fsvs (1) - Linux Manuals
fsvs: fast versioning tool
NAME
FSVS - fast versioning tool
SYNOPSIS
fsvs command [options] [args]
The following commands are understood by fsvs:
Local configuration and information:
- urls
- Define working copy base directories by their URL(s)
- status
- Get a list of changed entries
- info
- Display detailed information about single entries
- log
- Fetch the log messages from the repository
- diff
- Get differences between files (local and remote)
- copyfrom-detect
- Ask fsvs about probably copied/moved/renamed entries; see cp
Defining which entries to take:
- ignore and rign
- Define ignore patterns
- unversion
- Remove entries from versioning
- add
- Add entries that would be ignored
- cp, mv
- Tell fsvs that entries were copied
Commands working with the repository:
- commit
- Send changed data to the repository
- update
- Get updates from the repository
- checkout
- Fetch some part of the repository, and register it as working copy
- cat
- Get a file from the directory
- revert and uncp
- Undo local changes and entry markings
- remote-status
- Ask what an update would bring
Property handling:
- prop-set
- Set user-defined properties
- prop-get
- Ask value of user-defined properties
- prop-list
- Get a list of user-defined properties
Additional commands used for recovery and debugging:
- export
- Fetch some part of the repository
- sync-repos
- Drop local information about the entries, and fetch the current list from the repository.
Note:
- Multi-url-operations are relatively new; there might be rough edges.
The return code is 0 for success, or 2 for an error. 1 is returned if the option Stopping status reports as soon as changes are found is used, and changes are found; see also Filtering entries.
Universal options
-V -- show version
-V makes fsvs print the version and a copyright notice, and exit.-d and -D -- debugging
If fsvs was compiled using --enable-debug you can enable printing of debug messages (to STDOUT) with -d. Per default all messages are printed; if you're only interested in a subset, you can use -D start-of-function-name.
fsvs -d -D waa_ status
Furthermore you can specify the debug output destination with the option debug_output. This can be a simple filename (which gets truncated), or, if it starts with a |, a command that the output gets piped into.
If the destination cannot be opened (or none is given), debug output goes to STDOUT.
Note:
An example: writing the last 200 lines of debug output into a file.
It requires a specification at the end, which can be any combination of any, text, new, deleted (or removed), meta, mtime, group, mode, changed or owner.
By giving eg. the value text, with a status action only entries that are new or changed are shown; with mtime,group only entries whose group or modification time has changed are printed.
Note:
If an entry gets replaced with an entry of a different type (eg. a directory gets replaced by a file), that counts as deleted and new.
If you use -v, it's used as a any internally.
If you use the string none, it resets the bitmask to no entries shown; then you can built a new mask. So owner,none,any,none,delete would show deleted entries. If the value after all commandline parsing is none, it is reset to the default.
The general format here is specification = action, where specification is a string matching the start of at least one of the defined situations, and action is one of these:
If specification matches more than one situation, all of them are set; eg. for meta=ignore all of meta-mtime, meta-user etc. are ignored.
If at least a single warning that is not ignored is encountered during the program run, a list of warnings along with the number of messages it would have printed with the setting always is displayed, to inform the user of possible problems.
The following situations can be handled with this: meta-mtime, meta-user, meta-group, meta-umask These warnings are issued if a meta-data property that was fetched from the repository couldn't be parsed. This can only happen if some other program or a user changes properties on entries.
no-urllist This warning is issued if a info action is executed, but no URLs have been defined yet.
charset-invalid If the function nl_langinfo(3) couldn't return the name of the current character encoding, a default of UTF-8 is used. You might need that for a minimal system installation, eg. on recovery.
chmod-eperm, chown-eperm If you update a working copy as normal user, and get to update a file which has another owner but you may modify, you'll get errors because neither the user, group, nor mode can be set.
chmod-other, chown-other If you get another error than EPERM in the situation above, you might find these useful.
overlayed-entries This is not yet used.
mixed-rev-wc If you specify some revision number on a revert, it will complain that mixed-revision working copies are not allowed. By using this specification you cannot enable mixed-revision working copies, of course, but you can avoid getting told every time.
propname-reserved It is normally not allowed to set a property with the prop-set action with a name matching some reserved prefixes.
diff-status GNU diff has defined that it returns an exit code 2 in case of an error; sadly it returns that also for binary files, so that a simply fsvs diff some-binary-file text-file would abort without printing the diff for the second file. So the exit status of diff is per default ignored, but can be used by setting this option to eg. stop.
