array_agg(e) {
2616 let arg_owned = Some(arg.clone());
2617 let filter_owned = filter.cloned();
2618 for (i, spec) in aggs.iter().enumerate() {
2619 if spec.first_ordered
2620 && spec.name == "array_agg"
2621 && spec.arg == arg_owned
2622 && spec.order_by == *order_by
2623 && spec.filter == filter_owned
2624 {
2625 return Expr::Column(spg_sql::ast::ColumnName {
2626 qualifier: None,
2627 name: format!("__agg_{i}"),
2628 });
2629 }
2630 }
2631 }
2632 // v7.24 (round-16 A) — ordered aggregate: match on the inner
2633 // call PLUS the ordering keys.
2634 if let Expr::AggregateOrdered {
2635 call,
2636 order_by,
2637 distinct,
2638 filter,
2639 } = e
2640 && let Expr::FunctionCall { name, args } = call.as_ref()
2641 {
2642 let lower = name.to_ascii_lowercase();
2643 if is_aggregate_name(&lower) {
2644 let canonical: &str = if lower == "every" { "bool_and" } else { &lower };
2645 // Mirror collect_aggregates: ordered-set aggregates take the
2646 // value from the sort spec and the in-parens arg as direct.
2647 let (arg, direct_arg) = if is_within_group_name(canonical) {
2648 (
2649 order_by.first().map(|o| o.expr.clone()),
2650 args.first().cloned(),
2651 )
2652 } else {
2653 (args.first().cloned(), None)
2654 };
2655 let arg2 = if agg_uses_second_arg(canonical) {
2656 args.get(1).cloned()
2657 } else {
2658 None
2659 };
2660 let filter_owned = filter.as_deref().cloned();
2661 for (i, spec) in aggs.iter().enumerate() {
2662 if spec.name == canonical
2663 && spec.arg == arg
2664 && spec.arg2 == arg2
2665 && spec.distinct == *distinct
2666 && spec.order_by == *order_by
2667 && spec.filter == filter_owned
2668 && spec.direct_arg == direct_arg
2669 {
2670 return Expr::Column(spg_sql::ast::ColumnName {
2671 qualifier: None,
2672 name: format!("__agg_{i}"),
2673 });
2674 }
2675 }
2676 }
2677 }
2678 // Match aggregate FunctionCalls first — they sit outside group_by.
2679 if let Expr::FunctionCall { name, args } = e {
2680 let lower = name.to_ascii_lowercase();
2681 if is_aggregate_name(&lower) {
2682 let arg = if lower == "count_star" {
2683 None
2684 } else {
2685 args.first().cloned()
2686 };
2687 // v7.17.0 — match the spec we registered for
2688 // string_agg(value, separator) on the full pair; v7.32 also
2689 // the regression family and json_object_agg.
2690 let arg2 = if agg_uses_second_arg(&lower) {
2691 args.get(1).cloned()
2692 } else {
2693 None
2694 };
2695 // v7.17.0 — `every` collapses into `bool_and` at
2696 // collection; mirror that here so the rewrite finds
2697 // the matching synth column.
2698 let canonical: &str = if lower == "every" {
2699 "bool_and"
2700 } else {
2701 lower.as_str()
2702 };
2703 for (i, spec) in aggs.iter().enumerate() {
2704 if spec.name == canonical
2705 && spec.arg == arg
2706 && spec.arg2 == arg2
2707 && !spec.distinct
2708 && spec.order_by.is_empty()
2709 {
2710 return Expr::Column(spg_sql::ast::ColumnName {
2711 qualifier: None,
2712 name: format!("__agg_{i}"),
2713 });
2714 }
2715 }
2716 }
2717 }
2718 // Match a group_by expression by AST equality.
2719 for (i, g) in group_exprs.iter().enumerate() {
2720 if g == e {
2721 return Expr::Column(spg_sql::ast::ColumnName {
2722 qualifier: None,
2723 name: format!("__grp_{i}"),
2724 });
2725 }
2726 }
2727 // Recurse into children.
