Merge pull request #8 from ser-vasilich/fixes/seven-bug-probes

fix(8 bugs) + test(coverage +1.3pp): probes, type-preservation, broad workout

Author: singaraiona

.gitignore

@@ -19,3 +19,15 @@ rf_test_*.csv
 
 CLAUDE.md
 docs/plans/
+
+# IDE state
+.idea/
+.vscode/
+
+# gcov / lcov artifacts
+*.gcda
+*.gcno
+*.gcov
+coverage*.info
+coverage_html/
+rayforce.cov

src/app/repl.h

@@ -40,6 +40,13 @@ typedef struct ray_repl {
 ray_repl_t* ray_repl_create(ray_poll_t* poll);
 void       ray_repl_destroy(ray_repl_t* repl);
 void       ray_repl_run(ray_repl_t* repl);
+
+/* Run a Rayfall script file in batch (script) mode.  Contract:
+ *   - returns 0 on success
+ *   - returns 1 on any eval error (script execution stops at first
+ *     error; subsequent forms are not run)
+ * Distinct from ray_repl_run / stdin pipe which use REPL semantics
+ * (errors are printed but do not terminate the loop). */
 int        ray_repl_run_file(const char* path);
 
 #endif /* RAY_IO_REPL_H */

src/lang/eval.c

@@ -740,6 +740,8 @@ ray_t* call_fn1(ray_t* fn, ray_t* arg) {
     if (fn_is_restricted(fn)) return ray_error("access", "restricted");
     if (fn->type == RAY_UNARY) {
         ray_unary_fn f = (ray_unary_fn)(uintptr_t)fn->i64;
+        if ((fn->attrs & RAY_FN_ATOMIC) && is_collection(arg))
+            return atomic_map_unary(f, arg);
         return f(arg);
     }
     if (fn->type == RAY_LAMBDA) {
@@ -1661,18 +1663,16 @@ op_callf: {
         switch (fn_obj->type) {
         case RAY_UNARY:
             if (fn_is_restricted(fn_obj)) { for (int32_t i = 0; i < n; i++) ray_release(fn_args[i]); result = ray_error("access", "restricted"); break; }
-            if (n < 1) { result = ray_error("arity", "expected 1 arg, got 0"); break; }
+            if (n != 1) { for (int32_t i = 0; i < n; i++) ray_release(fn_args[i]); result = ray_error("arity", "expected 1 arg, got %d", n); break; }
             result = ((ray_unary_fn)(uintptr_t)fn_obj->i64)(fn_args[0]);
             ray_release(fn_args[0]);
-            for (int32_t i = 1; i < n; i++) ray_release(fn_args[i]);
             break;
         case RAY_BINARY:
             if (fn_is_restricted(fn_obj)) { for (int32_t i = 0; i < n; i++) ray_release(fn_args[i]); result = ray_error("access", "restricted"); break; }
-            if (n < 2) { for (int32_t i = 0; i < n; i++) ray_release(fn_args[i]); result = ray_error("arity", "expected 2 args, got %d", n); break; }
+            if (n != 2) { for (int32_t i = 0; i < n; i++) ray_release(fn_args[i]); result = ray_error("arity", "expected 2 args, got %d", n); break; }
             result = ((ray_binary_fn)(uintptr_t)fn_obj->i64)(fn_args[0], fn_args[1]);
             ray_release(fn_args[0]);
             ray_release(fn_args[1]);
-            for (int32_t i = 2; i < n; i++) ray_release(fn_args[i]);
             break;
         case RAY_VARY:
             if (fn_is_restricted(fn_obj)) { for (int32_t i = 0; i < n; i++) ray_release(fn_args[i]); result = ray_error("access", "restricted"); break; }
@@ -2021,8 +2021,8 @@ static void ray_register_builtins(void) {
     register_binary_op("<=",  RAY_FN_ATOMIC, ray_lte_fn,    OP_LE);
     register_binary_op("==",  RAY_FN_ATOMIC, ray_eq_fn,  OP_EQ);
     register_binary_op("!=",  RAY_FN_ATOMIC, ray_neq_fn,    OP_NE);
-    register_binary_op("and", RAY_FN_NONE,   ray_and_fn, OP_AND);
-    register_binary_op("or",  RAY_FN_NONE,   ray_or_fn,  OP_OR);
+    register_vary("and", RAY_FN_NONE, ray_and_vary_fn);
+    register_vary("or",  RAY_FN_NONE, ray_or_vary_fn);
     register_unary_op("not",  RAY_FN_NONE,   ray_not_fn, OP_NOT);
     register_unary_op("neg",  RAY_FN_ATOMIC, ray_neg_fn, OP_NEG);
     register_unary("round",   RAY_FN_ATOMIC, ray_round_fn);
@@ -2392,7 +2392,7 @@ ray_t* ray_eval(ray_t* obj) {
 
