3 BUGS ---------------------------------------------------------------
4 Fix hardlink reporting 2002/03/25
5 Fix progress indicator to not corrupt log
7 Do not rely on having a group called "nobody"
8 Incorrect timestamps (Debian #100295)
11 FEATURES ------------------------------------------------------------
12 server-imposed bandwidth limits
14 Use chroot only if supported
15 Allow supplementary groups in rsyncd.conf 2002/04/09
16 Handling IPv6 on old machines
18 Add ACL support 2001/12/02
19 Lazy directory creation
20 Conditional -z for old protocols
21 proxy authentication 2002/01/23
24 Allow forcing arbitrary permissions 2002/03/12
25 --diff david.e.sewell 2002/03/15
26 Add daemon --no-detach and --no-fork options
27 Create more granular verbosity jw 2003/05/15
29 DOCUMENTATION --------------------------------------------------------
31 Keep list of open issues and todos on the web site
32 Update web site from CVS
33 Perhaps redo manual as SGML
35 LOGGING --------------------------------------------------------------
36 Make dry run list all updates 2002/04/03
38 Improve error messages
39 Better statistics: Rasmus 2002/03/08
40 Perhaps flush stdout like syslog
41 Log deamon sessions that just list modules
42 Log child death on signal
43 Keep stderr and stdout properly separated (Debian #23626)
44 Log errors with function that reports process of origin
45 verbose output David Stein 2001/12/20
46 Add reason for transfer to file logging
47 debugging of daemon 2002/04/08
50 DEVELOPMENT --------------------------------------------------------
51 Handling duplicate names
52 Use generic zlib 2002/02/25
57 Add machines to build farm
59 PERFORMANCE ----------------------------------------------------------
60 File list structure in memory
61 Traverse just one directory at a time
63 Allow skipping MD4 file_sum 2002/04/08
67 TESTING --------------------------------------------------------------
69 Cross-test versions 2001/08/22
72 Create mutator program for testing
73 Create configure option to enable dangerous tests
74 If tests are skipped, say why.
75 Test daemon feature to disallow particular options.
76 Create pipe program for testing
77 Create test makefile target for some tests
78 Test "refuse options" works
80 RELATED PROJECTS -----------------------------------------------------
82 http://rsync.samba.org/rsync-and-debian/
84 rsyncsplit as alternative to real integration with gzip?
85 reverse rsync over HTTP Range
89 BUGS ---------------------------------------------------------------
91 Fix hardlink reporting 2002/03/25
92 (was: There seems to be a bug with hardlinks)
94 mbp/2 build$ ls -l /tmp/a /tmp/b -i
97 2568307 -rw-rw-r-- 3 mbp mbp 29 Mar 25 17:30 a1
98 2568307 -rw-rw-r-- 3 mbp mbp 29 Mar 25 17:30 a2
99 2568307 -rw-rw-r-- 3 mbp mbp 29 Mar 25 17:30 a3
100 2568310 -rw-rw-r-- 5 mbp mbp 29 Mar 25 17:30 a4
101 2568310 -rw-rw-r-- 5 mbp mbp 29 Mar 25 17:30 a5
102 2568310 -rw-rw-r-- 5 mbp mbp 29 Mar 25 17:30 b1
103 2568310 -rw-rw-r-- 5 mbp mbp 29 Mar 25 17:30 b2
104 2568310 -rw-rw-r-- 5 mbp mbp 29 Mar 25 17:30 b3
108 2568309 -rw-rw-r-- 3 mbp mbp 29 Mar 25 17:30 a1
109 2568309 -rw-rw-r-- 3 mbp mbp 29 Mar 25 17:30 a2