2728 match e {
2729 Expr::AggregateOrdered {
2730 call,
2731 order_by,
2732 distinct,
2733 filter,
2734 } => Expr::AggregateOrdered {
2735 call: Box::new(rewrite_expr(call, group_exprs, aggs)),
2736 distinct: *distinct,
2737 order_by: order_by
2738 .iter()
2739 .map(|o| spg_sql::ast::OrderBy {
2740 expr: rewrite_expr(&o.expr, group_exprs, aggs),
2741 desc: o.desc,
2742 nulls_first: o.nulls_first,
2743 })
2744 .collect(),
2745 // The filter is evaluated against SOURCE rows during
2746 // accumulation, never against synth rows — keep it as-is.
2747 filter: filter.clone(),
2748 },
2749 Expr::Binary { lhs, op, rhs } => Expr::Binary {
2750 lhs: Box::new(rewrite_expr(lhs, group_exprs, aggs)),
2751 op: *op,
2752 rhs: Box::new(rewrite_expr(rhs, group_exprs, aggs)),
2753 },
2754 Expr::Unary { op, expr } => Expr::Unary {
2755 op: *op,
2756 expr: Box::new(rewrite_expr(expr, group_exprs, aggs)),
2757 },
2758 Expr::Cast { expr, target } => Expr::Cast {
2759 expr: Box::new(rewrite_expr(expr, group_exprs, aggs)),
2760 target: *target,
2761 },
2762 Expr::IsNull { expr, negated } => Expr::IsNull {
2763 expr: Box::new(rewrite_expr(expr, group_exprs, aggs)),
2764 negated: *negated,
2765 },
2766 Expr::FunctionCall { name, args } => Expr::FunctionCall {
2767 name: name.clone(),
2768 args: args
2769 .iter()
2770 .map(|a| rewrite_expr(a, group_exprs, aggs))
2771 .collect(),
2772 },
2773 Expr::Like {
2774 expr,
2775 pattern,
2776 negated,
2777 case_insensitive,
2778 } => Expr::Like {
2779 expr: Box::new(rewrite_expr(expr, group_exprs, aggs)),
2780 pattern: Box::new(rewrite_expr(pattern, group_exprs, aggs)),
2781 negated: *negated,
2782 case_insensitive: *case_insensitive,
2783 },
2784 Expr::Extract { field, source } => Expr::Extract {
2785 field: *field,
2786 source: Box::new(rewrite_expr(source, group_exprs, aggs)),
2787 },
2788 // v7.25.2 (round-19 A) — subquery nodes: rewrite group-key
2789 // references INSIDE the body to `__grp_N` so the correlated
2790 // resolver can substitute them against the synthesised group
2791 // row (aggs are NOT matched inside the body — a COUNT in the
2792 // subquery is the subquery's own aggregate).
2793 Expr::ScalarSubquery(s) => {
2794 Expr::ScalarSubquery(Box::new(rewrite_group_keys_in_select(s, group_exprs)))
2795 }
2796 Expr::Exists { subquery, negated } => Expr::Exists {
2797 subquery: Box::new(rewrite_group_keys_in_select(subquery, group_exprs)),
2798 negated: *negated,
2799 },
2800 Expr::InSubquery {
2801 expr,
2802 subquery,
2803 negated,
2804 } => Expr::InSubquery {
2805 expr: Box::new(rewrite_expr(expr, group_exprs, aggs)),
2806 subquery: Box::new(rewrite_group_keys_in_select(subquery, group_exprs)),
2807 negated: *negated,
2808 },
2809 // v4.12 window / Literal / Column — clone-pass (these don't
2810 // participate in aggregate rewrite).
2811 Expr::WindowFunction { .. } | Expr::Literal(_) | Expr::Placeholder(_) | Expr::Column(_) => {
2812 e.clone()
2813 }
2814 // v7.10.10 — recurse children for array nodes.