     switch (head->type) {
         case RAY_UNARY: {
-            if (n < 2) { ray_release(head); ret = ray_error("domain", NULL); goto out; }
+            if (n != 2) { ray_release(head); ret = ray_error("arity", "expected 1 arg, got %d", (int)(n-1)); goto out; }
             if (fn_is_restricted(head)) { ray_release(head); ret = ray_error("access", "restricted"); goto out; }
             ray_unary_fn fn = (ray_unary_fn)(uintptr_t)head->i64;
             uint8_t fn_attrs = head->attrs;
@@ -2412,7 +2412,7 @@ ray_t* ray_eval(ray_t* obj) {
             ret = result; goto out;
         }
         case RAY_BINARY: {
-            if (n < 3) { ray_release(head); ret = ray_error("domain", NULL); goto out; }
+            if (n != 3) { ray_release(head); ret = ray_error("arity", "expected 2 args, got %d", (int)(n-1)); goto out; }
             if (fn_is_restricted(head)) { ray_release(head); ret = ray_error("access", "restricted"); goto out; }
             ray_binary_fn fn = (ray_binary_fn)(uintptr_t)head->i64;
             uint8_t fn_attrs = head->attrs;

src/lang/eval.h

@@ -201,6 +201,8 @@ ray_t* ray_neq_fn(ray_t* a, ray_t* b);
 /* Logic */
 ray_t* ray_and_fn(ray_t* a, ray_t* b);
 ray_t* ray_or_fn(ray_t* a, ray_t* b);
+ray_t* ray_and_vary_fn(ray_t** args, int64_t n);
+ray_t* ray_or_vary_fn(ray_t** args, int64_t n);
 ray_t* ray_not_fn(ray_t* x);
 ray_t* ray_neg_fn(ray_t* x);
 

src/ops/agg.c

@@ -239,9 +239,13 @@ ray_t* ray_first_fn(ray_t* x) {
     }
     if (ray_is_vec(x)) {
         if (ray_len(x) == 0) return ray_typed_null(-x->type);
-        /* For SYM, GUID, STR and other non-numeric types, use collection_elem directly */
-        if (x->type == RAY_SYM || x->type == RAY_I32 || x->type == RAY_I16 ||
-            x->type == RAY_GUID || x->type == RAY_STR) {
+        /* For non-I64/F64 types route through collection_elem which
+         * preserves the element type.  The DAG path widens to i64 for
+         * DATE/TIME/TIMESTAMP/BOOL/U8 — bypass it. */
+        if (x->type == RAY_SYM   || x->type == RAY_I32  || x->type == RAY_I16 ||
+            x->type == RAY_GUID  || x->type == RAY_STR  || x->type == RAY_BOOL ||
+            x->type == RAY_U8    || x->type == RAY_DATE || x->type == RAY_TIME ||
+            x->type == RAY_TIMESTAMP) {
             int alloc = 0;
             return collection_elem(x, 0, &alloc);
         }
@@ -275,8 +279,11 @@ ray_t* ray_last_fn(ray_t* x) {
     }
     if (ray_is_vec(x)) {
         if (ray_len(x) == 0) return ray_typed_null(-x->type);
-        if (x->type == RAY_SYM || x->type == RAY_I32 || x->type == RAY_I16 ||
-            x->type == RAY_GUID || x->type == RAY_STR) {
+        /* See ray_first_fn for rationale on the type whitelist. */
+        if (x->type == RAY_SYM   || x->type == RAY_I32  || x->type == RAY_I16 ||
+            x->type == RAY_GUID  || x->type == RAY_STR  || x->type == RAY_BOOL ||
+            x->type == RAY_U8    || x->type == RAY_DATE || x->type == RAY_TIME ||
+            x->type == RAY_TIMESTAMP) {
             int alloc = 0;
             return collection_elem(x, ray_len(x) - 1, &alloc);
         }

