CVS maintains a history of a source tree, in terms of a series of changes. It stamps each change with the time it was made and the user name of the person who made it. Usually, the person provides a bit of text describing why they made the change as well. Given that information, CVS can help developers answer questions like:
Before discussing too many vague terms and concepts, let's look over the essential CVS commands.
CVS records everyone's changes to a given project in a directory tree
called a repository. Before you can use CVS, you need to set the
CVSROOT environment variable to the repository's path. Whoever
is in charge of your project's configuration management will know what
this is; perhaps they've made a global definition for CVSROOT
somewhere.
In any case, on our system, the CVS repository is `/u/src/master'. In that case, you would need to enter the commands
setenv CVSROOT /u/src/masterif your shell is
csh or one of its descendents, or
CVSROOT=/u/src/master export CVSROOTif your shell is Bash or some other Bourne shell variant.
If you forget to do this, CVS will complain when you try to use it:
$ cvs checkout httpc cvs checkout: No CVSROOT specified! Please use the `-d' option cvs [checkout aborted]: or set the CVSROOT environment variable. $
CVS doesn't work on ordinary directory trees; you need to work within a
directory that CVS created for you. Just as you check out a book from a
library before taking it home to read it, you use the cvs
checkout command to get a directory tree from CVS before working on it.
For example, suppose you are working on a project named httpc, a
trivial HTTP client:
$ cd $ cvs checkout httpc cvs checkout: Updating httpc U httpc/.cvsignore U httpc/Makefile U httpc/httpc.c U httpc/poll-server $
The command cvs checkout httpc means, "Check out the source tree
called httpc from the repository specified by the CVSROOT
environment variable."
CVS puts the tree in a subdirectory named `httpc':
$ cd httpc $ ls -l total 8 drwxr-xr-x 2 jimb 512 Oct 31 11:04 CVS -rw-r--r-- 1 jimb 89 Oct 31 10:42 Makefile -rw-r--r-- 1 jimb 4432 Oct 31 10:45 httpc.c -rwxr-xr-x 1 jimb 460 Oct 30 10:21 poll-server
Most of these files are your working copies of the httpc sources.
However, the subdirectory called `CVS' (at the top) is different.
CVS uses it to record extra information about each of the files in that
directory, to help it determine what changes you've made since you
checked it out.
Once CVS has created a working directory tree, you can edit, compile and test the files it contains in the usual way -- they're just files.
For example, suppose we try compiling the package we just checked out:
$ make gcc -g -Wall -lnsl -lsocket httpc.c -o httpc httpc.c: In function `tcp_connection': httpc.c:48: warning: passing arg 2 of `connect' from incompatible pointer type $
It seems that `httpc.c' hasn't been ported to this operating system
yet. We need to cast one of the arguments to connect. To fix
that, line 48 must change from this:
if (connect (sock, &name, sizeof (name)) >= 0)
to this:
if (connect (sock, (struct sockaddr *) &name, sizeof (name)) >= 0)
Now it should compile:
$ make gcc -g -Wall -lnsl -lsocket httpc.c -o httpc $ httpc GET http://www.cyclic.com ... HTML text for Cyclic Software's home page follows ... $
Since each developer uses their own working directory, the changes you
make to your working directory aren't automatically visible to the other
developers on your team. CVS doesn't publish your changes until you're
ready. When you're done testing your changes, you must commit
them to the repository to make them available to the rest of the group.
We'll describe the cvs commit command below.
However, what if another developer has changed the same files you have, or the same lines? Whose changes should prevail? It's generally impossible to answer this question automatically; CVS certainly isn't competent to make that judgment.
Thus, before you can commit your changes, CVS requires your sources to
be in sync with any changes committed by the other team members. The
cvs update command takes care of this:
$ cvs update cvs update: Updating . U Makefile RCS file: /u/src/master/httpc/httpc.c,v retrieving revision 1.6 retrieving revision 1.7 Merging differences between 1.6 and 1.7 into httpc.c M httpc.c $
Let's look at this line-by-line:
U Makefile
RCS file: ...
retrieving revision 1.6
retrieving revision 1.7
Merging differences between 1.6 and 1.7 into httpc.c
1.7 are revision
numbers, used to identify a specific point in a file's history.
