home | O'Reilly's CD bookshelfs | FreeBSD | Linux | Cisco | Cisco Exam    

Book HomeLinux in a NutshellSearch this book

14.3. CVS Administrator Reference

This section provides details on creating and configuring repositories and performing other CVS administrative tasks. A single computer can run multiple copies of the CVS server, and each server can serve multiple repositories.

14.3.3. Repository Structure

The CVS repository is implemented as a normal directory with special contents. This section describes the contents of the repository directory.

14.3.3.1. The CVSROOT directory

The CVSROOT directory contains the administrative files for the repository; other directories in the repository contain the modules. The administrative files permit (and ignore) blank lines and comment lines in addition to the lines with real configuration information on them. Comment lines start with a hash mark (`#').

Some of the administrative files contain filename patterns to match file and directory names. These patterns are regular expressions like those used in GNU Emacs. Table 14-2 contains the special constructions used most often.

Table 14-2. Filename Pattern Special Constructions

Construction

Description

^

Match the beginning of the string.

$

Match the end of the string.

.

Match any single character.

*

Modify the preceding construct to match zero or more repetitions.

CVS will perform a few important expansions in the contents of the administrative files before interpreting the results. First, the typical shell syntax for referring to a home directory is ~/, which expands to the home directory of the user running CVS; and ~user expands to the home directory of the specified user.

In addition, CVS provides a mechanism similar to the shell's environment variable expansion capability. Constructs such as ${variable} will be replaced by the value of the named variable. Variable names start with letters and consist entirely of letters, numbers, and underscores. Curly brackets may be omitted if the character immediately following the variable reference is not a valid variable name character. While this construct looks like a shell environment variable reference, the full environment is not available. Table 14-3 contains the built-in variables.

Table 14-3. Administrative File Variables

Variable

Description

CVSEDITOR

EDITOR

VISUAL

The editor CVS uses for log file editing.

CVSROOT

The repository locator in use.

USER

The name of the user (on the server, if using a remote repository) running CVS.

=var

The value of a user-defined variable named var. Values for these variables are provided by the global -s option.

In order to edit these files, check out the CVSROOT module from the repository, edit the files, and commit them back to the repository. You must commit the changes for them to affect CVS's behavior.

Table 14-4 describes the administrative files and their functions.

Table 14-4. CVSROOT Files

File

Description

checkoutlist

Extra files to be maintained in CVSROOT

commitinfo

Specifications for commit governors

config

Settings to affect the behavior of CVS

cvsignore

Filename patterns of files to ignore

cvswrappers

Specifications for checkout and commit filters

editinfo

Specifications for log editors (obsolete)

history

Log information for the history command

loginfo

Specify commit notifier program(s)

modules

Module definitions

notify

Notification processing specifications

passwd

A list of users and their CVS-specific passwords

rcsinfo

Template form for log messages

readers

A list of users having read-only access

taginfo

Tag processing specifications

users

Alternate user email addresses for use with notify

verifymsg

Specify log message evaluator program

writers

A list of users having read/write access

Since the editinfo file is obsolete, use the $EDITOR environment variable (or the -e option) to specify the editor and the verifymsg file to specify an evaluator.

Each line of the taginfo file contains a filename pattern and a command line to execute when files with matching names are tagged.

14.3.3.3. The commitinfo file

Whenever a commit is being processed, CVS consults this file to determine whether or not any precommit checking of the file is required. Each line of the file contains a directory name pattern, followed by the path of a program to invoke when files are commited in directories with matching names.

Aside from the usual filename-pattern syntax, there are two special patterns:

ALL

If this pattern is present in the file, then all files are passed to the specified checking program. CVS then looks for a pattern that matches the name of each particular file and runs the additional checks found, if any.

DEFAULT

If this pattern is present in the file, all files for which there was no pattern match are sent to the specified checking program. The automatic match of every file to the ALL entry, if any, does not count as a match when determining whether or not to send the file to the DEFAULT checking program.

