# HG changeset patch
# User Vadim Gelfer <vadim.gelfer@gmail.com>
# Date 1152309775 25200
# Node ID fa76c5d609c909a5fdcb64bb824b37001f0d9970
# Parent  6a961a54f953e6d0f2e6760f15f42f0000e1b774
bdiff: improve worst case behavior by 100x.

on 5.8MB (244.000 lines) text file with similar lines, hash before
this change made diff against empty file take 75 seconds.  this change
improves performance to 0.6 seconds.  result is that clone of smallish
repo (137MB) with some files like this takes 1 minute instead of 10
minutes.

common case of diff is 10% slower now, probably because of worse cache
locality. but diff does not affect overall performance in common case
(less than 1% of runtime is in diff when it is working ok), so this
tradeoff looks good.

diff -r 6a961a54f953 -r fa76c5d609c9 mercurial/bdiff.c
--- a/mercurial/bdiff.c	Fri Jul 07 11:23:53 2006 +0200
+++ b/mercurial/bdiff.c	Fri Jul 07 15:02:55 2006 -0700
@@ -65,7 +65,7 @@
 
 int splitlines(const char *a, int len, struct line **lr)
 {
-	int h, i;
+	int g, h, i;
 	const char *p, *b = a;
 	struct line *l;
 
@@ -82,7 +82,16 @@
 	/* build the line array and calculate hashes */
 	h = 0;
 	for (p = a; p < a + len; p++) {
-		h = *p + rol32(h, 7); /* a simple hash from GNU diff */
+		/*
+		 * a simple hash from GNU diff, with better collision
+		 * resistance from hashpjw. this slows down common
+		 * case by 10%, but speeds up worst case by 100x.
+		 */
+		h = *p + rol32(h, 7);
+		if ((g = h & 0xf0000000)) {
+			h ^= g >> 24;
+			h ^= g;
+		}
 		if (*p == '\n' || p == a + len - 1) {
 			l->len = p - b + 1;
 			l->h = h * l->len;