Note that CVS merges changes into your working copy only; the repository and the other developers' working directories are left undisturbed. It is up to you to test the merged text, and make sure it's valid.
M httpc.c
In this case, `httpc.c' now contains both your modifications and those of the other user.
Since CVS has merged someone else's changes into your source, it's best to make sure things still work:
$ make gcc -g -Wall -Wmissing-prototypes -lnsl -lsocket httpc.c -o httpc $ httpc GET http://www.cyclic.com ... HTML text for Cyclic Software's home page follows ... $
It seems to still work.
Now that you have brought your sources up to date with the rest of the
group and tested them, you are ready to commit your changes to the
repository and make them visible to the rest of the group. The only
file you've modified is `httpc.c', but it's always safe to run
cvs update to get a list of the modified files from CVS:
$ cvs update cvs update: Updating . M httpc.c $As expected, the only file CVS mentions is `httpc.c'; it says it contains changes which you have not yet committed. You can commit them like so:
$ cvs commit httpc.cAt this point, CVS will start up your favorite editor and prompt you for a log message describing the change. When you exit the editor, CVS will commit your change:
Checking in httpc.c; /u/src/master/httpc/httpc.c,v <-- httpc.c new revision: 1.8; previous revision: 1.7 $
Now that you have committed your changes, they are visible to the rest
of the group. When another developer runs cvs update, CVS will
merge your changes to `httpc.c' into their working directory.
At this point, you might well be curious what changes the other
developer made to `httpc.c'. To look at the log entries for a
given file, you can use the cvs log command:
$ cvs log httpc.c RCS file: /u/src/master/httpc/httpc.c,v Working file: httpc.c head: 1.8 branch: locks: strict access list: symbolic names: keyword substitution: kv total revisions: 8; selected revisions: 8 description: The one and only source file for the trivial HTTP client ---------------------------- revision 1.8 date: 1996/10/31 20:11:14; author: jimb; state: Exp; lines: +1 -1 (tcp_connection): Cast address structure when calling connect. ---------------------------- revision 1.7 date: 1996/10/31 19:18:45; author: fred; state: Exp; lines: +6 -2 (match_header): Make this test case-insensitive. ---------------------------- revision 1.6 date: 1996/10/31 19:15:23; author: jimb; state: Exp; lines: +2 -6 ... $
Most of the text here you can ignore; the portion to look at carefully is the series of log entries after the first line of hyphens. The log entries appear in reverse chronological order, under the assumption that more recent changes are usually more interesting. Each entry describes one change to the file, and may be parsed as follows:
The cvs log command can select log entries by date range, or by
revision number; see the manual for details on this.
If you would actually like to see the change in question, you can use
the cvs diff command. For example, if you would like to see the
changes Fred committed as revision 1.7, you can use the following
command:
$ cvs diff -c -r 1.6 -r 1.7 httpc.c
Before we look at the output from this command, let's look at what the various parts mean:
-c
cvs diff use a more human-readable format for
its output. (I'm not sure why it isn't the default.)
-r 1.6 -r 1.7
-r 1.6 -r 1.8 would display both fred's
changes and your most recent change. (You can even request that changes
be displayed backwards -- as if they were being undone -- by
specifying the revisions backwards: -r 1.7 -r 1.6. This
sounds odd, but it is useful sometimes.)
httpc.c
Here is the output from the command:
Index: httpc.c
===================================================================
RCS file: /u/src/master/httpc/httpc.c,v
retrieving revision 1.6
retrieving revision 1.7
diff -c -r1.6 -r1.7
*** httpc.c 1996/10/31 19:15:23 1.6
--- httpc.c 1996/10/31 19:18:45 1.7
***************
*** 62,68 ****
}
! /* Return non-zero iff HEADER is a prefix of TEXT. HEADER should be
null-terminated; LEN is the length of TEXT. */
static int
match_header (char *header, char *text, size_t len)
--- 62,69 ----
}
! /* Return non-zero iff HEADER is a prefix of TEXT, ignoring
! differences in case. HEADER should be lower-case, and
null-terminated; LEN is the length of TEXT. */
static int
match_header (char *header, char *text, size_t len)
***************
*** 76,81 ****
--- 77,84 ----
for (i = 0; i < header_len; i++)
{
char t = text[i];
+ if ('A' <= t && t <= 'Z')
+ t += 'a' - 'A';
if (header[i] != t)
return 0;
}
$
This output takes a bit of effort to get used to, but it is definitely worth understanding.