CVS constructs the command line for the checking program by appending the full path to the directory within the repository and the list of files being committed (this means you can specify the first few command-line arguments to the program, if necessary). If the checking program exits with a nonzero status, the commit is aborted.

The programs that run via this mechanism run on the server computer when a remote repository is used. Here is an example of a commitinfo file:

ALL $CVSROOT/CVSROOT/commit-ALL.pl
DEFAULT $CVSROOT/CVSROOT/commit-DEFAULT.pl
CVSROOT$ $CVSROOT/CVSROOT/commit-CVSROOT.pl

This example assumes you will create the script files in the CVSROOT module and add them to the checkoutlist file.

14.3.3.4. The config file

Repository configuration is specified in the config administrative file.

LockDir=dir

Directs CVS to put its lock files in the alternate directory given instead of in the repository itself, allowing users without write access to the repository (but with write access to dir) to read from the repository.

Version 1.10 doesn't support alternate directories for lock files and reports an error if this option is set. Older versions of CVS (1.9 and older) don't support this option either and will not report an error. Do not mix versions that support alternate directories for lock files with versions that don't, since lock files in both places defeat the purpose of having them.

RCSBIN=dir

Obsolete (used in versions 1.9.12 to 1.9.18). This option used to tell CVS where to find RCS programs. Since all RCS-related functions are now handled internally, this option does nothing.

SystemAuth=value

CVS tries to authenticate users via the CVSROOT/passwd file first, and if that fails and this option is set to yes, CVS tries to authenticate via the system's user database. This option is used with the password server. The default is yes.

TopLevelAdmin=value

If this option is set to yes, an additional CVS directory is created at the top-level directory when checkout is run. This allows the client software to detect the repository locator in that directory (see Section 14.4.1, "Repository Locators"). The default is no.

This option is useful if you check out multiple modules to the same sandbox directory. If it is enabled, you won't have to provide a repository locator after the first checkout; CVS infers it from the information in the top-level CVS directory created during the first checkout.

14.3.3.6. The cvswrappers file

While the cvsignore file allows CVS to ignore certain files, the cvswrappers file allows you to give CVS default options for commands that work with files. Lines in this file consist of a sh-style filename pattern followed by a -k (keyword substitution mode) option and/or an -m (update method) option. The legal values for -k are described in Table 14-19. The legal values for -m are COPY and MERGE.

If -m COPY is specified, CVS will not attempt to merge the files. Instead, it presents the user with conflicting versions of the file, and he can choose one or the other or resolve the conflict manually.

For example, to treat all files ending in .exe as binary, add this line to the file:

*.exe -k b

14.3.3.8. The loginfo file

The loginfo administrative file works much like the commitinfo file and can use the special patterns ALL and DEFAULT. This file allows you to do something with commit log messages and related information.

The programs called during loginfo processing receive the log message on standard input. Table 14-6 shows the three codes that can pass additional information to the called programs via command-line arguments.

Table 14-6. Special loginfo Variables

Variable

Description

s

Filename

V

Pre-commit revision number

v

Post-commit revision number

If a percent sign (%) followed by the desired variable is placed after the command path, CVS inserts the corresponding information as a whitespace-separated list with one entry for each file, preceded by the repository path (as with commitinfo). There can be only one percent sign on the command line, so if you want information from more than one variable, place the variable names inside curly brackets: %{...}. In this case, each file-specific entry has one field for each variable, separated by commas. For example, the code %{sVv} expands into a list like this:

/usr/local/cvsrep/hello Makefile,1.1,1.2 hello.c,1.8,1.9

It can be helpful to send email notifications each time someone commits a file to the repository. Developers can monitor this stream of notices to determine when they should pull the latest development code into their private sandboxes. For example, consider a developer doing some preparatory work in his sandbox while he awaits stabilization and addition of another developer's new library. As soon as the new library is added and committed, email notification goes out, and the waiting developer sees the code is ready to use. So, he runs cvs upd -d in the appropriate directory to pull in the new library code and then sets about integrating it with his work.

