🐐 Hash tables

centvrion/compiler/runtime/cent_runtime.c

@@ -1172,44 +1172,123 @@ static int _cent_key_eq(CentValue a, CentValue b) {
   return 0;
 }
 
+/* splitmix64 finalizer — good distribution for sequential ints */
+static uint32_t _cent_hash_int(long v) {
+  uint64_t x = (uint64_t)v;
+  x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9ULL;
+  x = (x ^ (x >> 27)) * 0x94d049bb133111ebULL;
+  x =  x ^ (x >> 31);
+  return (uint32_t)x;
+}
+
+/* FNV-1a */
+static uint32_t _cent_hash_str(const char *s) {
+  uint32_t h = 2166136261u;
+  for (; *s; s++) {
+    h ^= (uint8_t)*s;
+    h *= 16777619u;
+  }
+  return h;
+}
+
+static uint32_t _cent_hash_key(CentValue k) {
+  if (k.type == CENT_INT) return _cent_hash_int(k.ival);
+  if (k.type == CENT_STR) return _cent_hash_str(k.sval);
+  cent_type_error("dict key must be a numeral or string");
+  return 0;
+}
+
+static int _next_pow2(int n) {
+  int p = 1;
+  while (p < n) p <<= 1;
+  return p;
+}
+
+/* Probe for `key` in the bucket array. Returns the bucket slot — either
+   one whose stored index points to a matching key (hit), or an empty
+   slot (-1) where the key would be inserted. nbuckets is a power of 2. */
+static int _cent_dict_probe(const CentDict *d, CentValue key, uint32_t h) {
+  uint32_t mask = (uint32_t)d->nbuckets - 1;
+  uint32_t i = h & mask;
+  while (1) {
+    int idx = d->buckets[i];
+    if (idx < 0) return (int)i;
+    if (_cent_key_eq(d->keys[idx], key)) return (int)i;
+    i = (i + 1) & mask;
+  }
+}
+
+static void _cent_dict_rehash(CentDict *d, int new_nbuckets) {
+  int *new_buckets = cent_arena_alloc(cent_arena, new_nbuckets * sizeof(int));
+  for (int i = 0; i < new_nbuckets; i++) new_buckets[i] = -1;
+  uint32_t mask = (uint32_t)new_nbuckets - 1;
+  for (int idx = 0; idx < d->len; idx++) {
+    uint32_t h = _cent_hash_key(d->keys[idx]);
+    uint32_t i = h & mask;
+    while (new_buckets[i] >= 0) i = (i + 1) & mask;
+    new_buckets[i] = idx;
+  }
+  d->buckets  = new_buckets;
+  d->nbuckets = new_nbuckets;
+}
+
 CentValue cent_dict_new(int cap) {
   if (cap < 4) cap = 4;
+  int nbuckets = _next_pow2(cap * 2);
   CentValue *keys = cent_arena_alloc(cent_arena, cap * sizeof(CentValue));
   CentValue *vals = cent_arena_alloc(cent_arena, cap * sizeof(CentValue));
-  return cent_dict_val(keys, vals, 0, cap);
+  int       *buckets = cent_arena_alloc(cent_arena, nbuckets * sizeof(int));
+  for (int i = 0; i < nbuckets; i++) buckets[i] = -1;
+  return cent_dict_val(keys, vals, buckets, 0, cap, nbuckets);
 }
 
 void cent_dict_set(CentValue *dict, CentValue key, CentValue val) {
   if (dict->type != CENT_DICT)
     cent_type_error("dict-set requires a dict");
-  for (int i = 0; i < dict->dval.len; i++) {
+  CentDict *d = &dict->dval;
-    if (_cent_key_eq(dict->dval.keys[i], key)) {
+
-      dict->dval.vals[i] = val;
+  uint32_t h = _cent_hash_key(key);
-      return;
+  int slot = _cent_dict_probe(d, key, h);
-    }
+  int idx  = d->buckets[slot];
+  if (idx >= 0) {
+    d->vals[idx] = val;
+    return;
   }
-  if (dict->dval.len >= dict->dval.cap) {
+
-    int new_cap = dict->dval.cap * 2;
+  /* Grow the keys/vals arrays first so the new entry has a stable index. */
+  if (d->len >= d->cap) {
+    int new_cap = d->cap * 2;
     CentValue *new_keys = cent_arena_alloc(cent_arena, new_cap * sizeof(CentValue));
     CentValue *new_vals = cent_arena_alloc(cent_arena, new_cap * sizeof(CentValue));
-    memcpy(new_keys, dict->dval.keys, dict->dval.len * sizeof(CentValue));
+    memcpy(new_keys, d->keys, d->len * sizeof(CentValue));
-    memcpy(new_vals, dict->dval.vals, dict->dval.len * sizeof(CentValue));
+    memcpy(new_vals, d->vals, d->len * sizeof(CentValue));
-    dict->dval.keys = new_keys;
+    d->keys = new_keys;
-    dict->dval.vals = new_vals;
+    d->vals = new_vals;
-    dict->dval.cap  = new_cap;
+    d->cap  = new_cap;
+  }
+
+  int new_idx = d->len;
+  d->keys[new_idx] = key;
+  d->vals[new_idx] = val;
+  d->len++;
+
+  /* If load factor would exceed 0.75, rehash — this re-inserts every
+     entry including the one we just appended, so we're done. Otherwise
+     the slot picked by the earlier probe is still valid. */
+  if (d->len * 4 >= d->nbuckets * 3) {
+    _cent_dict_rehash(d, d->nbuckets * 2);
+  } else {
+    d->buckets[slot] = new_idx;
   }
-  dict->dval.keys[dict->dval.len] = key;
-  dict->dval.vals[dict->dval.len] = val;
-  dict->dval.len++;
 }
 