The interesting portion starts with the first two lines beginning with
*** and ---; those describe the older and newer files
compared. The remainder consists of two hunks, each of which
starts with a line of asterisks. Here is the first hunk:
***************
*** 62,68 ****
}
! /* Return non-zero iff HEADER is a prefix of TEXT. HEADER should be
null-terminated; LEN is the length of TEXT. */
static int
match_header (char *header, char *text, size_t len)
--- 62,69 ----
}
! /* Return non-zero iff HEADER is a prefix of TEXT, ignoring
! differences in case. HEADER should be lower-case, and
null-terminated; LEN is the length of TEXT. */
static int
match_header (char *header, char *text, size_t len)
Text from the older version appears after the *** 62,68 *** line;
text from the new appears after the --- 62,69 --- line. The pair
of numbers in each indicates the range of lines shown. CVS provides
context around the change, and marks the actual lines affected with
`!' characters. Thus, one can see that the single line in the top
half was replaced with the two lines in the bottom half.
Here is the second hunk:
***************
*** 76,81 ****
--- 77,84 ----
for (i = 0; i < header_len; i++)
{
char t = text[i];
+ if ('A' <= t && t <= 'Z')
+ t += 'a' - 'A';
if (header[i] != t)
return 0;
}
This hunk describes the insertion of two lines, marked with `+' characters. CVS omits the old text in this case, because it would be redundant. CVS uses a similar hunk format to describe deletions.
Like the Unix diff command, output from cvs diff is
usually called a patch, because developers have traditionally used
the format to distribute bug fixes or small new features. While
reasonably readable to humans, a patch contains enough information for
a program to apply the changes it describes to an unmodified text file.
In fact, the Unix patch command does exactly this, given a patch
as input.
CVS treats file creation and deletion like other changes, recording such events in the files' histories. One way to look at this is to say that CVS records the history of directories as well as the files they contain.
CVS doesn't assume that newly created files should be placed under its
control; this would do the wrong thing in many circumstances. For
example, one needn't record changes to object files and executables,
since their contents can always be recreated from the source files (one
hopes). Instead, if you create a new file, cvs update will flag
it with a `?' character until you tell CVS what you intend to do
with it.
To add a file to a project, you must first create the file, and then use
the cvs add command to mark it for addition. Then, the next call
to cvs commit will add the file to the repository. For example,
here's how you might add a README file to the httpc project:
$ ls CVS Makefile httpc.c poll-server $ vi README ... enter a description of httpc ... $ ls CVS Makefile README httpc.c poll-server $ cvs update cvs update: Updating . ? README --- CVS doesn't know about this file yet. $ cvs add README cvs add: scheduling file `README' for addition cvs add: use 'cvs commit' to add this file permanently $ cvs update --- Now what does CVS think? cvs update: Updating . A README --- The file is marked for addition. $ cvs commit README ... CVS prompts you for a log entry ... RCS file: /u/jimb/cvs-class/rep/httpc/README,v done Checking in README; /u/src/master/httpc/README,v <-- README initial revision: 1.1 done $
CVS treats deleted files similarly. If you delete a file and then run
cvs update, CVS doesn't assume you intend the file for deletion.
Instead, it does something benign -- it recreates the file with its
last recorded contents, and flags it with a `U' character, as for
any other update. (This means that if you want to undo the changes
you've made to a file in your working directory, you can simply delete
the files, and then let cvs update recreate them.)
To remove a file from a project, you must first delete the file, and
then use the cvs rm command to mark it for deletion. Then, the
next call to cvs commit will delete the file from the repository.
Committing a file marked with cvs rm does not destroy the file's
history. It simply adds a new revision, which is marked as
"non-existent." The repository still has records of the file's prior
contents, and can recall them as needed -- for example, by cvs
diff or cvs log.
There are several strategies for renaming files; the simplest is to
simply rename the file in your working directory, and run cvs rm
on the old name, and cvs add on the new name. The disadvantage
of this approach is that the log entries for the old file's content do
not carry over to the new file. Other strategies avoid this quirk,
but have other, stranger problems.
