| 1 | -*- indented-text -*- |
| 2 | |
| 3 | URGENT --------------------------------------------------------------- |
| 4 | |
| 5 | |
| 6 | IMPORTANT ------------------------------------------------------------ |
| 7 | |
| 8 | Cross-test versions |
| 9 | |
| 10 | Part of the regression suite should be making sure that we don't |
| 11 | break backwards compatibility: old clients vs new servers and so |
| 12 | on. Ideally we would test the cross product of versions. |
| 13 | |
| 14 | It might be sufficient to test downloads from well-known public |
| 15 | rsync servers running different versions of rsync. This will give |
| 16 | some testing and also be the most common case for having different |
| 17 | versions and not being able to upgrade. |
| 18 | |
| 19 | use chroot |
| 20 | |
| 21 | If the platform doesn't support it, then don't even try. |
| 22 | |
| 23 | If running as non-root, then don't fail, just give a warning. |
| 24 | (There was a thread about this a while ago?) |
| 25 | |
| 26 | http://lists.samba.org/pipermail/rsync/2001-August/thread.html |
| 27 | http://lists.samba.org/pipermail/rsync/2001-September/thread.html |
| 28 | |
| 29 | --files-from |
| 30 | |
| 31 | Avoids traversal. Better option than a pile of --include statements |
| 32 | for people who want to generate the file list using a find(1) |
| 33 | command or a script. |
| 34 | |
| 35 | File list structure in memory |
| 36 | |
| 37 | Rather than one big array, perhaps have a tree in memory mirroring |
| 38 | the directory tree. |
| 39 | |
| 40 | This might make sorting much faster! (I'm not sure it's a big CPU |
| 41 | problem, mind you.) |
| 42 | |
| 43 | It might also reduce memory use in storing repeated directory names |
| 44 | -- again I'm not sure this is a problem. |
| 45 | |
| 46 | Performance |
| 47 | |
| 48 | Traverse just one directory at a time. Tridge says it's possible. |
| 49 | |
| 50 | At the moment rsync reads the whole file list into memory at the |
| 51 | start, which makes us use a lot of memory and also not pipeline |
| 52 | network access as much as we could. |
| 53 | |
| 54 | |
| 55 | Handling duplicate names |
| 56 | |
| 57 | We need to be careful of duplicate names getting into the file list. |
| 58 | See clean_flist(). This could happen if multiple arguments include |
| 59 | the same file. Bad. |
| 60 | |
| 61 | I think duplicates are only a problem if they're both flowing |
| 62 | through the pipeline at the same time. For example we might have |
| 63 | updated the first occurrence after reading the checksums for the |
| 64 | second. So possibly we just need to make sure that we don't have |
| 65 | both in the pipeline at the same time. |
| 66 | |
| 67 | Possibly if we did one directory at a time that would be sufficient. |
| 68 | |
| 69 | Alternatively we could pre-process the arguments to make sure no |
| 70 | duplicates will ever be inserted. There could be some bad cases |
| 71 | when we're collapsing symlinks. |
| 72 | |
| 73 | We could have a hash table. |
| 74 | |
| 75 | The root of the problem is that we do not want more than one file |
| 76 | list entry referring to the same file. At first glance there are |
| 77 | several ways this could happen: symlinks, hardlinks, and repeated |
| 78 | names on the command line. |
| 79 | |
| 80 | If names are repeated on the command line, they may be present in |
| 81 | different forms, perhaps by traversing directory paths in different |
| 82 | ways, traversing paths including symlinks. Also we need to allow |
| 83 | for expansion of globs by rsync. |
| 84 | |
| 85 | At the moment, clean_flist() requires having the entire file list in |
| 86 | memory. Duplicate names are detected just by a string comparison. |
| 87 | |
| 88 | We don't need to worry about hard links causing duplicates because |
| 89 | files are never updated in place. Similarly for symlinks. |
| 90 | |
| 91 | I think even if we're using a different symlink mode we don't need |
| 92 | to worry. |
| 93 | |
| 94 | Unless we're really clever this will introduce a protocol |
| 95 | incompatibility, so we need to be able to accept the old format as |
| 96 | well. |
| 97 | |
| 98 | |
| 99 | Memory accounting |
| 100 | |
| 101 | At exit, show how much memory was used for the file list, etc. |
| 102 | |
| 103 | Also we do a wierd exponential-growth allocation in flist.c. I'm |
| 104 | not sure this makes sense with modern mallocs. At any rate it will |