It is simple to set up this kind of notification. Just add a line like this to the CVSROOT/loginfo file:

DEFAULT mail -s %s developers@company.com

Often, the email address is a mailing list, which has all the interested parties (developers or otherwise) on the distribution list. If you want to send messages to multiple email addresses, you can write a script to do that and have that script called via this file. Alternatively, you can use the log.pl program that comes as part of the CVS source distribution (located at /usr/local/src/cvs-1.10.8/contrib/log.pl, assuming CVS was unpacked into /usr/local/src). Instructions for its use are provided as comments in the file.

14.3.3.9. The modules file

The top-level directories in a repository are called modules. In addition to these physical modules, CVS provides a mechanism to create logical modules through the modules administrative file. Here are the three kinds of logical modules:

Alias

Alias modules are defined by lines of the form:

module_name -a alias_module ...

You can use the alias module name in CVS commands in the same way you use the modules named after the -a option.

Regular

Regular modules are defined by lines of the form:

module_name [options] directory file ...

Checking out module_name results in the specified files from directory being checked out into a directory named module_name. The intervening directories (if any) are not reflected in the sandbox.

Ampersand

Ampersand modules are defined by lines of the form:

module_name [options] &other_module ...

Checking out such a module results in a directory named module_name, which in turn contains copies of the other_module modules.

Table 14-7 shows the options that can define modules.

Table 14-7. Module Options

Option

Description

-d name

Overrides the default working directory name for the module

-e prog

Runs the program prog when files are exported from the module; the module name is passed to prog as the sole argument

-i prog

Runs the program prog when files are committed to the module; the repository directory of the committed files is passed in to prog as the sole argument

-i prog

Runs the program prog when files are checked out from the module; the module name is passed in to prog as the sole argument

-s status

Assigns a status descriptor to the module

-t prog

Runs the program prog when files are tagged in the module using rtag; the module name and the symbolic tag are passed to prog

-u prog

Runs the program prog when files are updated in the module's top-level directory; the full path to the module within the repository is passed to prog as the sole argument

Alias modules provide alternative names for other modules or shortcuts for referring to collections or subdirectories of other modules. Alias module definitions function like macro definitions in that they cause commands to run as if the expanded list of modules and directories was on the command line. Alias modules do not cause the modules of their definition to be grouped together under the alias name (use ampersand modules for that). For example, the definition:

h -a hello

makes the name h a synonym for the hello module. This definition:

project -a library client server

allows you to check out all three modules of the project as a unit. If an entry in the definition of an alias module is preceded by an exclamation point (!), then the named directory is excluded from the module.

Regular modules allow you to create modules that are subsets of other modules. For example, the definition:

header library library.h

creates a module that just contains the header file from the library module.

Ampersand modules are true logical modules. There are no top-level directories for them in the repository, but you can check them out to sandboxes, and directories with their names will then appear. The modules listed in the definition are below that directory. For example:

project &library &client &server

is almost the same as the alias module example given earlier, except that the submodules are checked out inside a subdirectory named project.

In this file, long definitions may be split across multiple lines by terminating all but the last line with backslashes (\).

14.3.3.11. The passwd file

If you access the repository via a pserver repository locator (see Section 14.4.1, "Repository Locators"), then CVS can have its own private authentication information, separate from the system's user database. This information is stored in the CVSROOT/passwd administrative file.

This feature provides anonymous CVS access over the Internet. By creating an entry for a public user (usually anoncvs or anonymous), the pserver can be used by many people sharing the public account. If you don't want to create a system user with the same name as the public user, or if you have such a user but it has a different purpose, you can employ a user alias to map it to something else:

anonymous:TY7QWpLw8bvus:cvsnoname

Then, make sure you create the cvsnoname user on the system. You can use /bin/false as the login shell and the repository's root directory as the home directory for the user.

