bdiff: don't check border condition in loop

`plast = a + len - 1`. So, this "for" loop iterates from "a" to "plast",
inclusive. So, `p == plast` can only be true on the final iteration
of the loop. So checking for it on every loop iteration is wasteful.

This patch simply decreases the upper bound of the loop by 1 and
adds an explicit check after iteration for the `p == plast` case.
We can't simply add 1 to the initial value for "i" because that
doesn't do the correct thing on empty input strings.

`perfbdiff -m 3041e4d59df2` on the Firefox repo becomes significantly
faster:

! wall 0.072763 comb 0.070000 user 0.070000 sys 0.000000 (best of 100)
! wall 0.053221 comb 0.060000 user 0.060000 sys 0.000000 (best of 100)

For the curious, this code has its origins in 8b067bde6679, which is
the changeset that introduced bdiff.c in 2005.

Also, GNU diffutils is able to perform a similar line-based diff in
under 20ms. So there's likely more perf wins to be found in this code.
One of them is the hashing algorithm. But it looks like mpm spent
some time testing hash collisions in d0c48891dd4a. I'd like to do the
same before switching away from lyhash, just to be on the safe side.

mercurial/bdiff.c

@@ -1,296 +1,298
 /*
  bdiff.c - efficient binary diff extension for Mercurial
 
  Copyright 2005, 2006 Matt Mackall <[e-mail]>
 
  This software may be used and distributed according to the terms of
  the GNU General Public License, incorporated herein by reference.
 
  Based roughly on Python difflib
 */
 
 #include <stdlib.h>
 #include <string.h>
 #include <limits.h>
 
 #include "compat.h"
 #include "bitmanipulation.h"
 #include "bdiff.h"
 
 struct pos {
 	int pos, len;
 };
 
 int bdiff_splitlines(const char *a, ssize_t len, struct bdiff_line **lr)
 {
 	unsigned hash;
 	int i;
 	const char *p, *b = a;
 	const char * const plast = a + len - 1;
 	struct bdiff_line *l;
 
 	/* count the lines */
 	i = 1; /* extra line for sentinel */
-	for (p = a; p < a + len; p++)
-		if (*p == '\n' || p == plast)
+	for (p = a; p < plast; p++)
+		if (*p == '\n')
+			i++;
+	if (p == plast)
 			i++;
 
 	*lr = l = (struct bdiff_line *)malloc(sizeof(struct bdiff_line) * i);
 	if (!l)
 		return -1;
 
 	/* build the line array and calculate hashes */
 	hash = 0;
 	for (p = a; p < a + len; p++) {
 		/* Leonid Yuriev's hash */
 		hash = (hash * 1664525) + (unsigned char)*p + 1013904223;
 
 		if (*p == '\n' || p == plast) {
 			l->hash = hash;
 			hash = 0;
 			l->len = p - b + 1;
 			l->l = b;
 			l->n = INT_MAX;
 			l++;
 			b = p + 1;
 		}
 	}
 
 	/* set up a sentinel */
 	l->hash = 0;
 	l->len = 0;
 	l->l = a + len;
 	return i - 1;
 }
 
 static inline int cmp(struct bdiff_line *a, struct bdiff_line *b)
 {
 	return a->hash != b->hash || a->len != b->len || memcmp(a->l, b->l, a->len);
 }
 
 static int equatelines(struct bdiff_line *a, int an, struct bdiff_line *b,
 	int bn)
 {
 	int i, j, buckets = 1, t, scale;
 	struct pos *h = NULL;
 
 	/* build a hash table of the next highest power of 2 */
 	while (buckets < bn + 1)
 		buckets *= 2;
 
 	/* try to allocate a large hash table to avoid collisions */
 	for (scale = 4; scale; scale /= 2) {
 		h = (struct pos *)malloc(scale * buckets * sizeof(struct pos));
 		if (h)
 			break;
 	}
 
 	if (!h)
 		return 0;
 
 	buckets = buckets * scale - 1;
 
 	/* clear the hash table */
 	for (i = 0; i <= buckets; i++) {
 		h[i].pos = -1;
 		h[i].len = 0;
 	}
 
 	/* add lines to the hash table chains */
 	for (i = 0; i < bn; i++) {
 		/* find the equivalence class */
 		for (j = b[i].hash & buckets; h[j].pos != -1;
 		     j = (j + 1) & buckets)
 			if (!cmp(b + i, b + h[j].pos))
 				break;
 
 		/* add to the head of the equivalence class */
 		b[i].n = h[j].pos;
 		b[i].e = j;
 		h[j].pos = i;
 		h[j].len++; /* keep track of popularity */
 	}
 
 	/* compute popularity threshold */
 	t = (bn >= 31000) ? bn / 1000 : 1000000 / (bn + 1);
 
 	/* match items in a to their equivalence class in b */
 	for (i = 0; i < an; i++) {
 		/* find the equivalence class */
 		for (j = a[i].hash & buckets; h[j].pos != -1;
 		     j = (j + 1) & buckets)
 			if (!cmp(a + i, b + h[j].pos))
 				break;
 
 		a[i].e = j; /* use equivalence class for quick compare */
 		if (h[j].len <= t)
 			a[i].n = h[j].pos; /* point to head of match list */
 		else
 			a[i].n = -1; /* too popular */
 	}
 