| 105 | make us allocate a huge amount of memory for large file lists. |
| 106 | |
| 107 | |
| 108 | Hard-link handling |
| 109 | |
| 110 | At the moment hardlink handling is very expensive, so it's off by |
| 111 | default. It does not need to be so. |
| 112 | |
| 113 | Since most of the solutions are rather intertwined with the file |
| 114 | list it is probably better to fix that first, although fixing |
| 115 | hardlinks is possibly simpler. |
| 116 | |
| 117 | We can rule out hardlinked directories since they will probably |
| 118 | screw us up in all kinds of ways. They simply should not be used. |
| 119 | |
| 120 | At the moment rsync only cares about hardlinks to regular files. I |
| 121 | guess you could also use them for sockets, devices and other beasts, |
| 122 | but I have not seen them. |
| 123 | |
| 124 | When trying to reproduce hard links, we only need to worry about |
| 125 | files that have more than one name (nlinks>1 && !S_ISDIR). |
| 126 | |
| 127 | The basic point of this is to discover alternate names that refer to |
| 128 | the same file. All operations, including creating the file and |
| 129 | writing modifications to it need only to be done for the first name. |
| 130 | For all later names, we just create the link and then leave it |
| 131 | alone. |
| 132 | |
| 133 | If hard links are to be preserved: |
| 134 | |
| 135 | Before the generator/receiver fork, the list of files is received |
| 136 | from the sender (recv_file_list), and a table for detecting hard |
| 137 | links is built. |
| 138 | |
| 139 | The generator looks for hard links within the file list and does |
| 140 | not send checksums for them, though it does send other metadata. |
| 141 | |
| 142 | The sender sends the device number and inode with file entries, so |
| 143 | that files are uniquely identified. |
| 144 | |
| 145 | The receiver goes through and creates hard links (do_hard_links) |
| 146 | after all data has been written, but before directory permissions |
| 147 | are set. |
| 148 | |
| 149 | At the moment device and inum are sent as 4-byte integers, which |
| 150 | will probably cause problems on large filesystems. On Linux the |
| 151 | kernel uses 64-bit ino_t's internally, and people will soon have |
| 152 | filesystems big enough to use them. We ought to follow NFS4 in |
| 153 | using 64-bit device and inode identification, perhaps with a |
| 154 | protocol version bump. |
| 155 | |
| 156 | Once we've seen all the names for a particular file, we no longer |
| 157 | need to think about it and we can deallocate the memory. |
| 158 | |
| 159 | We can also have the case where there are links to a file that are |
| 160 | not in the tree being transferred. There's nothing we can do about |
| 161 | that. Because we rename the destination into place after writing, |
| 162 | any hardlinks to the old file are always going to be orphaned. In |
| 163 | fact that is almost necessary because otherwise we'd get really |
| 164 | confused if we were generating checksums for one name of a file and |
| 165 | modifying another. |
| 166 | |
| 167 | At the moment the code seems to make a whole second copy of the file |
| 168 | list, which seems unnecessary. |
| 169 | |
| 170 | We should have a test case that exercises hard links. Since it |
| 171 | might be hard to compare ./tls output where the inodes change we |
| 172 | might need a little program to check whether several names refer to |
| 173 | the same file. |
| 174 | |
| 175 | IPv6 |
| 176 | |
| 177 | Implement suggestions from http://www.kame.net/newsletter/19980604/ |
| 178 | and ftp://ftp.iij.ad.jp/pub/RFC/rfc2553.txt |
| 179 | |
| 180 | If a host has multiple addresses, then listen try to connect to all |
| 181 | in order until we get through. (getaddrinfo may return multiple |
| 182 | addresses.) This is kind of implemented already. |
| 183 | |
| 184 | Possibly also when starting as a server we may need to listen on |
| 185 | multiple passive addresses. This might be a bit harder, because we |
| 186 | may need to select on all of them. Hm. |
| 187 | |
| 188 | Define a syntax for IPv6 literal addresses. Since they include |
| 189 | colons, they tend to break most naming systems, including ours. |
| 190 | Based on the HTTP IPv6 syntax, I think we should use |
| 191 | |
| 192 | rsync://[::1]/foo/bar |
| 193 | [::1]::bar |
| 194 | |
| 195 | which should just take a small change to the parser code. |
| 196 | |
| 197 | Errors |
| 198 | |
