Hash checks strings against min and max length for the set.

strlen() is also cheap if it has a SIMD implementation, so we can
afford this optimization to reject some strings immediately.

src/ph.c

@@ -49,6 +49,8 @@ struct ph {
 #define PH_MAGIC	0x00cd8c1d
 	int32_t		*inter;
 	unsigned	*tbl;
+	size_t		minlen;
+	size_t		maxlen;
 	int32_t		mask;
 };
 
@@ -108,6 +110,7 @@ len_desc_cmp(const void *b1, const void *b2)
 struct ph *
 PH_Generate(char * const * const strings, unsigned n)
 {
+	size_t min = SIZE_MAX, max = 0;
 	int *found = NULL;
 	struct bucket *buckets = NULL;
 	int32_t *inter = NULL;
@@ -138,6 +141,7 @@ PH_Generate(char * const * const strings, unsigned n)
 		goto exit;
 
 	for (i = 0; i < n; i++) {
+		size_t len = strlen(strings[i]);
 		uint32_t h = hash(FNV32_OFFSET_BASIS, strings[i]) & mask;
 		assert(h < sz);
 
@@ -149,6 +153,11 @@ PH_Generate(char * const * const strings, unsigned n)
 			goto exit;
 		buckets[h].idx[buckets[h].n] = i;
 		buckets[h].n++;
+
+		if (len < min)
+			min = len;
+		if (len > max)
+			max = len;
 	}
 
 	qsort(buckets, sz, sizeof(*buckets), len_desc_cmp);
@@ -209,6 +218,8 @@ PH_Generate(char * const * const strings, unsigned n)
 	ph->inter = inter;
 	ph->tbl = tbl;
 	ph->mask = mask;
+	ph->minlen = min;
+	ph->maxlen = max;
 
  exit:
 	AN(buckets);
@@ -233,6 +244,7 @@ PH_Lookup(const struct ph * const restrict ph,
 	  char * const restrict * const restrict strings,
 	  const char * const restrict subject)
 {
+	size_t len;
 	int32_t h;
 	unsigned idx;
 
@@ -243,6 +255,12 @@ PH_Lookup(const struct ph * const restrict ph,
 	if (ph == NULL)
 		return (UINT_MAX);
 
+	len = strlen(subject);
+	if (len < ph->minlen)
+		return (UINT_MAX);
+	if (len > ph->maxlen)
+		return (UINT_MAX);
+
 	h = ph->inter[hash(FNV32_OFFSET_BASIS, subject) & ph->mask];
 	if (h < 0)
 		idx = ph->tbl[-h - 1];