To restrict the public user to read-only access, list it in the CVSROOT/readers administrative file.

Additionally, CVS's private user database is useful even if you don't want to set up anonymous CVS access. You can restrict access to a subset of the system's users, provide remote access to users who don't have general system access, or prevent a user's normal system password from being transmitted in the clear over the network (see Section 14.3.2, "Security Issues").

There is no cvs passwd command for setting CVS-specific passwords (located in the repository file CVSROOT/passwd). CVS-specific user and password management are manual tasks.

14.3.4. Hacking the Repository

Since the repository is a normal directory, albeit one with special contents, it is possible to cd into the directory and examine its contents and/or make changes to the files and directories there. For each file that has been added there will be a file with the same name followed by ,v in a corresponding directory in the repository. These are RCS (the format, not the program) files that contain multiple versions of the file.

NOTE

Since the activities discussed in this section involve making changes directly to the repository instead of working through CVS commands, you should exercise extreme caution and have current backups when following these instructions.

14.3.4.1. Restructuring a project

Restructuring the project by moving files and directories around (and possibly renaming them) in the repository will allow the files to retain their history. The standard way to rename a file when using CVS is to rename the file in the sandbox and do a cvs remove on the old name and a cvs add on the new name. This results in the file being disconnected from its history under the new name, so sometimes it is better to do the renaming directly in the repository, although doing this while people have active sandboxes is dangerous, since the sandboxes will contain information about a file that is no longer in the repository.

14.3.5. Importing

If you have an existing code base, you'll want to import it into CVS in a way that preserves the most historical information. This section provides instructions for importing projects into CVS from code snapshots or other version control systems. All of these, except the code snapshot import procedure, are based upon conversion to RCS files, followed by placing the RCS files in the proper location in the CVS repository.

14.3.5.1. Importing code snapshots

If you have maintained project history archives manually by taking periodic snapshots of the code, you can import the first snapshot, tag it with the date or version number, and then successively overlay the updated files from later archives. Each set can then be committed and tagged in order to bootstrap a repository that maintains the prior history.

For example, first unpack the distributions (this assumes they unpack to directories containing the version numbers):

user@localhost$ tar xvzf foo-1.0.tar.gz
user@localhost$ tar xvzf foo-1.1.tar.gz
user@localhost$ tar xvzf foo-2.0.tar.gz

Next, make a copy of the first version, import it into the CVS repository, check it out to make a sandbox (since importing doesn't convert the source directory into a sandbox), and use cvs tag to give it a symbolic name reflecting the project version:

user@localhost$ mkdir foo
user@localhost$ cp -R -p foo-1.0/* foo
user@localhost$ cd foo
user@localhost$ cvs import -m 'Imported version 1.0' foo vendor start
user@localhost$ cd ..
user@localhost$ mv foo foo.bak
user@localhost$ cvs checkout foo
user@localhost$ cd foo
user@localhost$ cvs tag foo-1_0
user@localhost$ cd ..

Now, apply the differences between version 1.0 and 1.1 to the sandbox, commit the changes, and create a tag:

user@localhost$ diff -Naur foo-1.0 foo-1.1 | (cd foo; patch -Np1)
user@localhost$ cd foo
user@localhost$ cvs commit -m 'Imported version 1.1'
user@localhost$ cvs tag foo-1_1
user@localhost$ cd ..

Now, apply the differences between version 1.1 and 2.0 to the sandbox, commit the changes, and create a tag:

user@localhost$ diff -Naur foo-1.1 foo-2.0 | (cd foo; patch -Np1)
user@localhost$ cd foo
user@localhost$ cvs commit -m 'Imported version 2.0'
user@localhost$ cvs tag foo-2_0

Now, you can use the log command to view the history of the files, browse past versions of the files, and continue development under version control.

