X-Git-Url: https://mattmccutchen.net/rsync/rsync.git/blobdiff_plain/9486289ce452ab82ca8b3f16fae8c2ef5247a923..2b6b4d539b0aab98b1cbb9d8c639cb12473aae94:/match.c

diff --git a/match.c b/match.c
index bc759e24..12102158 100644
--- a/match.c
+++ b/match.c
@@ -100,7 +100,7 @@ static void matched(int f,struct sum_struct *s,struct map_struct *buf,
 		rprintf(FINFO,"match at %d last_match=%d j=%d len=%d n=%d\n",
 			(int)offset,(int)last_match,i,(int)s->sums[i].len,(int)n);
 
-	send_token(f,i,buf,last_match,n,i==-1?0:s->sums[i].len);
+	send_token(f,i,buf,last_match,n,i<0?0:s->sums[i].len);
 	data_transfer += n;
 
 	if (n > 0)
@@ -212,7 +212,17 @@ static void hash_search(int f,struct sum_struct *s,
 		} else {
 			--k;
 		}
-		
+
+		/* By matching early we avoid re-reading the
+		   data 3 times in the case where a token
+		   match comes a long way after last
+		   match. The 3 reads are caused by the
+		   running match, the checksum update and the
+		   literal send. */
+		if (offset-last_match >= CHUNK_SIZE+s->n && 
+		    (end-offset > CHUNK_SIZE)) {
+			matched(f,s,buf,offset - s->n, -2);
+		}
 	} while (++offset < end);
 	
 	matched(f,s,buf,len,-1);