parser: support pg19 null treatment in aggregates

crates/squawk_parser/src/grammar.rs

@@ -2165,6 +2165,7 @@ fn call_expr_args(p: &mut Parser<'_>, lhs: CompletedMarker) -> CompletedMarker {
 fn opt_agg_clauses(p: &mut Parser<'_>) {
     opt_within_clause(p);
     opt_filter_clause(p);
+    opt_null_treatment(p);
     opt_over_clause(p);
 }
 
@@ -2181,13 +2182,10 @@ fn opt_filter_clause(p: &mut Parser<'_>) {
 }
 
 fn opt_over_clause(p: &mut Parser<'_>) {
-    if p.at(OVER_KW) || p.at(RESPECT_KW) || p.at(IGNORE_KW) {
+    if p.at(OVER_KW) {
         // OVER window_name
         // OVER ( window_definition )
         let m = p.start();
-        if p.eat(RESPECT_KW) || p.eat(IGNORE_KW) {
-            p.expect(NULLS_KW);
-        }
         p.expect(OVER_KW);
         if p.eat(L_PAREN) {
             window_spec(p);
@@ -2199,6 +2197,22 @@ fn opt_over_clause(p: &mut Parser<'_>) {
     }
 }
 
+fn opt_null_treatment(p: &mut Parser<'_>) {
+    if !p.at(RESPECT_KW) && !p.at(IGNORE_KW) {
+        return;
+    }
+
+    let m = p.start();
+    let kind = if p.eat(RESPECT_KW) {
+        RESPECT_NULLS
+    } else {
+        p.bump(IGNORE_KW);
+        IGNORE_NULLS
+    };
+    p.expect(NULLS_KW);
+    m.complete(p, kind);
+}
+
 fn opt_within_clause(p: &mut Parser<'_>) {
     if p.at(WITHIN_KW) {
         let m = p.start();

crates/squawk_syntax/src/postgresql.ungram

@@ -65,7 +65,7 @@ ArgList =
 | '(' ('distinct' | 'all')? args:(Arg (',' Arg)*)? ')'
 
 CallExpr =
-  Expr ArgList WithinClause? FilterClause? OverClause?
+  Expr ArgList WithinClause? FilterClause? NullTreatment? OverClause?
 | ExtractFn
 | GraphTableFn
 | JsonExistsFn
@@ -1728,7 +1728,17 @@ FilterClause =
   'filter' '(' 'where' Expr ')'
 
 OverClause =
-  (('respect' | 'ignore') 'nulls')? 'over' (NameRef | '(' WindowSpec? ')')
+  'over' (NameRef | '(' WindowSpec? ')')
+
+NullTreatment =
+  IgnoreNulls
+| RespectNulls
+
+IgnoreNulls =
+  'ignore' 'nulls'
+
+RespectNulls =
+  'respect' 'nulls'
 
 WithinClause =
   'within' 'group' '(' OrderByClause ')'

postgres/kwlist.h

@@ -1,7 +1,7 @@
 // synced from:
-//   commit: 2c4bd2bf5700db98be0602854a8b7fa2c16b5f4a
-//   committed at: 2026-05-23T09:39:58+09:00
-//   file: https://github.com/postgres/postgres/blob/2c4bd2bf5700db98be0602854a8b7fa2c16b5f4a/src/include/parser/kwlist.h
+//   commit: 6468f7a853c3c75066410cfb54ecdb3050ec7132
+//   committed at: 2026-06-25T14:25:57-07:00
+//   file: https://github.com/postgres/postgres/blob/6468f7a853c3c75066410cfb54ecdb3050ec7132/src/include/parser/kwlist.h
 //
 // update via:
 //   cargo xtask sync-pg

postgres/regression_suite/aggregates.sql

@@ -141,22 +141,33 @@ SELECT covar_pop(1::float8,'inf'::float8), covar_samp(3::float8,'inf'::float8);
 SELECT covar_pop(1::float8,'nan'::float8), covar_samp(3::float8,'nan'::float8);
 
 -- check some cases that formerly had poor roundoff-error behavior
+-- note: regr_r2() differs from corr() for a horizontal line, per spec
 SELECT corr(0.09, g), regr_r2(0.09, g)
   FROM generate_series(1, 30) g;
 SELECT corr(g, 0.09), regr_r2(g, 0.09), regr_slope(g, 0.09), regr_intercept(g, 0.09)
   FROM generate_series(1, 30) g;
 SELECT corr(1.3 + g * 1e-16, 1.3 + g * 1e-16)
   FROM generate_series(1, 3) g;
-SELECT corr(1e-100 + g * 1e-105, 1e-100 + g * 1e-105)
+
+-- check some cases that formerly suffered from internal overflow/underflow
+SELECT corr(1e-100 + g * 1e-105, 1e-100 + g * 1e-105),
+       regr_r2(1e-100 + g * 1e-105, 1e-100 + g * 1e-105)
   FROM generate_series(1, 3) g;
-SELECT corr(1e-100 + g * 1e-105, 1e-100 + g * 1e-105)
+SELECT corr(1e-100 + g * 1e-105, 1e-100 + g * 1e-105),
+       regr_r2(1e-100 + g * 1e-105, 1e-100 + g * 1e-105)
   FROM generate_series(1, 30) g;
+SELECT corr(1e100 + g * 1e95, 1e100 + g * 1e95),
+       regr_r2(1e100 + g * 1e95, 1e100 + g * 1e95)
+  FROM generate_series(1, 2) g;
+SELECT regr_intercept(y, x) FROM (VALUES (-1e150, 0), (2e150, 3e150)) v(x, y);
+SELECT regr_intercept(y, x) FROM (VALUES (-1e-131, 0), (2e-131, 3e-131)) v(x, y);
 
 -- these examples pose definitional questions for NaN inputs,
 -- which we resolve by saying that an all-NaN input column is not all equal
-SELECT corr(g, 'NaN') FROM generate_series(1, 30) g;
-SELECT corr(0.1, 'NaN') FROM generate_series(1, 30) g;
-SELECT corr('NaN', 'NaN') FROM generate_series(1, 30) g;
+SELECT corr(g, 'NaN'), regr_r2(g, 'NaN') FROM generate_series(1, 30) g;
+SELECT corr(0.1, 'NaN'), regr_r2(0.1, 'NaN') FROM generate_series(1, 30) g;
+SELECT corr('NaN', 0.1), regr_r2('NaN', 0.1) FROM generate_series(1, 30) g;
+SELECT corr('NaN', 'NaN'), regr_r2('NaN', 'NaN') FROM generate_series(1, 30) g;
 
 -- test accum and combine functions directly
 CREATE TABLE regr_test (x float8, y float8);