2815 Expr::Array(items) => Expr::Array(
2816 items
2817 .iter()
2818 .map(|elem| rewrite_expr(elem, group_exprs, aggs))
2819 .collect(),
2820 ),
2821 Expr::ArraySubscript { target, index } => Expr::ArraySubscript {
2822 target: Box::new(rewrite_expr(target, group_exprs, aggs)),
2823 index: Box::new(rewrite_expr(index, group_exprs, aggs)),
2824 },
2825 Expr::AnyAll {
2826 expr,
2827 op,
2828 array,
2829 is_any,
2830 } => Expr::AnyAll {
2831 expr: Box::new(rewrite_expr(expr, group_exprs, aggs)),
2832 op: *op,
2833 array: Box::new(rewrite_expr(array, group_exprs, aggs)),
2834 is_any: *is_any,
2835 },
2836 Expr::InList {
2837 expr,
2838 list,
2839 negated,
2840 } => Expr::InList {
2841 expr: Box::new(rewrite_expr(expr, group_exprs, aggs)),
2842 list: list
2843 .iter()
2844 .map(|item| rewrite_expr(item, group_exprs, aggs))
2845 .collect(),
2846 negated: *negated,
2847 },
2848 Expr::Case {
2849 operand,
2850 branches,
2851 else_branch,
2852 } => Expr::Case {
2853 operand: operand
2854 .as_deref()
2855 .map(|o| Box::new(rewrite_expr(o, group_exprs, aggs))),
2856 branches: branches
2857 .iter()
2858 .map(|(w, t)| {
2859 (
2860 rewrite_expr(w, group_exprs, aggs),
2861 rewrite_expr(t, group_exprs, aggs),
2862 )
2863 })
2864 .collect(),
2865 else_branch: else_branch
2866 .as_deref()
2867 .map(|e| Box::new(rewrite_expr(e, group_exprs, aggs))),
2868 },
2869 }
2870}
2871
2872/// v7.25.2 (round-19 A) — rewrite group-key references inside a
2873/// subquery body to `__grp_N` synthetic columns (aggregates are
2874/// not touched: empty spec list). Runs through the canonical
2875/// Select walker so every expression slot is covered.
2876fn rewrite_group_keys_in_select(
2877 s: &spg_sql::ast::SelectStatement,
2878 group_exprs: &[Expr],
2879) -> spg_sql::ast::SelectStatement {
2880 let mut out = s.clone();
2881 let _ = crate::walk_select_exprs_mut(&mut out, &mut |e| {
2882 *e = rewrite_expr(e, group_exprs, &[]);
2883 Ok(())
2884 });
2885 out
2886}
2887
2888/// Canonical string key for a tuple of group values. Used as map key.
2889/// Per-value group-key encoding (shared by owned and borrowed paths).
2890fn encode_one(out: &mut String, v: &Value) {
2891 use core::fmt::Write;
2892 match v {
2893 Value::Null => out.push_str("N|"),
2894 // v7.36 (perf — mailrs Phase 1) — switch the integer / float
2895 // encoders to `write!`. `n.to_string()` allocates a fresh
2896 // `String` per cell just to push its bytes into the
2897 // (already-cleared) reuse buffer — for the 25 k-row JOIN
2898 // probe in `count_messages` that's 25 k heap allocs per
2899 // query. `write!(&mut String, ...)` formats straight into
2900 // the buffer; no intermediate alloc.
2901 Value::SmallInt(n) => {
2902 let _ = write!(out, "s{n}|");
2903 }
2904 Value::Int(n) => {
2905 let _ = write!(out, "I{n}|");
2906 }
2907 Value::BigInt(n) => {
2908 let _ = write!(out, "B{n}|");
2909 }
2910 Value::Float(x) => {
2911 let _ = write!(out, "F{x}|");
2912 }
2913 Value::Bool(b) => {
2914 out.push(if *b { 'T' } else { 'f' });
2915 out.push('|');
2916 }
2917 Value::Text(s) => {
2918 out.push('S');
2919 out.push_str(s);
2920 out.push('|');
2921 }
2922 Value::Vector(v) => {
2923 out.push('V');
2924 for x in v {
2925 out.push_str(&x.to_string());
2926 out.push(',');
2927 }
2928 out.push('|');
2929 }
2930 // v6.0.1: GROUP BY on a `VECTOR(N) USING SQ8` column.
2931 // Two cells with byte-identical `(min, max, bytes)`
2932 // share the same group; equivalence is byte-equality
2933 // (same as f32 grouping today — neither path tries to
2934 // normalise nan/-0).
2935 Value::Sq8Vector(q) => {
2936 out.push('Q');
2937 out.push_str(&q.min.to_string());
2938 out.push('@');
2939 out.push_str(&q.max.to_string());
2940 out.push(':');
2941 for b in &q.bytes {
2942 out.push_str(&b.to_string());
2943 out.push(',');
2944 }
2945 out.push('|');
2946 }
2947 // v6.0.3: GROUP BY on a `VECTOR(N) USING HALF` column.