Also an environment variable FSVS_WARNINGS is used and parsed; it is simply a whitespace-separated list of option specifications.
If you have more than a single URL in use for your working copy, update normally updates all entries from all URLs. By using this parameter you can tell FSVS to update only the specified URLs.
The parameter can be used repeatedly; the value can have multiple URLs, separated by whitespace or one of ',;'.
This would get HEAD of base_install and gcc, and set the target revision of the boot URL at 32.
Note:
For a list of these please see Further options for FSVS..
With this command you can explicitly define entries to be versioned, even if they have a matching ignore pattern. They will be sent to the repository on the next commit, just like other new entries, and will therefore be reported as New .
So you tell fsvs to not ignore these entries:
Note:
This command flags the given paths locally as removed. On the next commit they will be deleted in the repository, and the local information of them will be removed, but not the entries themselves. So they will show up as New again, and you get another chance at ignoring them.
Then you do a
Now fsvs would report these files as New , as it does no longer know anything about them; but that can be cured by
The example also shows why the given paths are not just entered as separate ignore patterns - they are just single cases of a (probably) much broader pattern.
Note:
This is used mainly for debugging. It traverses the filesystem and build a new entries file. In production it should not be used - as the revision of the entries is unknown, we can only use 0 - and loose information this way!
This command delays execution until the time has passed at least to the next second after writing the dir and urls files. So, where previously the delay option was used, this can be substituted by the given command followed by the delay command.
The advantage is over the Waiting for a time change after working copy operations option is, that read-only commands can be used in the meantime.
An example:
In the testing framework it is used to save a bit of time; in normal operation, where fsvs commands are not so tightly packed, it is normally preferable to use the delay option.
Fetches a file with the specified revision or, if not given, BASE, from the repository, and outputs it to STDOUT.
Sets one or more URLs for the current working directory (or the directory path), and does an checkout of these URLs.
Example:
The distinction whether a directory is given or not is done based on the result of URL-parsing -- if it looks like an URL, it is used as an URL.
If no directory is given, . is used; this differs from the usual subversion usage, but might be better suited for usage as a recovery tool (where versioning / is common). Opinions welcome.
The given path must exist, and should be empty -- fsvs will abort on conflicts, ie. if files that should be created already exist.
For a format definition of the URLs please see the chapter Format of URLs and the urls and update commands.
Furthermore you might be interested in Using an alternate root directory and Recovery for a non-booting system.
Commits the current state into the repository. It is possible to commit only parts of a working copy into the repository.
Your working copy is /etc , and you've set it up and committed already. Now you've changed /etc/hosts , and /etc/inittab . Since these are non-related changes, you'd like them to be in separate commits.
So you simply run these commands:
If you're currently in /etc , you can even drop the /etc/ in front, and just use the filenames.
Please see status for explanations on -v and -C . For advanced backup usage see also FSVS_PROP_COMMIT_PIPE.
The copy command marks DEST as a copy of SRC at revision rev, so that on the next commit of DEST the corresponding source path is sent as copy source.
The default value for rev is BASE, ie. the revision the SRC (locally) is at.
Please note that this command works always on a directory structure - if you say to copy a directory, the whole structure is marked as copy. That means that if some entries below the copy are missing, they are reported as removed from the copy on the next commit.
Note:
If this command are used without parameters, the currently defined relations are printed; please keep in mind that the key is the destination name, ie. the 2nd line of each pair!
The input format for load is newline-separated - first a SRC line, followed by a DEST line, then an line with just a dot ('.') as delimiter. If you've got filenames with newlines or other special characters, you have to give the paths as arguments.
Internally the paths are stored relative to the working copy base directory, and they're printed that way, too.
Later definitions are appended to the internal database; to undo mistakes, use the revert action.
Note:
As subversion currently treats a rename as copy+delete, the mv command is an alias to cp.
If you have a need to give the filenames dump or load as first parameter for copyfrom relations, give some path, too, as in ./dump.
Note:
This command tells fsvs to look through the new entries, and see whether it can find some that seem to be copied from others already known.
This is just for information purposes and doesn't change any FSVS state, unless some option/parameter is set. (TODO)
The list format is on purpose incompatible with the load syntax, as the best match normally has to be taken manually.
If verbose is used, an additional value giving the percentage of matching blocks, and the count of possibly copied entries is printed.