14.3.5.2. Importing from RCS

If you are migrating from RCS to CVS, following these instructions will result in a usable CVS repository. This procedure involves direct modification of the CVS repository, so it should be undertaken with caution.

Before beginning, make sure none of the files to be imported into CVS are locked by RCS. Then, create a new CVS repository and module (or a new module within an existing repository). Next, create directories in the CVS repository to mirror the project's directory structure. Finally, copy all the version files (,v) from the project (which may be in RCS subdirectories) into the appropriate directories in the repository (without RCS subdirectories).

For example, first move aside the directory under RCS control, create an empty directory to build the new CVS structure, import the directory, and then check it out to make a sandbox:

user@localhost$ mv foo foo-rcs
user@localhost$ mkdir foo
user@localhost$ cd foo
user@localhost$ cvs import -m 'New empty project' foo vendor start
user@localhost$ cd ..
user@localhost$ mv foo foo.bak
user@localhost$ cvs checkout foo

Next, make directories and add them to the repository to match the structure in the RCS project:

user@localhost$ cd foo
user@localhost$ mkdir dir
user@localhost$ cvs add dir
user@localhost$ cd ..

Now, copy the ,v files from the RCS project into the repository for the CVS project:

user@localhost$ cp -p foo-rcs/*,v $CVSROOT/foo
user@localhost$ cp -p foo-rcs/dir/*,v $CVSROOT/foo/dir

Finally, issue the cvs update command in the sandbox directory to bring in the latest versions of all the files:

user@localhost$ cd foo
user@localhost$ cvs upd

14.3.6. Using an Interim Shared Sandbox

Sometimes projects will develop unintended environmental dependencies over time, especially when there is no pressure for the code to be relocatable. A project developed outside version control may even be initially developed in place (at its intended installation location). While these practices are not recommended, they do occur in real-world situations; CVS can be helpful in improving the situation, by encouraging relocatability from the beginning of a project.

The default mode of operation for CVS is multiple independent sandboxes, all coordinated with a central shared repository. Code that runs in this environment is necessarily (at least partially) relocatable. So, using CVS from the beginning of a project helps ensure flexibility.

However, if a project is already well underway, an interim approach can be used. For example, you could convert the development area to a single shared sandbox by importing the code into CVS and checking it back out again:

user@localhost$ cd /usr/local/bar
user@localhost$ cvs import bar vendor start
user@localhost$ cd ..
user@localhost$ mv bar bar.bak
user@localhost$ cvs checkout bar

Chances are good that this approach is too aggressive and will check in more files than absolutely necessary. You can either go back and hack the repository to remove the files that shouldn't be there or just issue the cvs remove command to delete them as you discover them.

In addition, there will probably be some binary files in the sandbox that were imported as text files. Wherever you see a binary file that needs to remain in the repository, you should issue the command cvs admin -kb file, then make a fresh copy from the project backup. Finally, issue the command cvs commit file to commit the fixed file back to the repository.

Having version control in place before making flexibility enhancements is a good idea, since it makes it easier to find (and possibly reverse) changes that cause trouble.

The repository locator (see Section 14.4.1, "Repository Locators") is specified via the -d option or the $CVSROOT environment variable. It is stored in the various sandbox CVS/root files. If you are using the password server (pserver), the user ID of the person checking out the sandbox will be remembered. If more than one person is working with a particular sandbox, they will have to share an account for CVS access.

One way to do this is to have a neutral user account, with a password known by everyone with CVS access. Everyone can then issue the cvs login command with the same user ID and password and have access to the repository. Once you are no longer using a shared sandbox, this workaround won't be necessary. However, during the time you are using a shared sandbox, it is important that the developers type their real user IDs into their log messages, since all the changes will appear to be made by the common user.

14.3.8. Administrator Commands

Table 14-8 lists the commands that CVS administrators can use to manage their repositories.

