23. File Handling

The operating system stores data permanently in named files, so most of the text you edit with Emacs comes from a file and is ultimately stored in a file.

To edit a file, you must tell Emacs to read the file and prepare a buffer containing a copy of the file's text. This is called visiting the file. Editing commands apply directly to text in the buffer; that is, to the copy inside Emacs. Your changes appear in the file itself only when you save the buffer back into the file.

In addition to visiting and saving files, Emacs can delete, copy, rename, and append to files, keep multiple versions of them, and operate on file directories.

23.1 File Names

Most Emacs commands that operate on a file require you to specify the file name. (Saving and reverting are exceptions; the buffer knows which file name to use for them.) You enter the file name using the minibuffer (see section The Minibuffer). Completion is available (see section Completion) to make it easier to specify long file names. When completing file names, Emacs ignores those whose file-name extensions appear in the variable completion-ignored-extensions; see Completion Options.

For most operations, there is a default file name which is used if you type just RET to enter an empty argument. Normally the default file name is the name of the file visited in the current buffer; this makes it easy to operate on that file with any of the Emacs file commands.

Each buffer has a default directory which is normally the same as the directory of the file visited in that buffer. When you enter a file name without a directory, the default directory is used. If you specify a directory in a relative fashion, with a name that does not start with a slash, it is interpreted with respect to the default directory. The default directory is kept in the variable default-directory, which has a separate value in every buffer.

The command M-x pwd displays the current buffer's default directory, and the command M-x cd sets it (to a value read using the minibuffer). A buffer's default directory changes only when the cd command is used. A file-visiting buffer's default directory is initialized to the directory of the file it visits. If you create a buffer with C-x b, its default directory is copied from that of the buffer that was current at the time.

For example, if the default file name is `/u/rms/gnu/gnu.tasks' then the default directory is normally `/u/rms/gnu/'. If you type just `foo', which does not specify a directory, it is short for `/u/rms/gnu/foo'. `../.login' would stand for `/u/rms/.login'. `new/foo' would stand for the file name `/u/rms/gnu/new/foo'.

The default directory actually appears in the minibuffer when the minibuffer becomes active to read a file name. This serves two purposes: it shows you what the default is, so that you can type a relative file name and know with certainty what it will mean, and it allows you to edit the default to specify a different directory. This insertion of the default directory is inhibited if the variable insert-default-directory is set to nil.

Note that it is legitimate to type an absolute file name after you enter the minibuffer, ignoring the presence of the default directory name as part of the text. The final minibuffer contents may look invalid, but that is not so. For example, if the minibuffer starts out with `/usr/tmp/' and you add `/x1/rms/foo', you get `/usr/tmp//x1/rms/foo'; but Emacs ignores everything through the first slash in the double slash; the result is `/x1/rms/foo'. See section Minibuffers for File Names.

You can use `~/' in a file name to mean your home directory, or `~user-id/' to mean the home directory of a user whose login name is user-id(2).

`$' in a file name is used to substitute an environment variable. The environment variable name consists of all the alphanumeric characters after the `$'; alternatively, it can be enclosed in braces after the `$'. For example, if you have used the shell command export FOO=rms/hacks to set up an environment variable named FOO, then you can use `/u/$FOO/test.c' or `/u/${FOO}/test.c' as an abbreviation for `/u/rms/hacks/test.c'. If the environment variable is not defined, no substitution occurs: `/u/$notdefined' stands for itself (assuming the environment variable notdefined is not defined).

Note that shell commands to set environment variables affect Emacs only when done before Emacs is started.

To access a file with `$' in its name, if the `$' causes expansion, type `$$'. This pair is converted to a single `$' at the same time as variable substitution is performed for a single `$'. Alternatively, quote the whole file name with `/:' (see section Quoted File Names). File names which begin with a literal `~' should also be quoted with `/:'.

The Lisp function that performs the `$'-substitution is called substitute-in-file-name. The substitution is performed only on file names read as such using the minibuffer.

You can include non-ASCII characters in file names if you set the variable file-name-coding-system to a non-nil value. See section Coding Systems for File Names.

23.2 Visiting Files

C-x C-f

Visit a file (find-file).

C-x C-r

Visit a file for viewing, without allowing changes to it (find-file-read-only).

C-x C-v

Visit a different file instead of the one visited last (find-alternate-file).

C-x 4 f

Visit a file, in another window (find-file-other-window). Don't alter what is displayed in the selected window.

