fix(query): reject STR equality keys in asof-join and window-join

src/ops/query.c

@@ -14876,6 +14876,21 @@ static ray_t* window_join_impl(ray_t** args, int64_t n, int mode) {
                 for (int i = 0; i < 4; i++) ray_release(eargs[i]);
                 return ray_error("domain", "window-join: equality key column not found in both tables");
             }
+            /* STR equality-key columns are silently mismatched: window-join
+             * sorts and probes eq cells through read_col_i64, which has no
+             * RAY_STR case (it reads one raw byte per row), so once there are
+             * enough distinct keys the byte collisions cross-contaminate
+             * groups and the aggregates are WRONG.  Decline both sides — the
+             * base equi-join has a STR-aware kernel, these window kernels do
+             * not. */
+            int8_t lct = left_eq[e]->type, rct = right_eq[e]->type;
+            if (RAY_IS_PARTED(lct)) lct = (int8_t)RAY_PARTED_BASETYPE(lct);
+            if (RAY_IS_PARTED(rct)) rct = (int8_t)RAY_PARTED_BASETYPE(rct);
+            if (lct == RAY_STR || rct == RAY_STR) {
+                scratch_free(eq_hdr);
+                for (int i = 0; i < 4; i++) ray_release(eargs[i]);
+                return ray_error("nyi", "window-join: string equality key columns are not supported");
+            }
         }
 
         /* Parse every (name, (op src)) pair from the agg dict.  The dict's
@@ -15436,6 +15451,18 @@ static ray_t* window_join_impl(ray_t** args, int64_t n, int mode) {
         if (!nm) { scratch_free(eqops_hdr); ray_graph_free(g); if (_bxeq) ray_release(_bxeq); return ray_error("domain", "window-join: unknown equality key symbol"); }
         eq_ops[i] = ray_scan(g, ray_str_ptr(nm));
         if (!eq_ops[i]) { scratch_free(eqops_hdr); ray_graph_free(g); if (_bxeq) ray_release(_bxeq); return ray_error("domain", "window-join: equality key column not found"); }
+        ray_t* lcol = ray_table_get_col(left_tbl, eq_elems[i]->i64);
+        ray_t* rcol = ray_table_get_col(right_tbl, eq_elems[i]->i64);
+        int8_t lct = lcol ? lcol->type : (int8_t)0;
+        int8_t rct = rcol ? rcol->type : (int8_t)0;
+        if (RAY_IS_PARTED(lct)) lct = (int8_t)RAY_PARTED_BASETYPE(lct);
+        if (RAY_IS_PARTED(rct)) rct = (int8_t)RAY_PARTED_BASETYPE(rct);
+        if (lct == RAY_STR || rct == RAY_STR) {
+            scratch_free(eqops_hdr);
+            ray_graph_free(g);
+            if (_bxeq) ray_release(_bxeq);
+            return ray_error("nyi", "window-join: string equality key columns are not supported");
+        }
     }
 
     if (_bxeq) ray_release(_bxeq);
@@ -15546,6 +15573,23 @@ static ray_t* ray_asof_join_core(ray_t* keys_vec, ray_t* left_tbl, ray_t* right_
         if (!nm) { scratch_free(eqops_hdr); ray_graph_free(g); if (_bxk) ray_release(_bxk); return ray_error("domain", "asof-join: unknown equality key symbol"); }
         eq_ops[i] = ray_scan(g, ray_str_ptr(nm));
         if (!eq_ops[i]) { scratch_free(eqops_hdr); ray_graph_free(g); if (_bxk) ray_release(_bxk); return ray_error("domain", "asof-join: equality key column not found"); }
+        /* STR equality-key columns are silently mismatched: the asof kernel
+         * reads eq cells through read_col_i64 (asof_eq_lread), which has no
+         * RAY_STR case, so string keys compare as raw bytes and yield WRONG
+         * results (a key that should match can null out).  The base equi-join
+         * has a separate STR-aware kernel; asof does not, so decline both
+         * sides rather than return corrupt data. */
+        ray_t* lcol = ray_table_get_col(left_tbl, eq_syms[i]->i64);
+        ray_t* rcol = ray_table_get_col(right_tbl, eq_syms[i]->i64);
+        int8_t lct = lcol ? lcol->type : (int8_t)0;
+        int8_t rct = rcol ? rcol->type : (int8_t)0;
+        if (RAY_IS_PARTED(lct)) lct = (int8_t)RAY_PARTED_BASETYPE(lct);
+        if (RAY_IS_PARTED(rct)) rct = (int8_t)RAY_PARTED_BASETYPE(rct);
+        if (lct == RAY_STR || rct == RAY_STR) {
+            scratch_free(eqops_hdr);
+            ray_graph_free(g); if (_bxk) ray_release(_bxk);
+            return ray_error("nyi", "asof-join: string equality key columns are not supported");
+        }
     }
 
     if (_bxk) ray_release(_bxk);