110 2568309 -rw-rw-r-- 3 mbp mbp 29 Mar 25 17:30 a3
111 2568311 -rw-rw-r-- 5 mbp mbp 29 Mar 25 17:30 a4
112 2568311 -rw-rw-r-- 5 mbp mbp 29 Mar 25 17:30 a5
113 2568311 -rw-rw-r-- 5 mbp mbp 29 Mar 25 17:30 b1
114 2568311 -rw-rw-r-- 5 mbp mbp 29 Mar 25 17:30 b2
115 2568311 -rw-rw-r-- 5 mbp mbp 29 Mar 25 17:30 b3
116 mbp/2 build$ rm -r /tmp/b && ./rsync -avH /tmp/a/ /tmp/b
117 building file list ... done
118 created directory /tmp/b
124 wrote 350 bytes read 52 bytes 804.00 bytes/sec
125 total size is 232 speedup is 0.58
126 mbp/2 build$ rm -r /tmp/b
127 mbp/2 build$ ls -l /tmp/b
128 ls: /tmp/b: No such file or directory
129 mbp/2 build$ rm -r /tmp/b && ./rsync -avH /tmp/a/ /tmp/b
130 rm: cannot remove `/tmp/b': No such file or directory
131 mbp/2 build$ rm -f -r /tmp/b && ./rsync -avH /tmp/a/ /tmp/b
132 building file list ... done
133 created directory /tmp/b
139 wrote 350 bytes read 52 bytes 804.00 bytes/sec
140 total size is 232 speedup is 0.58
141 mbp/2 build$ ls -l /tmp/b
143 -rw-rw-r-- 3 mbp mbp 29 Mar 25 17:30 a1
144 -rw-rw-r-- 3 mbp mbp 29 Mar 25 17:30 a2
145 -rw-rw-r-- 3 mbp mbp 29 Mar 25 17:30 a3
146 -rw-rw-r-- 5 mbp mbp 29 Mar 25 17:30 a4
147 -rw-rw-r-- 5 mbp mbp 29 Mar 25 17:30 a5
148 -rw-rw-r-- 5 mbp mbp 29 Mar 25 17:30 b1
149 -rw-rw-r-- 5 mbp mbp 29 Mar 25 17:30 b2
150 -rw-rw-r-- 5 mbp mbp 29 Mar 25 17:30 b3
151 mbp/2 build$ ls -l /tmp/a
153 -rw-rw-r-- 3 mbp mbp 29 Mar 25 17:30 a1
154 -rw-rw-r-- 3 mbp mbp 29 Mar 25 17:30 a2
155 -rw-rw-r-- 3 mbp mbp 29 Mar 25 17:30 a3
156 -rw-rw-r-- 5 mbp mbp 29 Mar 25 17:30 a4
157 -rw-rw-r-- 5 mbp mbp 29 Mar 25 17:30 a5
158 -rw-rw-r-- 5 mbp mbp 29 Mar 25 17:30 b1
159 -rw-rw-r-- 5 mbp mbp 29 Mar 25 17:30 b2
160 -rw-rw-r-- 5 mbp mbp 29 Mar 25 17:30 b3
165 Fix progress indicator to not corrupt log
167 Progress indicator can produce corrupt output when transferring directories:
170 main/binary-arm/admin/
171 main/binary-arm/base/
172 main/binary-arm/comm/8.56kB/s 0:00:52
173 main/binary-arm/devel/
175 main/binary-arm/editors/
176 main/binary-arm/electronics/s 0:00:53
177 main/binary-arm/games/
178 main/binary-arm/graphics/
179 main/binary-arm/hamradio/
180 main/binary-arm/interpreters/
181 main/binary-arm/libs/6.61kB/s 0:00:54
182 main/binary-arm/mail/
183 main/binary-arm/math/
184 main/binary-arm/misc/
191 I don't think we handle this properly on systems that don't have the
192 call. Are there any such?
197 Do not rely on having a group called "nobody"
199 http://www.linuxbase.org/spec/refspecs/LSB_1.1.0/gLSB/usernames.html
201 On Debian it's "nogroup"
206 Incorrect timestamps (Debian #100295)
208 A bit hard to believe, but apparently it happens.
215 Don't detach, because this messes up --srvany.
217 http://sources.redhat.com/ml/cygwin/2001-08/msg00234.html
223 FEATURES ------------------------------------------------------------
225 server-imposed bandwidth limits
232 There are already some patches to do this.
234 BitKeeper uses a server whose login shell is set to bkd. That's
235 probably a reasonable approach.
240 Use chroot only if supported
242 If the platform doesn't support it, then don't even try.
244 If running as non-root, then don't fail, just give a warning.
245 (There was a thread about this a while ago?)