Example:
The abbreviations are: md5 The MD5 of the new file is identical to that of one or more already committed files; there is no percentage.
inode The device/inode number is identical to the given known entry; this could mean that the old entry has been renamed or hardlinked. Note: Not all filesystems have persistent inode numbers (eg. NFS) - so depending on your filesystems this might not be a good indicator!
name The entry has the same name as another entry.
manber Analysing files of similar size shows some percentage of (variable-sized) common blocks (ignoring the order of the blocks).
dirlist The new directory has similar files to the old directory.
Note:
If too many possible matches are found, not all may be printed; only the indicator ... is shown at the end.
The uncopy command removes a copyfrom mark from the destination entry. This will make the entry unknown again, and reported as New on the next invocations.
Only the base of a copy can be un-copied; if a directory structure was copied, and the given entry is just implicitly copied, this command will give you an error.
This is not folded in revert, because it's not clear whether revert should restore the original copyfrom data or remove the copy attribute; by using a special command this is no longer ambiguous.
Example:
This command gives you diffs between local and repository files.
With -v the meta-data is additionally printed, and changes shown.
If you don't give the revision arguments, you get a diff of the base revision in the repository (the last commit) against your current local file. With one revision, you diff this repository version against you local file. With both revisions given, the difference between these repository versions is calculated.
You'll need the diff program, as the files are simply passed as parameters to it.
The default is to do non-recursive diffs; so fsvs diff . will output the changes in all files in the current directory.
The output for non-files is not defined.
For entries marked as copy the diff against the (clean) source entry is printed.
Please see also Options relating to the 'diff' action and Using colordiff.
If you want to export a directory from your repository without having to have an WAA-area, you can use this command. This restores all meta-data - owner, group, access mask and modification time. Its primary use is for data recovery.
The data gets written (in the correct directory structure) below the current working directory; if entries already exist, the export will stop, so this should be an empty directory.
This command shows general or specific help (for the given command). A similar function is available by using -h or -? after a command.
This command adds patterns to the end of the ignore list, or, with prepend , puts them at the beginning of the list. With at=x the patterns are inserted at the position x , counting from 0.
fsvs dump prints the patterns to STDOUT . If there are special characters like CR or LF embedded in the pattern without encoding (like \r or \n), the output will be garbled.
The patterns may include * and ? as wildcards in one directory level, or ** for arbitrary strings.
These patterns are only matched against new files; entries that are already versioned are not invalidated. If the given path matches a new directory, entries below aren't found, either; but if this directory or entries below are already versioned, the pattern doesn't work, as the match is restricted to the directory.
So:
Other special variants are available, see the documentation Using ignore patterns .
Examples:
Note:
If you use more than a single working copy for the same data, it will be stored in different paths - and that makes absolute ignore patterns infeasible. But relative ignore patterns are anchored at the beginning of the WC root - which is a bit tiring if you're deep in your WC hierarchy and want to ignore some files.
To make that easier you can use the rel-ignore (abbreviated as ri) command; this converts all given path-specifications (that may include wildcards as per the shell pattern specification above) to WC-relative values before storing them.
Example for /etc as working copy root:
Note:
For more details about ignoring files please see the ignore command and Using ignore patterns.
Use this command to show information regarding one or more entries in your working copy. Currently you must be at the working copy root; but that will change. You can use -v to obtain slightly more information.
This may sometimes be helpful for locating bugs, or to obtain the URL and revision a working copy is currently at.
Example:
The default is to print information about the given entry only. With a single -R you'll get this data about all entries of a given directory; with a second -R you'll get the whole (sub-)tree.
This command views the log information associated with the given path, or, if none, the highest priority URL.
The optional rev1 and rev2 can be used to restrict the revisions that are shown; if no values are given, the logs are given starting from HEAD downwards, and then a limit on the number of revisions is applied (but see the limit option).
If you use the -v -option, you get the files changed in each revision printed, too.
There is an option controlling the output format; see 'fsvs log' output format.
TODOs:
You get the data of the property printed to STDOUT.
Note:
This command sets an arbitrary property value for the given path(s).
Note:
This command removes property value for the given path(s).
See also prop-set.
Lists the names of all properties for the given entry. With -v, the value is printed as well; special characters will be translated, to not mess with your terminal.
If you need raw output, post a patch for --raw, or loop with prop-get.
This command looks into the repository and tells you which files would get changed on an update - it's a dry-run for update .
Per default it compares to HEAD, but you can choose another revision with the -r parameter.
When FSVS tries to update local files which have been changed, a conflict might occur. (For various ways of handling these please see the conflict option.)
This command lets you mark such conflicts as resolved.
This command undoes local modifications:
See also HOWTO: Understand the entries' statii.
If a directory is given on the command line all known entries in this directory are reverted to the old state; this behaviour can be modified with -R/-N, or see below.
