3 URGENT ---------------------------------------------------------------
6 IMPORTANT ------------------------------------------------------------
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.
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.
21 If the platform doesn't support it, then don't even try.
23 If running as non-root, then don't fail, just give a warning.
24 (There was a thread about this a while ago?)
26 http://lists.samba.org/pipermail/rsync/2001-August/thread.html
27 http://lists.samba.org/pipermail/rsync/2001-September/thread.html
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)
38 Traverse just one directory at a time. Tridge says it's possible.
40 At the moment rsync reads the whole file list into memory at the
41 start, which makes us use a lot of memory and also not pipeline
42 network access as much as we could.
45 Handling duplicate names
47 We need to be careful of duplicate names getting into the file list.
48 See clean_flist(). This could happen if multiple arguments include
51 I think duplicates are only a problem if they're both flowing
52 through the pipeline at the same time. For example we might have
53 updated the first occurrence after reading the checksums for the
54 second. So possibly we just need to make sure that we don't have
55 both in the pipeline at the same time.
57 Possibly if we did one directory at a time that would be sufficient.
59 Alternatively we could pre-process the arguments to make sure no
60 duplicates will ever be inserted. There could be some bad cases
61 when we're collapsing symlinks.
63 We could have a hash table.
65 The root of the problem is that we do not want more than one file
66 list entry referring to the same file. At first glance there are
67 several ways this could happen: symlinks, hardlinks, and repeated
68 names on the command line.
70 If names are repeated on the command line, they may be present in
71 different forms, perhaps by traversing directory paths in different
72 ways, traversing paths including symlinks. Also we need to allow
73 for expansion of globs by rsync.
75 At the moment, clean_flist() requires having the entire file list in
76 memory. Duplicate names are detected just by a string comparison.
78 We don't need to worry about hard links causing duplicates because
79 files are never updated in place. Similarly for symlinks.
81 I think even if we're using a different symlink mode we don't need
84 Unless we're really clever this will introduce a protocol
85 incompatibility, so we need to be able to accept the old format as
91 At exit, show how much memory was used for the file list, etc.
93 Also we do a wierd exponential-growth allocation in flist.c. I'm
94 not sure this makes sense with modern mallocs. At any rate it will
95 make us allocate a huge amount of memory for large file lists.
97 We can try using the GNU/SVID/XPG mallinfo() function to get some
103 At the moment hardlink handling is very expensive, so it's off by
104 default. It does not need to be so.
106 Since most of the solutions are rather intertwined with the file
107 list it is probably better to fix that first, although fixing
108 hardlinks is possibly simpler.
110 We can rule out hardlinked directories since they will probably
111 screw us up in all kinds of ways. They simply should not be used.
113 At the moment rsync only cares about hardlinks to regular files. I
114 guess you could also use them for sockets, devices and other beasts,
115 but I have not seen them.
117 When trying to reproduce hard links, we only need to worry about
118 files that have more than one name (nlinks>1 && !S_ISDIR).
120 The basic point of this is to discover alternate names that refer to
121 the same file. All operations, including creating the file and
122 writing modifications to it need only to be done for the first name.
123 For all later names, we just create the link and then leave it
126 If hard links are to be preserved:
128 Before the generator/receiver fork, the list of files is received
129 from the sender (recv_file_list), and a table for detecting hard
132 The generator looks for hard links within the file list and does
133 not send checksums for them, though it does send other metadata.
135 The sender sends the device number and inode with file entries, so
136 that files are uniquely identified.
138 The receiver goes through and creates hard links (do_hard_links)
139 after all data has been written, but before directory permissions
142 At the moment device and inum are sent as 4-byte integers, which
143 will probably cause problems on large filesystems. On Linux the
144 kernel uses 64-bit ino_t's internally, and people will soon have
145 filesystems big enough to use them. We ought to follow NFS4 in
146 using 64-bit device and inode identification, perhaps with a
147 protocol version bump.
149 Once we've seen all the names for a particular file, we no longer
150 need to think about it and we can deallocate the memory.