247 http://lists.samba.org/pipermail/rsync/2001-August/thread.html
248 http://lists.samba.org/pipermail/rsync/2001-September/thread.html
253 Allow supplementary groups in rsyncd.conf 2002/04/09
255 Perhaps allow supplementary groups to be specified in rsyncd.conf;
256 then make the first one the primary gid and all the rest be
262 Handling IPv6 on old machines
264 The KAME IPv6 patch is nice in theory but has proved a bit of a
265 nightmare in practice. The basic idea of their patch is that rsync
266 is rewritten to use the new getaddrinfo()/getnameinfo() interface,
267 rather than gethostbyname()/gethostbyaddr() as in rsync 2.4.6.
268 Systems that don't have the new interface are handled by providing
269 our own implementation in lib/, which is selectively linked in.
271 The problem with this is that it is really hard to get right on
272 platforms that have a half-working implementation, so redefining
273 these functions clashes with system headers, and leaving them out
274 breaks. This affects at least OSF/1, RedHat 5, and Cobalt, which
275 are moderately improtant.
277 Perhaps the simplest solution would be to have two different files
278 implementing the same interface, and choose either the new or the
279 old API. This is probably necessary for systems that e.g. have
280 IPv6, but gethostbyaddr() can't handle it. The Linux manpage claims
281 this is currently the case.
283 In fact, our internal sockets interface (things like
284 open_socket_out(), etc) is much narrower than the getaddrinfo()
285 interface, and so probably simpler to get right. In addition, the
286 old code is known to work well on old machines.
288 We could drop the rather large lib/getaddrinfo files.
295 Implement suggestions from http://www.kame.net/newsletter/19980604/
296 and ftp://ftp.iij.ad.jp/pub/RFC/rfc2553.txt
298 If a host has multiple addresses, then listen try to connect to all
299 in order until we get through. (getaddrinfo may return multiple
300 addresses.) This is kind of implemented already.
302 Possibly also when starting as a server we may need to listen on
303 multiple passive addresses. This might be a bit harder, because we
304 may need to select on all of them. Hm.
306 Define a syntax for IPv6 literal addresses. Since they include
307 colons, they tend to break most naming systems, including ours.
308 Based on the HTTP IPv6 syntax, I think we should use
310 rsync://[::1]/foo/bar [::1]::bar
312 which should just take a small change to the parser code.
317 Add ACL support 2001/12/02
319 Transfer ACLs. Need to think of a standard representation.
320 Probably better not to even try to convert between NT and POSIX.
321 Possibly can share some code with Samba.
326 Lazy directory creation
328 With the current common --include '*/' --exclude '*' pattern, people
329 can end up with many empty directories. We might avoid this by
330 lazily creating such directories.
335 Conditional -z for old protocols
337 After we get the @RSYNCD greeting from the server, we know it's
338 version but we have not yet sent the command line, so we could just
339 remove the -z option if the server is too old.
341 For ssh invocation it's not so simple, because we actually use the
342 command line to start the remote process. However, we only actually
343 do compression in token.c, and we could therefore once we discover
344 the remote version emit an error if it's too old. I'm not sure if
345 that's a good tradeoff or not.
350 proxy authentication 2002/01/23
352 Allow RSYNC_PROXY to be http://user:pass@proxy.foo:3128/, and do
353 HTTP Basic Proxy-Authentication.
355 Multiple schemes are possible, up to and including the insanity that
356 is NTLM, but Basic probably covers most cases.
363 Add --with-socks, and then perhaps a command-line option to put them
364 on or off. This might be more reliable than LD_PRELOAD hacks.
371 rsync to a FAT partition on a Unix machine doesn't work very well at
372 the moment. I think we get errors about invalid filenames and
373 perhaps also trying to do atomic renames.
375 I guess the code to do this is currently #ifdef'd on Windows;
376 perhaps we ought to intelligently fall back to it on Unix too.
381 Allow forcing arbitrary permissions 2002/03/12
383 On 12 Mar 2002, Dave Dykstra <dwd@bell-labs.com> wrote:
384 > If we would add an option to do that functionality, I
385 > would vote for one that was more general which could mask
386 > off any set of permission bits and possibly add any set of
387 > bits. Perhaps a chmod-like syntax if it could be
388 > implemented simply.