src/ops/arith.c

@@ -330,8 +330,15 @@ ray_t* ray_mod_fn(ray_t* a, ray_t* b) {
 
 ray_t* ray_neg_fn(ray_t* x) {
     if (RAY_ATOM_IS_NULL(x)) { ray_retain(x); return x; }
-    if (x->type == -RAY_I64) return make_i64(-x->i64);
     if (x->type == -RAY_F64) return make_f64(-x->f64);
+    /* Negate via unsigned to avoid signed-overflow UB on INT_MIN.
+     * Wraparound is defined for unsigned types; (T)(uT)(-(uT)x) yields
+     * the same wrapped value the corresponding two's-complement
+     * arithmetic would produce — so (neg INT_MIN) returns INT_MIN
+     * (overflow-wrap) consistently with binary `(- 0 INT_MIN)`. */
+    if (x->type == -RAY_I64) return make_i64((int64_t)(-(uint64_t)x->i64));
+    if (x->type == -RAY_I32) return make_i32((int32_t)(-(uint32_t)x->i32));
+    if (x->type == -RAY_I16) return make_i16((int16_t)(-(uint16_t)x->i16));
     return ray_error("type", NULL);
 }
 
@@ -359,13 +366,15 @@ ray_t* ray_ceil_fn(ray_t* x) {
     return ray_error("type", NULL);
 }
 
-/* abs: absolute value, preserves type */
+/* abs: absolute value, preserves type.  Uses unsigned-wrap negation
+ * for the negative branch — same overflow-wrap semantics as `neg`,
+ * so (abs INT_MIN) returns INT_MIN rather than UB. */
 ray_t* ray_abs_fn(ray_t* x) {
     if (RAY_ATOM_IS_NULL(x)) { ray_retain(x); return x; }
     if (x->type == -RAY_F64) return make_f64(fabs(x->f64));
-    if (x->type == -RAY_I64) return make_i64(x->i64 < 0 ? -x->i64 : x->i64);
-    if (x->type == -RAY_I32) return make_i64(x->i32 < 0 ? -(int64_t)x->i32 : x->i32);
-    if (x->type == -RAY_I16) return make_i64(x->i16 < 0 ? -(int64_t)x->i16 : x->i16);
+    if (x->type == -RAY_I64) return make_i64(x->i64 < 0 ? (int64_t)(-(uint64_t)x->i64) : x->i64);
+    if (x->type == -RAY_I32) return make_i32(x->i32 < 0 ? (int32_t)(-(uint32_t)x->i32) : x->i32);
+    if (x->type == -RAY_I16) return make_i16(x->i16 < 0 ? (int16_t)(-(uint16_t)x->i16) : x->i16);
     return ray_error("type", NULL);
 }
 

src/ops/cmp.c

@@ -40,9 +40,37 @@ int char_str_cmp(ray_t* a, ray_t* b, int *out) {
     return 0;
 }
 
+/* Lexicographic compare of two SYM atoms.  Fast path: equal interned
+ * ids ⇒ identical text ⇒ 0, no global-table lookup.  Slow path: pull
+ * the backing STR via ray_sym_str and delegate to ray_str_cmp, which
+ * uses the 12-byte SSO inline path for short symbols.
+ *
+ * If a sym_str lookup fails (NULL — e.g. corrupted intern table or
+ * uninitialised state) we fall back to comparing the raw interned ids
+ * rather than declaring the unequal symbols equal.  Stable, never
+ * silently collapses distinct symbols. */
+int sym_atom_cmp(ray_t* a, ray_t* b) {
+    if (a->i64 == b->i64) return 0;
+    ray_t* sa = ray_sym_str(a->i64);
+    ray_t* sb = ray_sym_str(b->i64);
+    int r;
+    if (sa && sb) {
+        r = ray_str_cmp(sa, sb);
+    } else {
+        /* Fallback: order by interned id (stable, total).  Same sign
+         * convention as memcmp: negative if a < b, positive if a > b. */
+        r = (a->i64 < b->i64) ? -1 : 1;
+    }
+    if (sa) ray_release(sa);
+    if (sb) ray_release(sb);
+    return r;
+}
+
 /* Comparison */
 ray_t* ray_gt_fn(ray_t* a, ray_t* b) {
     { int c; if (char_str_cmp(a, b, &c) == 0) return make_bool(c > 0 ? 1 : 0); }
+    if (a->type == -RAY_SYM && b->type == -RAY_SYM)
+        return make_bool(sym_atom_cmp(a, b) > 0 ? 1 : 0);
     if (a->type == -RAY_GUID && b->type == -RAY_GUID)
         return make_bool(memcmp(ray_data(a->obj), ray_data(b->obj), 16) > 0 ? 1 : 0);
     /* Temporal comparison (same or cross-temporal via nanosecond conversion) */
@@ -63,6 +91,8 @@ ray_t* ray_gt_fn(ray_t* a, ray_t* b) {
 