Table 14-8. Administrator Commands

Command

Description

admin

adm

rcs

Perform administrative functions

init

Create a new repository

server

Run in server mode

14.3.8.1. admin

admin
  [ -b[rev] ]
  [ -cstring ]
  [ -kkflag ]
  [ -l[rev] ]
  [ -L ]
  [ -mrev:msg ]
  [ -nname[:[rev]] ]
  [ -Nname[:[rev]] ]
  [ -orange ]
  [ -q ]
  [ -sstate[:rev]
  [ -t[file] ]
  [ -t-string ]
  [ -u[rev] ]
  [ -U ]
  [ files ... ]

The admin is used to perform administrative functions. If a cvsadmin user group exists, then only those users in that group will be able to run admin with options other than -k. Additional options that may be used with the admin command are listed in Table 14-9.

Table 14-9. admin Options

Option

Description

-b[rev]

Set the default branch.

-cstring

Obsolete. Set the comment leader.

-kkflag

Set the default keyword substitution mode.

-l[rev]

Lock the specified revision.

-L

Enable strict locking.

-mrev:msg

Change the revision's log message.

-nname[:[rev]]

Give the branch or revision specified the symbolic name name.

-Nname[:[rev]]

The same as -n, except that if name is already in use, it is moved.

-orange

Delete revisions permanently.

-q

Don't print diagnostics.

-sstate[:rev]

Change the state of a revision.

-t[file]

Set the descriptive text in the RCS file.

-t-string

Set the descriptive text in the RCS file to string.

-u[rev]

Unlock the specified revision.

-U

Disable strict locking.

If the revision specified for -l is a branch, the latest revision on that branch will be used. If no revision is given, the latest revision on the default branch is used.

If the name given for -n is already in use, an error is generated. You can use -N to move a tag (change the revision associated with the tag); however, you should usually use cvs tag or cvs rtag instead.

The -o option is very dangerous and results in a permanent loss of information from the repository. Use it with extreme caution and only after careful consideration. See Table 14-10 for the various ways to specify ranges. There must not be any branches or locks on the revisions to be removed. Beware of interactions between this command and symbolic names.

If no file is specified to the -t option, CVS reads from standard input until it reaches the end of the file or a period on a line by itself.

The determination of the target revision for the -u option is the same as for -l.

Table 14-10. Range Formats

Format

Description

rev1::rev2

Eliminate versions between rev1 and rev2, retaining only enough information to go directly from rev1 to rev2. The two specified versions are retained.

::rev

The same as rev1::rev2, except the first revision is the branchpoint revision.

rev::

The same as rev1::rev2, except the second revision is the end of the branch, and it is deleted instead of retained.

rev

Delete the specified revision.

rev1: rev2

The same as rev1::rev2, except the two named revisions are deleted as well.

:rev

The same as ::rev2, except the named revision is deleted as well.

rev:

The same as rev1::, except the named revision is deleted as well.

The options in Table 14-11 are present in CVS for historical reasons and should not be used (using these options may corrupt the repository).

Table 14-11. Obsolete admin Options

Option

Description

-alogins

Append the logins to the RCS file's access list.

-Aoldfile

Append the access list of oldfile to the access list of the RCS file.

-e[logins]

Erases logins from the RCS file's access list, or erases all if a list is not provided.

-i

Create and initialize a new RCS file. Don't use this option. Instead, use add to add files to a CVS repository.

-I

Run interactively. This option doesn't work with client/server CVS and is likely to be removed in a future version.

-Vn

Obsolete. This option was used to specify that the RCS files used by CVS should be made compatible with a specific version of RCS.

-xsuffixes

This option used to be described as determining the filename suffix for RCS files, but CVS has always only used ,v as the RCS file suffix.



Library Navigation Links

Copyright © 2001 O'Reilly & Associates. All rights reserved.







??????????????@Mail.ru