390 I think that would be good too. For example, people uploading files
391 to a web server might like to say
393 rsync -avzP --chmod a+rX ./ sourcefrog.net:/home/www/sourcefrog/
395 Ideally the patch would implement as many of the gnu chmod semantics
396 as possible. I think the mode parser should be a separate function
397 that passes back something like (mask,set) description to the rest
398 of the program. For bonus points there would be a test case for the
401 Possibly also --chown
408 --diff david.e.sewell 2002/03/15
410 Allow people to specify the diff command. (Might want to use wdiff,
413 Just diff the temporary file with the destination file, and delete
414 the tmp file rather than moving it into place.
416 Interaction with --partial.
418 Security interactions with daemon mode?
423 Add daemon --no-detach and --no-fork options
425 Very useful for debugging. Also good when running under a
426 daemon-monitoring process that tries to restart the service when the
432 Create more granular verbosity jw 2003/05/15
434 Control output with the --report option.
436 The option takes as a single argument (no whitespace) a
437 comma delimited lists of keywords.
439 This would separate debugging from "logging" as well as
440 fine grained selection of statistical reporting and what
443 http://lists.samba.org/archive/rsync/2003-May/006059.html
447 DOCUMENTATION --------------------------------------------------------
454 Keep list of open issues and todos on the web site
459 Update web site from CVS
464 Perhaps redo manual as SGML
466 The man page is getting rather large, and there is more information
467 that ought to be added.
469 TexInfo source is probably a dying format.
471 Linuxdoc looks like the most likely contender. I know DocBook is
472 favoured by some people, but it's so bloody verbose, even with emacs
477 LOGGING --------------------------------------------------------------
479 Make dry run list all updates 2002/04/03
483 Mark Santcroos points out that -n fails to list files which have
484 only metadata changes, though it probably should.
486 There may be a Debian bug about this as well.
493 At exit, show how much memory was used for the file list, etc.
495 Also we do a wierd exponential-growth allocation in flist.c. I'm
496 not sure this makes sense with modern mallocs. At any rate it will
497 make us allocate a huge amount of memory for large file lists.
502 Improve error messages
504 If we hang or get SIGINT, then explain where we were up to. Perhaps
505 have a static buffer that contains the current function name, or
506 some kind of description of what we were trying to do. This is a
507 little easier on people than needing to run strace/truss.
509 "The dungeon collapses! You are killed." Rather than "unexpected
510 eof" give a message that is more detailed if possible and also more
513 If we get an error writing to a socket, then we should perhaps
514 continue trying to read to see if an error message comes across
515 explaining why the socket is closed. I'm not sure if this would
516 work, but it would certainly make our messages more helpful.
518 What happens if a directory is missing -x attributes. Do we lose
519 our load? (Debian #28416) Probably fixed now, but a test case would
527 Better statistics: Rasmus 2002/03/08
530 hey, how about an rsync option that just gives you the
531 summary without the list of files? And perhaps gives
532 more information like the number of new files, number
533 of changed, deleted, etc. ?
536 nice idea there is --stats but at the moment it's very
537 tridge-oriented rather than user-friendly it would be
538 nice to improve it that would also work well with
544 Perhaps flush stdout like syslog
546 Perhaps flush stdout after each filename, so that people trying to
547 monitor progress in a log file can do so more easily. See
548 http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=48108
553 Log deamon sessions that just list modules
555 At the connections that just get a list of modules are not logged,
561 Log child death on signal
563 If a child of the rsync daemon dies with a signal, we should notice
564 that when we reap it and log a message.
569 Keep stderr and stdout properly separated (Debian #23626)
574 Log errors with function that reports process of origin
576 Use a separate function for reporting errors; prefix it with
577 "rsync:" or "rsync(remote)", or perhaps even "rsync(local
583 verbose output David Stein 2001/12/20
585 Indicate whether files are new, updated, or deleted
587 At end of transfer, show how many files were or were not transferred
593 Add reason for transfer to file logging
595 Explain *why* every file is transferred or not (e.g. "local mtime
596 123123 newer than 1283198")
601 debugging of daemon 2002/04/08
603 Add an rsyncd.conf parameter to turn on debugging on the server.
610 Change to using gettext(). Probably need to ship this for platforms
613 Solicit translations.
615 Does anyone care? Before we bother modifying the code, we ought to
616 get the manual translated first, because that's possibly more useful
617 and at any rate demonstrates desire.