| 199 | If we hang or get SIGINT, then explain where we were up to. Perhaps |
| 200 | have a static buffer that contains the current function name, or |
| 201 | some kind of description of what we were trying to do. This is a |
| 202 | little easier on people than needing to run strace/truss. |
| 203 | |
| 204 | "The dungeon collapses! You are killed." Rather than "unexpected |
| 205 | eof" give a message that is more detailed if possible and also more |
| 206 | helpful. |
| 207 | |
| 208 | File attributes |
| 209 | |
| 210 | Device major/minor numbers should be at least 32 bits each. See |
| 211 | http://lists.samba.org/pipermail/rsync/2001-November/005357.html |
| 212 | |
| 213 | Transfer ACLs. Need to think of a standard representation. |
| 214 | Probably better not to even try to convert between NT and POSIX. |
| 215 | Possibly can share some code with Samba. |
| 216 | |
| 217 | Empty directories |
| 218 | |
| 219 | With the current common --include '*/' --exclude '*' pattern, people |
| 220 | can end up with many empty directories. We might avoid this by |
| 221 | lazily creating such directories. |
| 222 | |
| 223 | zlib |
| 224 | |
| 225 | Perhaps don't use our own zlib. Will we actually be incompatible, |
| 226 | or just be slightly less efficient? |
| 227 | |
| 228 | logging |
| 229 | |
| 230 | Perhaps flush stdout after each filename, so that people trying to |
| 231 | monitor progress in a log file can do so more easily. See |
| 232 | http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=48108 |
| 233 | |
| 234 | At the connections that just get a list of modules are not logged, |
| 235 | but they should be. |
| 236 | |
| 237 | rsyncd over ssh |
| 238 | |
| 239 | There are already some patches to do this. |
| 240 | |
| 241 | proxy authentication |
| 242 | |
| 243 | Allow RSYNC_PROXY to be http://user:pass@proxy.foo:3128/, and do |
| 244 | HTTP Basic Proxy-Authentication. |
| 245 | |
| 246 | Multiple schemes are possible, up to and including the insanity that |
| 247 | is NTLM, but Basic probably covers most cases. |
| 248 | |
| 249 | SOCKS |
| 250 | |
| 251 | Add --with-socks, and then perhaps a command-line option to put them |
| 252 | on or off. This might be more reliable than LD_PRELOAD hacks. |
| 253 | |
| 254 | PLATFORMS ------------------------------------------------------------ |
| 255 | |
| 256 | Win32 |
| 257 | |
| 258 | Don't detach, because this messes up --srvany. |
| 259 | |
| 260 | http://sources.redhat.com/ml/cygwin/2001-08/msg00234.html |
| 261 | |
| 262 | According to "Effective TCP/IP Programming" (??) close() on a socket |
| 263 | has incorrect behaviour on Windows -- it sends a RST packet to the |
| 264 | other side, which gives a "connection reset by peer" error. On that |
| 265 | platform we should probably do shutdown() instead. However, on Unix |
| 266 | we are correct to call close(), because shutdown() discards |
| 267 | untransmitted data. |
| 268 | |
| 269 | DOCUMENTATION -------------------------------------------------------- |
| 270 | |
| 271 | Update README |
| 272 | |
| 273 | BUILD FARM ----------------------------------------------------------- |
| 274 | |
| 275 | Add machines |
| 276 | |
| 277 | AMDAHL UTS (Dave Dykstra) |
| 278 | |
| 279 | Cygwin (on different versions of Win32?) |
| 280 | |
| 281 | HP-UX variants (via HP?) |
| 282 | |
| 283 | SCO |
| 284 | |
| 285 | NICE ----------------------------------------------------------------- |
| 286 | |
| 287 | --no-detach and --no-fork options |
| 288 | |
| 289 | Very useful for debugging. Also good when running under a |
| 290 | daemon-monitoring process that tries to restart the service when the |
| 291 | parent exits. |
| 292 | |
| 293 | hang/timeout friendliness |
| 294 | |
| 295 | verbose output |
| 296 | |
| 297 | Indicate whether files are new, updated, or deleted |
| 298 | |
| 299 | internationalization |
| 300 | |
| 301 | Change to using gettext(). Probably need to ship this for platforms |
| 302 | that don't have it. |
| 303 | |
| 304 | Solicit translations. |
| 305 | |
| 306 | Does anyone care? |
| 307 | |
| 308 | rsyncsh |
| 309 | |
| 310 | Write a small emulation of interactive ftp as a Pythonn program |
| 311 | that calls rsync. Commands such as "cd", "ls", "ls *.c" etc map |
| 312 | fairly directly into rsync commands: it just needs to remember the |
| 313 | current host, directory and so on. We can probably even do |
| 314 | completion of remote filenames. |
| 315 | |
| 316 | %K% |