The reverted entries are printed, along with the status they had before the revert (because the new status is per definition unchanged).
If a revision is given, the entries' data is taken from this revision; furthermore, the new status of that entry is shown.
Note:
In contrast, if you update to an older version, you
If the user specifies a directory entry, see this table for the restoration results: command line switchresult -N this directory only (meta-data), none this directory, and direct children of the directory, -R this directory, and the complete tree below.
This command shows the entries that have changed since the last commit.
The output is formatted as follows:
The status column can show the following flags:
Here's a table with the characters and their positions:
Furthermore please take a look at Status output coloring.
This command loads the file list from the repository. A following commit will send all differences and make the repository data identical to the local.
This is normally not needed; the use cases are
It is (currently) important if you want to backup two similar machines. Then you can commit one machine into a subdirectory of your repository, make a copy of that directory for another machine, and sync this other directory on the other machine.
A commit then will transfer only _changed_ files; so if the two machines share 2GB of binaries (/usr , /bin , /lib , ...) then these 2GB are still shared in the repository, although over time they will deviate (as both committing machines know nothing of the other path with identical files).
This kind of backup could be substituted by several levels of repository paths, which get 'overlayed' in a defined priority. So the base directory, which all machines derive from, will be committed from one machine, and it's no longer necessary for all machines to send identical files into the repository.
The revision argument should only ever be used for debugging; if you fetch a filelist for a revision, and then commit against later revisions, problems are bound to occur.
Note:
This command does an update on all specified URLs for the current working copy, or, if none is given via -u, all URLs.
It first reads all changes in the repositories, overlays them (so that only the highest-priority entries are used), and fetches all necessary changes.
Initializes a working copy administrative area and connects the current working directory to REPOS_URL. All commits and updates will be done to this directory and against the given URL.
Example:
For a format definition of the URLs please see the chapter Format of URLs.
Note:
Example:
Empty lines are ignored.
As an optional parameter you can give a format statement; p , n , r, t and u are substituted by the priority, name, current revision, target revision and URL. Note: That's not a real printf()-format; only these and a few \ sequences are recognized.
Example:
The default format is 'N:%n,P:%p,D:%t,%u\\n'; for a more readable version you can use -v.
fsvs -o debug_output='| tail -200 > /tmp/debug.log' -d ....
-N, -R -- recursion
The -N and -R switches in effect just decrement/increment a counter; the behavious is chosen depending on that. So -N -N -N -R -R is equivalent to -N.
-q, -v -- verbose/quiet
Like the options for recursive behaviour (-R and -N) -v and -q just inc/decrement a counter. The higher the value, the more verbose.
-C -- checksum
-C chooses to use more change detection checks; please see the change_check option for more details.
-f -- filter entries
This parameter allows to do a bit of filtering of entries, or, for some operations, modification of the work done on given entries.
-W warning=action -- set warnings
Here you can define the behaviour for certain situations that should not normally happen, but which you might encounter.
-u URLname[@revision[:revision]] -- select URLs
Some commands can be reduced to a subset of defined URLs; the update command is a example.
fsvs up -u base_install,boot@32 -u gcc
-o [name[=value]] -- other options
This is used for setting some seldom used option, for which default can be set in a configuration file (to be implemented, currently only command-line).
add
fsvs add PATH [PATH...]
Example
Say, you're versioning your home directory, and gave an ignore pattern of ./.* to ignore all .* entries in your home-directory. Now you want .bashrc, .ssh/config, and your complete .kde3-tree saved, just like other data.
fsvs add .bashrc .ssh/config .kde3
fsvs ignore --insert t./.kde3
unversion
fsvs unversion PATH [PATH...]
Example
Say, you're versioning your home directory, and found that you no longer want .bash_history and .sh_history versioned. So you do
fsvs unversion .bash_history .sh_history
fsvs commit
fsvs ignore './.*sh_history'
_build_new_list
delay
fsvs commit /etc/X11 -m 'Backup of X11'
... read-only commands, like 'status'
fsvs delay /etc/X11
... read-write commands, like 'commit'
cat
fsvs cat [-r rev] path
checkout
fsvs checkout [path] URL [URLs...]
fsvs checkout . http://svn/repos/installation/machine-1/trunk
commit
fsvs commit [-m 'message'|-F filename] [-v] [-C [-C]] [PATH [PATH ...]]
fsvs commit -m 'Added some host' /etc/hosts
fsvs commit -m 'Tweaked default runlevel' /etc/inittab
cp
fsvs cp [-r rev] SRC DEST
fsvs cp dump
fsvs cp load
$ fsvs cp a b
$ rm a/1
$ fsvs ci a
$ fsvs ci b
copyfrom-detect
fsvs copyfrom-detect [paths...]