test/rfl/join/str_key_nyi.rfl

@@ -0,0 +1,33 @@
+;; str_key_nyi.rfl — STR equality-key columns are unsupported in the asof and
+;; window join kernels and must be DECLINED, not silently mis-answered.
+;;
+;; Root cause: both kernels read equality-key cells through read_col_i64
+;; (asof via asof_eq_lread; window when it sorts/probes the right side), and
+;; read_col_i64 has no RAY_STR case — a STR column falls into the byte-wide
+;; default.  asof then mismatches outright (a key that should match nulls
+;; out); window only reads one byte per row, so small key sets coincide by
+;; luck but enough distinct keys collide and cross-contaminate groups, so the
+;; aggregates are WRONG.  The base equi-join (inner/left/anti) uses a separate
+;; STR-aware kernel and stays correct, so the fix is scoped to asof + window:
+;; reject a STR eq-key on either side with `nyi` instead of returning corrupt
+;; data.  window-join1 shares the window kernel; the 4-argument compatibility
+;; window forms fall through to the asof executor, so they need the same guard.
+
+;; ── asof-join: STR equality key → nyi (both operand orders) ──────────
+(set tr (table [Sym Time Price] (list ["A" "A"] [12:00:01 12:00:04] [100 101])))(set qs (table [Sym Time Bid] (list ["A" "A"] [12:00:00 12:00:05] [99 100])))(asof-join [Sym Time] tr qs) !- nyi
+(set tr (table [Sym Time Price] (list ["A" "A"] [12:00:01 12:00:04] [100 101])))(set qs (table [Sym Time Bid] (list ["A" "A"] [12:00:00 12:00:05] [99 100])))(asof-join [Sym Time] qs tr) !- nyi
+
+;; ── window-join / window-join1: STR equality key → nyi ───────────────
+(set wt (table [Sym Time Price] (list ["a" "a"] [10:00:01.000 10:00:05.000] [100 200])))(set wq (table [Sym Time Bid] (list ["a" "a" "a"] [10:00:00.000 10:00:02.000 10:00:04.000] [99 100 101])))(set wi (map-left + [-2000 2000] (at wt 'Time)))(window-join [Sym Time] wi wt wq {minBid: (min Bid)}) !- nyi
+(set wt (table [Sym Time Price] (list ["a" "a"] [10:00:01.000 10:00:05.000] [100 200])))(set wq (table [Sym Time Bid] (list ["a" "a" "a"] [10:00:00.000 10:00:02.000 10:00:04.000] [99 100 101])))(set wi (map-left + [-2000 2000] (at wt 'Time)))(window-join1 [Sym Time] wi wt wq {minBid: (min Bid)}) !- nyi
+(set wt (table [Sym Time Price] (list ["a" "a"] [10:00:01.000 10:00:05.000] [100 200])))(set wq (table [Sym Time Bid] (list ["a" "a" "a"] [10:00:00.000 10:00:02.000 10:00:04.000] [99 100 101])))(at (window-join wt wq [Sym] 'Time) 'Bid) !- nyi
+(set wt (table [Sym Time Price] (list ["a" "a"] [10:00:01.000 10:00:05.000] [100 200])))(set wq (table [Sym Time Bid] (list ["a" "a" "a"] [10:00:00.000 10:00:02.000 10:00:04.000] [99 100 101])))(at (window-join1 wt wq [Sym] 'Time) 'Bid) !- nyi
+
+;; ── SYM equality keys keep working and stay order-independent ────────
+(set KS (as 'sym ["AAPL" "AAPL"]))(set trS (table [Sym Time Price] (list KS [12:00:01 12:00:04] [100 101])))(set qsS (table [Sym Time Bid] (list KS [12:00:00 12:00:05] [99 100])))(at (asof-join [Sym Time] trS qsS) 'Bid) -- [99 99]
+(set KS (as 'sym ["AAPL" "AAPL"]))(set trS (table [Sym Time Price] (list KS [12:00:01 12:00:04] [100 101])))(set qsU (table [Sym Time Bid] (list KS [12:00:05 12:00:00] [100 99])))(at (asof-join [Sym Time] trS qsU) 'Bid) -- [99 99]
+(set KW (as 'sym ["a" "a"]))(set wtS (table [Sym Time Price] (list KW [10:00:01.000 10:00:05.000] [100 200])))(set wqS (table [Sym Time Bid] (list (as 'sym ["a" "a" "a"]) [10:00:00.000 10:00:02.000 10:00:04.000] [99 100 101])))(set wiS (map-left + [-2000 2000] (at wtS 'Time)))(at (window-join [Sym Time] wiS wtS wqS {minBid: (min Bid)}) 'minBid) -- [99 100]
+
+;; ── base equi-join must NOT over-reject: STR keys stay correct ───────
+(set A (table [K V] (list ["x" "y"] [1 2])))(set B (table [K W] (list ["x" "z"] [10 20])))(at (inner-join [K] A B) 'V) -- [1]
+(set A (table [K V] (list ["x" "y"] [1 2])))(set B (table [K W] (list ["x" "z"] [10 20])))(at (anti-join [K] A B) 'V) -- [2]