621 DEVELOPMENT --------------------------------------------------------
623 Handling duplicate names
625 We need to be careful of duplicate names getting into the file list.
626 See clean_flist(). This could happen if multiple arguments include
629 I think duplicates are only a problem if they're both flowing
630 through the pipeline at the same time. For example we might have
631 updated the first occurrence after reading the checksums for the
632 second. So possibly we just need to make sure that we don't have
633 both in the pipeline at the same time.
635 Possibly if we did one directory at a time that would be sufficient.
637 Alternatively we could pre-process the arguments to make sure no
638 duplicates will ever be inserted. There could be some bad cases
639 when we're collapsing symlinks.
641 We could have a hash table.
643 The root of the problem is that we do not want more than one file
644 list entry referring to the same file. At first glance there are
645 several ways this could happen: symlinks, hardlinks, and repeated
646 names on the command line.
648 If names are repeated on the command line, they may be present in
649 different forms, perhaps by traversing directory paths in different
650 ways, traversing paths including symlinks. Also we need to allow
651 for expansion of globs by rsync.
653 At the moment, clean_flist() requires having the entire file list in
654 memory. Duplicate names are detected just by a string comparison.
656 We don't need to worry about hard links causing duplicates because
657 files are never updated in place. Similarly for symlinks.
659 I think even if we're using a different symlink mode we don't need
662 Unless we're really clever this will introduce a protocol
663 incompatibility, so we need to be able to accept the old format as
669 Use generic zlib 2002/02/25
671 Perhaps don't use our own zlib.
675 - will automatically be up to date with bugfixes in zlib
677 - can leave it out for small rsync on e.g. recovery disks
679 - can use a shared library
681 - avoids people breaking rsync by trying to do this themselves and
684 Should we ship zlib for systems that don't have it, or require
685 people to install it separately?
687 Apparently this will make us incompatible with versions of rsync
688 that use the patched version of rsync. Probably the simplest way to
689 do this is to just disable gzip (with a warning) when talking to old
697 Rather than storing the file list in memory, store it in a TDB.
699 This *might* make memory usage lower while building the file list.
701 Hashtable lookup will mean files are not transmitted in order,
704 This would neatly eliminate one of the major post-fork shared data
712 Build rsync with SPLINT to try to find security holes. Add
713 annotations as necessary. Keep track of the number of warnings
714 found initially, and see how many of them are real bugs, or real
715 security bugs. Knowing the percentage of likely hits would be
716 really interesting for other projects.
723 jra recommends Valgrind:
725 http://devel-home.kde.org/~sewardj/
730 Create release script
736 Build tar file; upload
738 Send announcement to mailing list and c.o.l.a.
740 Make freshmeat announcement
747 Add machines to build farm
749 Cygwin (on different versions of Win32?)
751 HP-UX variants (via HP?)
759 PERFORMANCE ----------------------------------------------------------
761 File list structure in memory
763 Rather than one big array, perhaps have a tree in memory mirroring
766 This might make sorting much faster! (I'm not sure it's a big CPU
769 It might also reduce memory use in storing repeated directory names
770 -- again I'm not sure this is a problem.
775 Traverse just one directory at a time
777 Traverse just one directory at a time. Tridge says it's possible.
779 At the moment rsync reads the whole file list into memory at the
780 start, which makes us use a lot of memory and also not pipeline
781 network access as much as we could.
788 At the moment hardlink handling is very expensive, so it's off by
789 default. It does not need to be so.
791 Since most of the solutions are rather intertwined with the file
792 list it is probably better to fix that first, although fixing
793 hardlinks is possibly simpler.
795 We can rule out hardlinked directories since they will probably
796 screw us up in all kinds of ways. They simply should not be used.
798 At the moment rsync only cares about hardlinks to regular files. I
799 guess you could also use them for sockets, devices and other beasts,
800 but I have not seen them.
802 When trying to reproduce hard links, we only need to worry about
803 files that have more than one name (nlinks>1 && !S_ISDIR).
805 The basic point of this is to discover alternate names that refer to
806 the same file. All operations, including creating the file and
807 writing modifications to it need only to be done for the first name.
808 For all later names, we just create the link and then leave it
811 If hard links are to be preserved:
813 Before the generator/receiver fork, the list of files is received
814 from the sender (recv_file_list), and a table for detecting hard
817 The generator looks for hard links within the file list and does
818 not send checksums for them, though it does send other metadata.
820 The sender sends the device number and inode with file entries, so
821 that files are uniquely identified.