You can add directories just as you would ordinary files;
cvs diff to retrieve the actual text of a change,
why should one bother writing a log entry? Obviously, log entries can
be shorter than a patch, and allow the reader to get a general
understanding of the change without delving into its details.
However, a good log entry describes the reason the developer made
the change. For example, a bad log entry for revision 1.7 shown above
might say, "Convert t to lower-case." This would be accurate, but
completely useless; cvs diff provides all the same information,
more clearly. A better log entry would be, "Make this test
case-insensitive," because it makes the purpose clear to anyone with a
general understanding of the code: HTTP clients should ignore case
differences when parsing reply headers.
As mentioned above, the cvs update command incorporates changes
made by other developers into your working directory. If both you and
another developer have modified the same file, CVS merges their changes
with yours.
It's straightforward to imagine how this works when the changes apply to distant regions of a file, but what happens when you and another developer have changed the same line? CVS calls this situation a conflict, and leaves it up to you to resolve it.
For example, suppose that you have just added some error checking to the
host name lookup code. Before you commit your change, you must run
cvs update, to bring your sources into sync:
$ cvs update cvs update: Updating . RCS file: /u/src/master/httpc/httpc.c,v retrieving revision 1.8 retrieving revision 1.9 Merging differences between 1.8 and 1.9 into httpc.c rcsmerge: warning: conflicts during merge cvs update: conflicts found in httpc.c C httpc.c $
In this case, another developer has changed the same region of the file you have, so CVS complains about a conflict. Instead of printing `M httpc.c', as it usually does, it prints `C httpc.c', to indicate that a conflict has occurred in that file.
To resolve the conflict, bring up the file in your editor. CVS marks the conflicting text this way:
/* Look up the IP address of the host. */
host_info = gethostbyname (hostname);
<<<<<<< httpc.c
if (! host_info)
{
fprintf (stderr, "%s: host not found: %s\n", progname, hostname);
exit (1);
}
=======
if (! host_info)
{
printf ("httpc: no host");
exit (1);
}
>>>>>>> 1.9
sock = socket (PF_INET, SOCK_STREAM, 0);
The text from your working file appears at the top, after the `<<<'
characters; below it is the conflicting text from the other developer.
The revision number `1.9' indicates that the conflicting change was
introduced in version 1.9 of the file, making it easier for you to
check the logs, or examine the entire change with cvs diff.
Once you've decided how the conflict should be resolved, remove the markers from the code, and put it in its proper state. In this case, since your error handling code is clearly superior, you could simply throw out the other developer's change, leaving the text like this:
/* Look up the IP address of the host. */
host_info = gethostbyname (hostname);
if (! host_info)
{
fprintf (stderr, "%s: host not found: %s\n", progname, hostname);
exit (1);
}
sock = socket (PF_INET, SOCK_STREAM, 0);
Once this is done, you can test your change and commit the file:
$ make gcc -g -Wall -Wmissing-prototypes -lnsl -lsocket httpc.c -o httpc $ httpc GET http://www.cyclic.com HTTP/1.0 200 Document follows Date: Thu, 31 Oct 1996 23:04:06 GMT ... $ httpc GET http://www.frobnitz.com httpc: host not found: www.frobnitz.com $ cvs commit httpc.c
It's important to understand what CVS does and doesn't consider a conflict. CVS does not understand the semantics of your program; it simply treats its source code as a tree of text files. If one developer adds a new argument to a function and fixes its callers, while another developer simultaneously adds a new call to that function, and does not pass the new argument, that is certainly a conflict -- the two changes are incompatible -- but CVS will not report it. CVS's understanding of conflicts is strictly textual.
In practice, fortunately, conflicts are rare. Usually, they seem to result from two developers attempting to address the same problem, a lack of communication between developers, or disagreement about the design of the program. Allocating tasks to developers in a reasonable way reduces the likelihood of conflicts.
Many version control systems allow a developer to lock a file, preventing others from making changes to it until she has committed her changes. While locking is appropriate in some situations, it is not clearly a better solution than the approach CVS takes. Changes can usually be merged correctly, and developers occasionally forget to release locks; in both cases, explicit locking causes unnecessary delays. Furthermore, locks prevent only textual conflicts; they do not prevent semantic conflicts of the sort described above, if the two developers make their changes to different files.