1 This patch changes the way the --checksum option works by having the
2 receiving side perform a checksum-read of every file in the file list
3 (if the sizes are equal) as the list is received, marking non-matching
4 items with a flag. The idea is that the checksum pass on the sender and
5 the receiver can then happen in parallel instead of having the reciever
6 to its checksum pass during its normal find-the-different-files pass.
8 I have benchmarked this a little, and it appears to slow things down.
10 --- orig/flist.c 2005-11-01 20:09:55
11 +++ flist.c 2005-09-16 16:41:25
12 @@ -36,6 +36,7 @@ extern int am_daemon;
14 extern int do_progress;
15 extern int always_checksum;
16 +extern int pre_checksum;
18 extern int ignore_errors;
19 extern int numeric_ids;
20 @@ -705,6 +706,16 @@ static struct file_struct *receive_file_
23 read_buf(f, sum, slen);
25 + char sum2[MD4_SUM_LENGTH];
27 + char *fname = f_name(file);
28 + if (stat(fname, &st) == 0 && st.st_size == file_length) {
29 + file_checksum(fname, sum2, st.st_size);
30 + if (memcmp(sum, sum2, slen) != 0)
31 + file->flags |= FLAG_SUM_DIFFERS;
36 if (!preserve_perms) {
37 --- orig/generator.c 2005-10-30 22:30:28
38 +++ generator.c 2005-09-16 16:41:25
39 @@ -69,6 +69,7 @@ extern int ignore_timeout;
40 extern int protocol_version;
41 extern int fuzzy_basis;
42 extern int always_checksum;
43 +extern int pre_checksum;
44 extern char *partial_dir;
45 extern char *basis_dir[];
46 extern int compare_dest;
47 @@ -359,7 +360,8 @@ void itemize(struct file_struct *file, i
50 /* Perform our quick-check heuristic for determining if a file is unchanged. */
51 -static int unchanged_file(char *fn, struct file_struct *file, STRUCT_STAT *st)
52 +static int unchanged_file(char *fn, int fnamecmp_type, struct file_struct *file,
55 if (st->st_size != file->length)
57 @@ -368,6 +370,8 @@ static int unchanged_file(char *fn, stru
58 of the file time to determine whether to sync */
59 if (always_checksum && S_ISREG(st->st_mode)) {
60 char sum[MD4_SUM_LENGTH];
61 + if (pre_checksum && fnamecmp_type == FNAMECMP_FNAME)
62 + return !(file->flags & FLAG_SUM_DIFFERS);
63 file_checksum(fn, sum, st->st_size);
64 return memcmp(sum, file->u.sum, protocol_version < 21 ? 2
65 : MD4_SUM_LENGTH) == 0;
66 @@ -923,7 +927,7 @@ static void recv_generator(char *fname,
70 - if (!unchanged_file(fnamecmpbuf, file, &st))
71 + if (!unchanged_file(fnamecmpbuf, 0, file, &st))
75 @@ -1051,7 +1055,7 @@ static void recv_generator(char *fname,
77 else if (fnamecmp_type == FNAMECMP_FUZZY)
79 - else if (unchanged_file(fnamecmp, file, &st)) {
80 + else if (unchanged_file(fnamecmp, fnamecmp_type, file, &st)) {
81 if (fnamecmp_type == FNAMECMP_FNAME) {
83 itemize(file, ndx, real_ret, &real_st,
84 --- orig/main.c 2005-11-02 07:22:12
85 +++ main.c 2005-09-16 16:41:25
86 @@ -45,6 +45,7 @@ extern int copy_links;
87 extern int keep_dirlinks;
88 extern int preserve_hard_links;
89 extern int protocol_version;
90 +extern int always_checksum;
92 extern int relative_paths;
93 extern int rsync_port;
94 @@ -60,8 +61,10 @@ extern char *filesfrom_host;
95 extern char *rsync_path;
96 extern char *shell_cmd;
97 extern char *batch_name;
98 +extern char curr_dir[MAXPATHLEN];
100 int local_server = 0;
101 +int pre_checksum = 0;
102 struct file_list *the_file_list;
104 /* There's probably never more than at most 2 outstanding child processes,
105 @@ -633,6 +636,7 @@ static void do_server_recv(int f_in, int
106 struct file_list *flist;
107 char *local_name = NULL;
109 + char olddir[sizeof curr_dir];
110 int save_verbose = verbose;
112 if (filesfrom_fd >= 0) {
113 @@ -677,6 +681,10 @@ static void do_server_recv(int f_in, int
117 + strlcpy(olddir, curr_dir, sizeof olddir);
118 + if (always_checksum && argc > 0)
119 + pre_checksum = push_dir(argv[0]);
121 flist = recv_file_list(f_in);
122 verbose = save_verbose;
124 @@ -685,6 +693,9 @@ static void do_server_recv(int f_in, int
126 the_file_list = flist;
132 local_name = get_local_name(flist,argv[0]);
134 @@ -733,6 +744,7 @@ int client_run(int f_in, int f_out, pid_
136 struct file_list *flist = NULL;
137 int exit_code = 0, exit_code2 = 0;
138 + char olddir[sizeof curr_dir];
139 char *local_name = NULL;
141 cleanup_child_pid = pid;
142 @@ -804,11 +816,18 @@ int client_run(int f_in, int f_out, pid_
146 + strlcpy(olddir, curr_dir, sizeof olddir);
147 + if (always_checksum)
148 + pre_checksum = push_dir(argv[0]);
150 if (write_batch && !am_server)
151 start_write_batch(f_in);
152 flist = recv_file_list(f_in);
153 the_file_list = flist;
158 if (flist && flist->count > 0) {
159 local_name = get_local_name(flist, argv[0]);
161 --- orig/rsync.h 2005-10-14 18:45:50
162 +++ rsync.h 2005-09-16 16:41:26
164 #define FLAG_DEL_HERE (1<<3) /* receiver/generator */
165 #define FLAG_SENT (1<<3) /* sender */
166 #define FLAG_HLINK_TOL (1<<4) /* receiver/generator */
167 +#define FLAG_SUM_DIFFERS (1<<5) /* receiver/generator */
169 /* update this if you make incompatible changes */
170 #define PROTOCOL_VERSION 29