 ray_t* ray_lt_fn(ray_t* a, ray_t* b) {
     { int c; if (char_str_cmp(a, b, &c) == 0) return make_bool(c < 0 ? 1 : 0); }
+    if (a->type == -RAY_SYM && b->type == -RAY_SYM)
+        return make_bool(sym_atom_cmp(a, b) < 0 ? 1 : 0);
     if (a->type == -RAY_GUID && b->type == -RAY_GUID)
         return make_bool(memcmp(ray_data(a->obj), ray_data(b->obj), 16) < 0 ? 1 : 0);
     if (is_temporal(a) && is_temporal(b)) {
@@ -82,6 +112,8 @@ ray_t* ray_lt_fn(ray_t* a, ray_t* b) {
 
 ray_t* ray_gte_fn(ray_t* a, ray_t* b) {
     { int c; if (char_str_cmp(a, b, &c) == 0) return make_bool(c >= 0 ? 1 : 0); }
+    if (a->type == -RAY_SYM && b->type == -RAY_SYM)
+        return make_bool(sym_atom_cmp(a, b) >= 0 ? 1 : 0);
     if (a->type == -RAY_GUID && b->type == -RAY_GUID)
         return make_bool(memcmp(ray_data(a->obj), ray_data(b->obj), 16) >= 0 ? 1 : 0);
     if (is_temporal(a) && is_temporal(b)) {
@@ -102,6 +134,8 @@ ray_t* ray_gte_fn(ray_t* a, ray_t* b) {
 
 ray_t* ray_lte_fn(ray_t* a, ray_t* b) {
     { int c; if (char_str_cmp(a, b, &c) == 0) return make_bool(c <= 0 ? 1 : 0); }
+    if (a->type == -RAY_SYM && b->type == -RAY_SYM)
+        return make_bool(sym_atom_cmp(a, b) <= 0 ? 1 : 0);
     if (a->type == -RAY_GUID && b->type == -RAY_GUID)
         return make_bool(memcmp(ray_data(a->obj), ray_data(b->obj), 16) <= 0 ? 1 : 0);
     if (is_temporal(a) && is_temporal(b)) {
@@ -215,6 +249,34 @@ ray_t* ray_or_fn(ray_t* a, ray_t* b) {
     return make_bool((is_truthy(a) || is_truthy(b)) ? 1 : 0);
 }
 
+/* Variadic left-fold over the binary kernels.  (and a b c) folds as
+ * (and (and a b) c) — same shape Lisp/Clojure use. */
+ray_t* ray_and_vary_fn(ray_t** args, int64_t n) {
+    if (n < 2) return ray_error("arity", "expected at least 2 args, got %lld", (long long)n);
+    ray_t* acc = ray_and_fn(args[0], args[1]);
+    if (!acc || RAY_IS_ERR(acc)) return acc;
+    for (int64_t i = 2; i < n; i++) {
+        ray_t* next = ray_and_fn(acc, args[i]);
+        ray_release(acc);
+        if (!next || RAY_IS_ERR(next)) return next;
+        acc = next;
+    }
+    return acc;
+}
+
+ray_t* ray_or_vary_fn(ray_t** args, int64_t n) {
+    if (n < 2) return ray_error("arity", "expected at least 2 args, got %lld", (long long)n);
+    ray_t* acc = ray_or_fn(args[0], args[1]);
+    if (!acc || RAY_IS_ERR(acc)) return acc;
+    for (int64_t i = 2; i < n; i++) {
+        ray_t* next = ray_or_fn(acc, args[i]);
+        ray_release(acc);
+        if (!next || RAY_IS_ERR(next)) return next;
+        acc = next;
+    }
+    return acc;
+}
+
 /* Unary */
 ray_t* ray_not_fn(ray_t* x) {
     /* Element-wise for bool vectors */

