From 62a2b529542b26c073d3a3aab68f8ced50597653 Mon Sep 17 00:00:00 2001
From: Joe Nelson <cred+github@begriffs.com>
Date: Sat, 22 Dec 2018 22:03:51 -0600
Subject: [PATCH] Allow random values to scale up to 64-bits

Improves precision in reservoir sampling
---
 rand.c   | 22 ++++++++++------------
 rand.h   |  4 +++-
 randln.c |  3 ++-
 3 files changed, 15 insertions(+), 14 deletions(-)

diff --git a/rand.c b/rand.c
index 7dfb20a..4b6a8ad 100644
--- a/rand.c
+++ b/rand.c
@@ -53,21 +53,17 @@ unsigned long defensive_seed(void)
  * Alternative to the garbage implementation of rand() in many
  * systems such as FreeBSD and OS X. Uses xorshift.
  *
- * Returns a value between 0 and DEFENSIVE_RAND_MAX, which on
- * many platforms is smaller than what unsigned long can hold.
+ * Returns a value between 0 and DEFENSIVE_RAND_MAX.
+ * From https://nullprogram.com/blog/2018/07/31/
  */
 unsigned long defensive_rand()
 {
-	/*
-	 * The long type should portably be at least 32 bits, and we
-	 * mask the calculations to 32 bits in case long happens to be
-	 * wider. The constants are tuned for a 32-bit type.
-	 */
-	unsigned long y = g_rand_state & DEFENSIVE_RAND_MAX;
-	y ^= (y << 13) & DEFENSIVE_RAND_MAX;
-	y ^= y >> 17;
-	y ^= (y << 5) & DEFENSIVE_RAND_MAX;
-	return (g_rand_state = y);
+    g_rand_state ^= g_rand_state >> 30;
+    g_rand_state *= 0xbf58476d1ce4e5b9UL;
+    g_rand_state ^= g_rand_state >> 27;
+    g_rand_state *= 0x94d049bb133111ebUL;
+    g_rand_state ^= g_rand_state >> 31;
+    return g_rand_state;
 }
 
 void defensive_srand(unsigned long s)
@@ -88,6 +84,8 @@ unsigned long djb2hash(const unsigned char *str)
 
 /* It's better to combine entropic values this way rather
  * than e.g. adding them together.
+ *
+ * From http://www.pcg-random.org/posts/developing-a-seed_seq-alternative.html
  */
 unsigned long mix(unsigned long x, unsigned long y)
 {
diff --git a/rand.h b/rand.h
index 206dff1..b9f6df5 100644
--- a/rand.h
+++ b/rand.h
@@ -1,7 +1,9 @@
 #ifndef RAND_H
 #define RAND_H
 
-#define DEFENSIVE_RAND_MAX 0xffffffffUL
+#include <limits.h>
+
+#define DEFENSIVE_RAND_MAX ULONG_MAX
 
 void defensive_srand(unsigned long);
 unsigned long defensive_seed(void);
diff --git a/randln.c b/randln.c
index 4bfa117..9adbc64 100644
--- a/randln.c
+++ b/randln.c
@@ -53,7 +53,8 @@ void via_fseek(FILE *fp)
 	if ((filesz = ftell(fp)) == -1)
 		die_perror(NULL);
 
-	pos = (int)((double)defensive_rand() / ((double)DEFENSIVE_RAND_MAX + 1) * filesz);
+	pos = ((double)defensive_rand() /
+		((double)DEFENSIVE_RAND_MAX)) * filesz;
 
 	if (fseek(fp, pos, SEEK_SET) != 0)
 		die_perror(NULL);