2948 // Byte-equality over the raw u16 bits; matches the SQ8
2949 // path's byte-key model.
2950 Value::HalfVector(h) => {
2951 out.push('H');
2952 for b in &h.bytes {
2953 out.push_str(&b.to_string());
2954 out.push(',');
2955 }
2956 out.push('|');
2957 }
2958 Value::Numeric { scaled, scale } => {
2959 out.push('D');
2960 out.push_str(&scaled.to_string());
2961 out.push('@');
2962 out.push_str(&scale.to_string());
2963 out.push('|');
2964 }
2965 Value::Date(d) => {
2966 out.push('d');
2967 out.push_str(&d.to_string());
2968 out.push('|');
2969 }
2970 Value::Timestamp(t) => {
2971 out.push('t');
2972 out.push_str(&t.to_string());
2973 out.push('|');
2974 }
2975 Value::Interval { months, micros } => {
2976 out.push('i');
2977 out.push_str(&months.to_string());
2978 out.push('m');
2979 out.push_str(µs.to_string());
2980 out.push('|');
2981 }
2982 Value::Json(s) => {
2983 out.push('j');
2984 out.push_str(s);
2985 out.push('|');
2986 }
2987 // v7.5.0 — Value is #[non_exhaustive] for downstream
2988 // forward-compat. Any future variant lacking explicit
2989 // handling here will share a debug-derived group key,
2990 // which is observably wrong but won't crash.
2991 _ => {
2992 out.push('?');
2993 out.push_str(&format!("{v:?}"));
2994 out.push('|');
2995 }
2996 }
2997}
2998
2999/// v7.30 (perf campaign) - encode from borrowed cells without
3000/// materialising an owned Vec<Value> first.
3001pub(crate) fn encode_key_refs(vals: &[&Value]) -> String {
3002 let mut out = String::new();
3003 for v in vals {
3004 encode_one(&mut out, v);
3005 }
3006 out
3007}
3008
3009/// v7.31 (perf 3e) — encode into a caller-owned scratch buffer.
3010/// The per-row key paths (group hash, DISTINCT set, join build/
3011/// probe) ran 24k+ String allocations per query through the
3012/// allocator just to LOOK UP a map; the scratch form allocates
3013/// only when a map actually has to take ownership (vacant insert).
3014pub(crate) fn encode_key_refs_into(vals: &[&Value], out: &mut String) {
3015 out.clear();
3016 for v in vals {
3017 encode_one(out, v);
3018 }
3019}
3020
3021pub(crate) fn encode_key(vals: &[Value]) -> String {
3022 let mut out = String::new();
3023 for v in vals {
3024 encode_one(&mut out, v);
3025 }
3026 out
3027}
3028
3029#[allow(clippy::cast_precision_loss)]
3030fn value_cmp(a: &Value, b: &Value) -> core::cmp::Ordering {
3031 use core::cmp::Ordering::Equal;
3032 match (a, b) {
3033 (Value::Null, Value::Null) => Equal,
3034 (Value::Null, _) => core::cmp::Ordering::Greater, // NULLs last
3035 (_, Value::Null) => core::cmp::Ordering::Less,
3036 (Value::Int(x), Value::Int(y)) => x.cmp(y),
3037 (Value::BigInt(x), Value::BigInt(y)) => x.cmp(y),
3038 (Value::Int(x), Value::BigInt(y)) => i64::from(*x).cmp(y),
3039 (Value::BigInt(x), Value::Int(y)) => x.cmp(&i64::from(*y)),
3040 (Value::Float(x), Value::Float(y)) => x.partial_cmp(y).unwrap_or(Equal),
3041 (Value::Int(x), Value::Float(y)) => f64::from(*x).partial_cmp(y).unwrap_or(Equal),
3042 (Value::Float(x), Value::Int(y)) => x.partial_cmp(&f64::from(*y)).unwrap_or(Equal),
3043 (Value::BigInt(x), Value::Float(y)) => (*x as f64).partial_cmp(y).unwrap_or(Equal),
3044 (Value::Float(x), Value::BigInt(y)) => x.partial_cmp(&(*y as f64)).unwrap_or(Equal),
3045 (Value::Text(x), Value::Text(y)) => x.cmp(y),
3046 (Value::Bool(x), Value::Bool(y)) => x.cmp(y),
3047 _ => Equal,
3048 }
3049}