 	/* discard hash tables */
 	free(h);
 	return 1;
 }
 
 static int longest_match(struct bdiff_line *a, struct bdiff_line *b,
 			struct pos *pos,
 			 int a1, int a2, int b1, int b2, int *omi, int *omj)
 {
 	int mi = a1, mj = b1, mk = 0, i, j, k, half;
 
 	/* window our search on large regions to better bound
 	   worst-case performance. by choosing a window at the end, we
 	   reduce skipping overhead on the b chains. */
 	if (a2 - a1 > 30000)
 		a1 = a2 - 30000;
 
 	half = (a1 + a2) / 2;
 
 	for (i = a1; i < a2; i++) {
 		/* skip all lines in b after the current block */
 		for (j = a[i].n; j >= b2; j = b[j].n)
 			;
 
 		/* loop through all lines match a[i] in b */
 		for (; j >= b1; j = b[j].n) {
 			/* does this extend an earlier match? */
 			for (k = 1; j - k >= b1 && i - k >= a1; k++) {
 				/* reached an earlier match? */
 				if (pos[j - k].pos == i - k) {
 					k += pos[j - k].len;
 					break;
 				}
 				/* previous line mismatch? */
 				if (a[i - k].e != b[j - k].e)
 					break;
 			}
 
 			pos[j].pos = i;
 			pos[j].len = k;
 
 			/* best match so far? we prefer matches closer
 			   to the middle to balance recursion */
 			if (k > mk || (k == mk && (i <= mi || i < half))) {
 				mi = i;
 				mj = j;
 				mk = k;
 			}
 		}
 	}
 
 	if (mk) {
 		mi = mi - mk + 1;
 		mj = mj - mk + 1;
 	}
 
 	/* expand match to include subsequent popular lines */
 	while (mi + mk < a2 && mj + mk < b2 &&
 	       a[mi + mk].e == b[mj + mk].e)
 		mk++;
 
 	*omi = mi;
 	*omj = mj;
 
 	return mk;
 }
 
 static struct bdiff_hunk *recurse(struct bdiff_line *a, struct bdiff_line *b,
 				struct pos *pos,
 			    int a1, int a2, int b1, int b2, struct bdiff_hunk *l)
 {
 	int i, j, k;
 
 	while (1) {
 		/* find the longest match in this chunk */
 		k = longest_match(a, b, pos, a1, a2, b1, b2, &i, &j);
 		if (!k)
 			return l;
 
 		/* and recurse on the remaining chunks on either side */
 		l = recurse(a, b, pos, a1, i, b1, j, l);
 		if (!l)
 			return NULL;
 
 		l->next = (struct bdiff_hunk *)malloc(sizeof(struct bdiff_hunk));
 		if (!l->next)
 			return NULL;
 
 		l = l->next;
 		l->a1 = i;
 		l->a2 = i + k;
 		l->b1 = j;
 		l->b2 = j + k;
 		l->next = NULL;
 
 		/* tail-recursion didn't happen, so do equivalent iteration */
 		a1 = i + k;
 		b1 = j + k;
 	}
 }
 
 int bdiff_diff(struct bdiff_line *a, int an, struct bdiff_line *b,
 		int bn, struct bdiff_hunk *base)
 {
 	struct bdiff_hunk *curr;
 	struct pos *pos;
 	int t, count = 0;
 
 	/* allocate and fill arrays */
 	t = equatelines(a, an, b, bn);
 	pos = (struct pos *)calloc(bn ? bn : 1, sizeof(struct pos));
 
 	if (pos && t) {
 		/* generate the matching block list */
 
 		curr = recurse(a, b, pos, 0, an, 0, bn, base);
 		if (!curr)
 			return -1;
 
 		/* sentinel end hunk */
 		curr->next = (struct bdiff_hunk *)malloc(sizeof(struct bdiff_hunk));
 		if (!curr->next)
 			return -1;
 		curr = curr->next;
 		curr->a1 = curr->a2 = an;
 		curr->b1 = curr->b2 = bn;
 		curr->next = NULL;
 	}
 
 	free(pos);
 
 	/* normalize the hunk list, try to push each hunk towards the end */
 	for (curr = base->next; curr; curr = curr->next) {
 		struct bdiff_hunk *next = curr->next;
 
 		if (!next)
 			break;
 
 		if (curr->a2 == next->a1 || curr->b2 == next->b1)
 			while (curr->a2 < an && curr->b2 < bn
 			       && next->a1 < next->a2
 			       && next->b1 < next->b2
 			       && !cmp(a + curr->a2, b + curr->b2)) {
 				curr->a2++;
 				next->a1++;
 				curr->b2++;
 				next->b1++;
 			}
 	}
 
 	for (curr = base->next; curr; curr = curr->next)
 		count++;
 	return count;
 }
 
 void bdiff_freehunks(struct bdiff_hunk *l)
 {
 	struct bdiff_hunk *n;
 	for (; l; l = n) {
 		n = l->next;
 		free(l);
 	}
 }