152 We can also have the case where there are links to a file that are
153 not in the tree being transferred. There's nothing we can do about
154 that. Because we rename the destination into place after writing,
155 any hardlinks to the old file are always going to be orphaned. In
156 fact that is almost necessary because otherwise we'd get really
157 confused if we were generating checksums for one name of a file and
160 At the moment the code seems to make a whole second copy of the file
161 list, which seems unnecessary.
163 We should have a test case that exercises hard links. Since it
164 might be hard to compare ./tls output where the inodes change we
165 might need a little program to check whether several names refer to
170 Implement suggestions from http://www.kame.net/newsletter/19980604/
171 and ftp://ftp.iij.ad.jp/pub/RFC/rfc2553.txt
173 If a host has multiple addresses, then listen try to connect to all
174 in order until we get through. (getaddrinfo may return multiple
175 addresses.) This is kind of implemented already.
177 Possibly also when starting as a server we may need to listen on
178 multiple passive addresses. This might be a bit harder, because we
179 may need to select on all of them. Hm.
181 Define a syntax for IPv6 literal addresses. Since they include
182 colons, they tend to break most naming systems, including ours.
183 Based on the HTTP IPv6 syntax, I think we should use
185 rsync://[::1]/foo/bar
188 which should just take a small change to the parser code.
192 If we hang or get SIGINT, then explain where we were up to. Perhaps
193 have a static buffer that contains the current function name, or
194 some kind of description of what we were trying to do. This is a
195 little easier on people than needing to run strace/truss.
197 "The dungeon collapses! You are killed." Rather than "unexpected
198 eof" give a message that is more detailed if possible and also more
203 Device major/minor numbers should be at least 32 bits each. See
204 http://lists.samba.org/pipermail/rsync/2001-November/005357.html
206 Transfer ACLs. Need to think of a standard representation.
207 Probably better not to even try to convert between NT and POSIX.
208 Possibly can share some code with Samba.
212 With the current common --include '*/' --exclude '*' pattern, people
213 can end up with many empty directories. We might avoid this by
214 lazily creating such directories.
218 Perhaps don't use our own zlib. Will we actually be incompatible,
219 or just be slightly less efficient?
223 Perhaps flush stdout after each filename, so that people trying to
224 monitor progress in a log file can do so more easily. See
225 http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=48108
227 At the connections that just get a list of modules are not logged,
232 There are already some patches to do this.
236 Allow RSYNC_PROXY to be http://user:pass@proxy.foo:3128/, and do
237 HTTP Basic Proxy-Authentication.
239 Multiple schemes are possible, up to and including the insanity that
240 is NTLM, but Basic probably covers most cases.
244 Add --with-socks, and then perhaps a command-line option to put them
245 on or off. This might be more reliable than LD_PRELOAD hacks.
247 PLATFORMS ------------------------------------------------------------
251 Don't detach, because this messes up --srvany.
253 http://sources.redhat.com/ml/cygwin/2001-08/msg00234.html
255 According to "Effective TCP/IP Programming" (??) close() on a socket
256 has incorrect behaviour on Windows -- it sends a RST packet to the
257 other side, which gives a "connection reset by peer" error. On that
258 platform we should probably do shutdown() instead. However, on Unix
259 we are correct to call close(), because shutdown() discards
262 DOCUMENTATION --------------------------------------------------------
266 BUILD FARM -----------------------------------------------------------
270 AMDAHL UTS (Dave Dykstra)
272 Cygwin (on different versions of Win32?)
274 HP-UX variants (via HP?)
278 NICE -----------------------------------------------------------------
280 --no-detach and --no-fork options
282 Very useful for debugging. Also good when running under a
283 daemon-monitoring process that tries to restart the service when the
286 hang/timeout friendliness
290 Indicate whether files are new, updated, or deleted
294 Change to using gettext(). Probably need to ship this for platforms
297 Solicit translations.
303 Write a small emulation of interactive ftp as a Pythonn program
304 that calls rsync. Commands such as "cd", "ls", "ls *.c" etc map
305 fairly directly into rsync commands: it just needs to remember the
306 current host, directory and so on. We can probably even do
307 completion of remote filenames.