src/ops/glob.c

@@ -0,0 +1,102 @@
+/*
+ *   Copyright (c) 2025-2026 Anton Kundenko <singaraiona@gmail.com>
+ *   All rights reserved.
+ */
+
+/*
+ * Iterative glob matcher.  Replaces three pre-existing implementations
+ * that diverged in syntax (eval used *,?,[abc]; DAG used SQL %,_) and
+ * one of which (strop.c::str_glob) blew up exponentially on patterns
+ * like "a*a*a*…a*b" against an a-only string.  This single file is
+ * the only matcher; both call sites delegate here.
+ */
+
+#include "ops/glob.h"
+
+/* Lowercase an ASCII byte; non-ASCII passes through unchanged. */
+static inline char to_lower(char c) {
+    return (c >= 'A' && c <= 'Z') ? (char)(c + 32) : c;
+}
+
+/* Match a single character against a class `[ ... ]`.  On entry *pi
+ * points at the byte after `[`.  On return *pi points one past `]`.
+ * Recognises `[abc]`, `[a-z]`, leading `!` for negation, embedded
+ * `]` is allowed as the first char (after optional `!`). */
+static bool match_class(const char* p, size_t pn, size_t* pi, char c, bool ci) {
+    size_t i = *pi;
+    bool neg = false;
+    if (i < pn && p[i] == '!') { neg = true; i++; }
+    bool matched = false;
+    bool first = true;
+    char ch = ci ? to_lower(c) : c;
+    while (i < pn && (first || p[i] != ']')) {
+        char lo = ci ? to_lower(p[i]) : p[i];
+        if (i + 2 < pn && p[i + 1] == '-' && p[i + 2] != ']') {
+            char hi = ci ? to_lower(p[i + 2]) : p[i + 2];
+            if (ch >= lo && ch <= hi) matched = true;
+            i += 3;
+        } else {
+            if (ch == lo) matched = true;
+            i++;
+        }
+        first = false;
+    }
+    if (i < pn && p[i] == ']') i++;  /* consume closing bracket */
+    *pi = i;
+    return neg ? !matched : matched;
+}
+
+static bool glob_impl(const char* s, size_t sn,
+                     const char* p, size_t pn, bool ci) {
+    size_t si = 0, pi = 0;
+    size_t star_pi = (size_t)-1, star_si = 0;
+
+    while (si < sn) {
+        if (pi < pn && p[pi] == '*') {
+            star_pi = pi++;        /* remember star, skip it */
+            star_si = si;
+        } else if (pi < pn && p[pi] == '?') {
+            pi++;
+            si++;
+        } else if (pi < pn && p[pi] == '[') {
+            size_t cls_pi = pi + 1;
+            if (match_class(p, pn, &cls_pi, s[si], ci)) {
+                pi = cls_pi;
+                si++;
+            } else if (star_pi != (size_t)-1) {
+                pi = star_pi + 1;
+                si = ++star_si;
+            } else {
+                return false;
+            }
+        } else if (pi < pn) {
+            char a = ci ? to_lower(s[si]) : s[si];
+            char b = ci ? to_lower(p[pi]) : p[pi];
+            if (a == b) {
+                pi++;
+                si++;
+            } else if (star_pi != (size_t)-1) {
+                pi = star_pi + 1;
+                si = ++star_si;
+            } else {
+                return false;
+            }
+        } else if (star_pi != (size_t)-1) {
+            pi = star_pi + 1;
+            si = ++star_si;
+        } else {
+            return false;
+        }
+    }
+    /* Consumed all of input — pattern must be at end, modulo trailing stars. */
+    while (pi < pn && p[pi] == '*') pi++;
+    return pi == pn;
+}
+
+bool ray_glob_match(const char* s, size_t sn, const char* p, size_t pn) {
+    return glob_impl(s, sn, p, pn, false);
+}
+
+bool ray_glob_match_ci(const char* s, size_t sn, const char* p, size_t pn) {
+    return glob_impl(s, sn, p, pn, true);
+}