$ fsvs copyfrom-list -v
newfile1
md5:oldfileA
newfile2
md5:oldfileB
md5:oldfileC
md5:oldfileD
newfile3
inode:oldfileI
manber=82.6:oldfileF
manber=74.2:oldfileG
manber=53.3:oldfileH
...
3 copyfrom relations found.
uncp
fsvs uncopy DEST [DEST ...]
$ fsvs copy SourceFile DestFile
# Whoops, was wrong!
$ fsvs uncopy DestFile
diff
fsvs diff [-v] [-r rev[:rev2]] [-R] PATH [PATH...]
export
fsvs export REPOS_URL [-r rev]
help
help [command]
ignore
fsvs ignore dump|load
fsvs ignore [prepend|append|at=n] pattern [pattern ...]
fsvs ignore ./tmp
fsvs ignore ./tmp/**
fsvs ignore ./proc
fsvs ignore ./dev/pts
fsvs ignore './var/log/*-*'
fsvs ignore './**~'
fsvs ignore './**/*.bak'
fsvs ignore prepend 't./**.txt'
fsvs ignore append 't./**.svg'
fsvs ignore at=1 './**.tmp'
fsvs ignore dump
fsvs ignore dump -v
echo './**.doc' | fsvs ignore load
rign
fsvs rel-ignore [prepend|append|at=n] path-spec [path-spec ...]
fsvs ri [prepend|append|at=n] path-spec [path-spec ...]
fsvs rel-ignore '/etc/X11/xorg.conf.*'
cd /etc/X11
fsvs rel-ignore 'xorg.conf.*'
info
fsvs info [-R [-R]] [PATH...]
$ fsvs info
URL: file:
.... 200 .
Type: directory
Status: 0x0
Flags: 0x100000
Dev: 0
Inode: 24521
Mode: 040755
UID/GID: 1000/1000
MTime: Thu Aug 17 16:34:24 2006
CTime: Thu Aug 17 16:34:24 2006
Revision: 4
Size: 200
log
fsvs log [-v] [-r rev1[:rev2]] [path]
prop-get
fsvs prop-get PROPERTY-NAME PATH...
prop-set
fsvs prop-set PROPERTY-NAME VALUE PATH...
prop-del
fsvs prop-del PROPERTY-NAME PATH...
prop-list
fsvs prop-list [-v] PATH...
remote-status
fsvs remote-status PATH [-r rev]
resolve
fsvs resolve PATH [PATH...]
revert
fsvs revert [-rRev] [-R] PATH [PATH...]
Difference to update
If you find that something doesn't work as it should, you can revert entries until you are satisfied, and directly commit the new state.
Currently only known entries are handled.
If you need a switch (like --delete in rsync(1) ) to remove unknown (new, not yet versioned) entries, to get the directory in the exact state it is in the repository, say so.
If a path is specified whose parent is missing, \c
fsvs complains. We plan to provide a switch (probably -p), which would create (a sparse) tree up to this entry.
Recursive behaviour
When the user specifies a non-directory entry (file, device, symlink), this entry is reverted to the old state. This is the easy case.
Working with copied entries
If an entry is marked as copied from another entry (and not committed!), a revert will fetch the original copyfrom source. To undo the copy setting use the uncp command.
status
fsvs status [-C [-C]] [-v] [PATHs...]
* Without -v With -v
* .... ......
* NmC? NtpPC?
* DPx! D x!
* R + R +
* d d
* n n
*
sync-repos
fsvs sync-repos [-r rev] [working copy base]
update
fsvs update [-r rev] [working copy base]
fsvs update [-u url@rev ...] [working copy base]
urls
fsvs urls URL [URLs...]
fsvs urls dump
fsvs urls load
fsvs urls http://svn/repos/installation/machine-1/trunk
echo '' | fsvs urls load
Loading URLs
You can load a list of URLs from STDIN; use the load subcommand for that.
( echo 'N:local,prio:10,http://svn/repos/install/machine-1/trunk' ;
echo 'P:50,name:common,http://svn/repos/install/common/trunk' ) |
fsvs urls load
Dumping the defined URLs
To see which URLs are in use for the current WC, you can use dump.
fsvs urls dump ' %u %n:%p\n'
http://svn/repos/installation/machine-1/trunk local:10
http://svn/repos/installation/common/trunk common:50