 CentValue cent_dict_get(CentValue dict, CentValue key) {
   if (dict.type != CENT_DICT)
     cent_type_error("dict-get requires a dict");
-  for (int i = 0; i < dict.dval.len; i++) {
+  uint32_t h = _cent_hash_key(key);
-    if (_cent_key_eq(dict.dval.keys[i], key))
+  int slot = _cent_dict_probe(&dict.dval, key, h);
-      return dict.dval.vals[i];
+  int idx  = dict.dval.buckets[slot];
-  }
+  if (idx >= 0) return dict.dval.vals[idx];
   cent_runtime_error("Key not found in dict");
   return cent_null();
 }

centvrion/compiler/runtime/cent_runtime.h

@@ -47,10 +47,13 @@ struct CentList {
 };
 
 struct CentDict {
-  CentValue *keys;
+  CentValue *keys;     /* insertion-order array, len entries     */
-  CentValue *vals;
+  CentValue *vals;     /* parallel to keys                       */
-  int        len;
+  int       *buckets;  /* hash table; values are indices into    */
-  int        cap;
+                       /* keys/vals, or -1 for empty             */
+  int        len;      /* number of entries                      */
+  int        cap;      /* capacity of keys/vals                  */
+  int        nbuckets; /* size of buckets, power of 2            */
 };
 
 struct CentValue {
@@ -135,13 +138,17 @@ static inline CentValue cent_func_val(CentFuncPtr fn, const char **param_names,
   r.fnval.param_count = param_count;
   return r;
 }
-static inline CentValue cent_dict_val(CentValue *keys, CentValue *vals, int len, int cap) {
+static inline CentValue cent_dict_val(CentValue *keys, CentValue *vals,
+                                      int *buckets, int len, int cap,
+                                      int nbuckets) {
   CentValue r;
   r.type = CENT_DICT;
-  r.dval.keys = keys;
+  r.dval.keys     = keys;
-  r.dval.vals = vals;
+  r.dval.vals     = vals;
-  r.dval.len  = len;
+  r.dval.buckets  = buckets;
-  r.dval.cap  = cap;
+  r.dval.len      = len;
+  r.dval.cap      = cap;
+  r.dval.nbuckets = nbuckets;
   return r;
 }
 

tests/08_test_tabulas_.py

@@ -189,3 +189,34 @@ class TestDictDisplay(unittest.TestCase):
   @parameterized.expand(dict_display_tests)
   def test_dict_display(self, source, nodes, value, output):
     run_test(self, source, nodes, value, output)
+
+
+class TestDictGrowth(unittest.TestCase):
+  def test_dict_growth_preserves_order_and_lookup(self):
+    # Inserts XX entries via PER; pushes the compiled dict through
+    # multiple rehashes (initial cap=4) and verifies that lookup, length,
+    # and insertion-order iteration all still hold afterwards.
+    source = (
+      "DESIGNA d VT TABVLA {}\n"
+      "PER i IN [I VSQVE XX] FAC {\n"
+      "DESIGNA d[i] VT i * II\n"
+      "}\n"
+      "DIC(d[X])\n"
+      "DIC(LONGITVDO(d))\n"
+      "DIC(CLAVES(d))"
+    )
+    nodes = Program([], [
+      Designa(ID("d"), DataDict([])),
+      PerStatement(
+        DataRangeArray(Numeral("I"), Numeral("XX")),
+        ID("i"),
+        [DesignaIndex(ID("d"), [ID("i")],
+                      BinOp(ID("i"), Numeral("II"), "SYMBOL_TIMES"))],
+      ),
+      ExpressionStatement(BuiltIn("DIC", [ArrayIndex(ID("d"), Numeral("X"))])),
+      ExpressionStatement(BuiltIn("DIC", [BuiltIn("LONGITVDO", [ID("d")])])),
+      ExpressionStatement(BuiltIn("DIC", [BuiltIn("CLAVES", [ID("d")])])),
+    ])
+    keys_str = "[" + " ".join(int_to_num(i, False) for i in range(1, 21)) + "]"
+    output = f"XX\nXX\n{keys_str}\n"
+    run_test(self, source, nodes, ValStr(keys_str), output)