823 The receiver goes through and creates hard links (do_hard_links)
824 after all data has been written, but before directory permissions
827 At the moment device and inum are sent as 4-byte integers, which
828 will probably cause problems on large filesystems. On Linux the
829 kernel uses 64-bit ino_t's internally, and people will soon have
830 filesystems big enough to use them. We ought to follow NFS4 in
831 using 64-bit device and inode identification, perhaps with a
832 protocol version bump.
834 Once we've seen all the names for a particular file, we no longer
835 need to think about it and we can deallocate the memory.
837 We can also have the case where there are links to a file that are
838 not in the tree being transferred. There's nothing we can do about
839 that. Because we rename the destination into place after writing,
840 any hardlinks to the old file are always going to be orphaned. In
841 fact that is almost necessary because otherwise we'd get really
842 confused if we were generating checksums for one name of a file and
845 At the moment the code seems to make a whole second copy of the file
846 list, which seems unnecessary.
848 We should have a test case that exercises hard links. Since it
849 might be hard to compare ./tls output where the inodes change we
850 might need a little program to check whether several names refer to
856 Allow skipping MD4 file_sum 2002/04/08
858 If we're doing a local transfer, or using -W, then perhaps don't
859 send the file checksum. If we're doing a local transfer, then
860 calculating MD4 checksums uses 90% of CPU and is unlikely to be
863 Indeed for transfers over zlib or ssh we can also rely on the
864 transport to have quite strong protection against corruption.
866 Perhaps we should have an option to disable this,
867 analogous to --whole-file, although it would default to
868 disabled. The file checksum takes up a definite space in
869 the protocol -- we can either set it to 0, or perhaps just
877 Perhaps borrow an assembler MD4 from someone?
879 Make sure we call MD4 with properly-sized blocks whenever possible
880 to avoid copying into the residue region?
887 Test whether this is actually faster than just using malloc(). If
888 it's not (anymore), throw it out.
892 TESTING --------------------------------------------------------------
896 Something that just keeps running rsync continuously over a data set
897 likely to generate problems.
902 Cross-test versions 2001/08/22
904 Part of the regression suite should be making sure that we
905 don't break backwards compatibility: old clients vs new
906 servers and so on. Ideally we would test both up and down
907 from the current release to all old versions.
909 Run current rsync versions against significant past releases.
911 We might need to omit broken old versions, or versions in which
912 particular functionality is broken
914 It might be sufficient to test downloads from well-known public
915 rsync servers running different versions of rsync. This will give
916 some testing and also be the most common case for having different
917 versions and not being able to upgrade.
919 The new --protocol option may help in this.
924 Test on kernel source
926 Download all versions of kernel; unpack, sync between them. Also
927 sync between uncompressed tarballs. Compare directories after
930 Use local mode; ssh; daemon; --whole-file and --no-whole-file.
932 Use awk to pull out the 'speedup' number for each transfer. Make
940 Sparse and non-sparse
945 Create mutator program for testing
947 Insert bytes, delete bytes, swap blocks, ...
952 Create configure option to enable dangerous tests
957 If tests are skipped, say why.
962 Test daemon feature to disallow particular options.
967 Create pipe program for testing
969 Create pipe program that makes slow/jerky connections for
970 testing Versions of read() and write() that corrupt the
971 stream, or abruptly fail
976 Create test makefile target for some tests
978 Separate makefile target to run rough tests -- or perhaps
979 just run them every time?
984 Test "refuse options" works
986 What about for --recursive?
988 If you specify an unrecognized option here, you should get an error.
990 We need a test case for this...
992 Was this broken when we changed to popt?
996 RELATED PROJECTS -----------------------------------------------------
1000 Write a small emulation of interactive ftp as a Pythonn program
1001 that calls rsync. Commands such as "cd", "ls", "ls *.c" etc map
1002 fairly directly into rsync commands: it just needs to remember the
1003 current host, directory and so on. We can probably even do
1004 completion of remote filenames.
1009 http://rsync.samba.org/rsync-and-debian/
1015 rsyncable gzip patch
1017 Exhaustive, tortuous testing
1024 rsyncsplit as alternative to real integration with gzip?
1029 reverse rsync over HTTP Range
1031 Goswin Brederlow suggested this on Debian; I think tridge and I
1032 talked about it previous in relation to rproxy.