C-x 5 f

Visit a file, in a new frame (find-file-other-frame). Don't alter what is displayed in the selected frame.

M-x find-file-literally

Visit a file with no conversion of the contents.

Visiting a file means reading its contents into an Emacs buffer so you can edit them. Emacs makes a new buffer for each file that you visit. We often say that this buffer "is visiting" that file, or that the buffer's "visited file" is that file. Emacs constructs the buffer name from the file name by throwing away the directory, keeping just the name proper. For example, a file named `/usr/rms/emacs.tex' would get a buffer named `emacs.tex'. If there is already a buffer with that name, Emacs constructs a unique name--the normal method is to append `<2>', `<3>', and so on, but you can select other methods (see section Making Buffer Names Unique).

Each window's mode line shows the name of the buffer that is being displayed in that window, so you can always tell what buffer you are editing.

The changes you make with editing commands are made in the Emacs buffer. They do not take effect in the file that you visited, or any permanent place, until you save the buffer. Saving the buffer means that Emacs writes the current contents of the buffer into its visited file. See section Saving Files.

If a buffer contains changes that have not been saved, we say the buffer is modified. This is important because it implies that some changes will be lost if the buffer is not saved. The mode line displays two stars near the left margin to indicate that the buffer is modified.

To visit a file, use the command C-x C-f (find-file). Follow the command with the name of the file you wish to visit, terminated by a RET.

The file name is read using the minibuffer (see section The Minibuffer), with defaulting and completion in the standard manner (see section File Names). While in the minibuffer, you can abort C-x C-f by typing C-g. File-name completion ignores certain file names; for more about this, see Completion Options.

Your confirmation that C-x C-f has completed successfully is the appearance of new text on the screen and a new buffer name in the mode line. If the specified file does not exist and you could not create it, or exists but you can't read it, then you get an error, with an error message displayed in the echo area.

If you visit a file that is already in Emacs, C-x C-f does not make another copy. It selects the existing buffer containing that file. However, before doing so, it checks whether the file itself has changed since you visited or saved it last. If the file has changed, Emacs offers to reread it.

If you try to visit a file larger than large-file-warning-threshold (the default is 10000000, which is about 10 megabytes), Emacs will ask you for confirmation first. You can answer y to proceed with visiting the file. Note, however, that Emacs cannot visit files that are larger than the maximum Emacs buffer size, which is around 256 megabytes on 32-bit machines (see section Using Multiple Buffers). If you try, Emacs will display an error message saying that the maximum buffer size has been exceeded.

On graphical displays there are two additional methods for visiting files. Firstly, when Emacs is built with a suitable GUI toolkit, commands invoked with the mouse (by clicking on the menu bar or tool bar) use the toolkit's standard File Selection dialog instead of prompting for the file name in the minibuffer. On Unix and GNU/Linux platforms, Emacs does that when built with GTK, LessTif, and Motif toolkits; on MS-Windows and Mac, the GUI version does that by default. For information on how to customize this, see Using Dialog Boxes.

Secondly, Emacs supports "drag and drop"; dropping a file into an ordinary Emacs window visits the file using that window. However, dropping a file into a window displaying a Dired buffer moves or copies the file into the displayed directory. For details, see Drag and Drop, and Other Dired Features.

What if you want to create a new file? Just visit it. Emacs displays `(New file)' in the echo area, but in other respects behaves as if you had visited an existing empty file. If you make any changes and save them, the file is created.

Emacs recognizes from the contents of a file which end-of-line convention it uses to separate lines--newline (used on GNU/Linux and on Unix), carriage-return linefeed (used on Microsoft systems), or just carriage-return (used on the Macintosh)--and automatically converts the contents to the normal Emacs convention, which is that the newline character separates lines. This is a part of the general feature of coding system conversion (see section Coding Systems), and makes it possible to edit files imported from different operating systems with equal convenience. If you change the text and save the file, Emacs performs the inverse conversion, changing newlines back into carriage-return linefeed or just carriage-return if appropriate.

If the file you specify is actually a directory, C-x C-f invokes Dired, the Emacs directory browser, so that you can "edit" the contents of the directory (see section Dired, the Directory Editor). Dired is a convenient way to view, delete, or operate on the files in the directory. However, if the variable find-file-run-dired is nil, then it is an error to try to visit a directory.

Files which are actually collections of other files, or file archives, are visited in special modes which invoke a Dired-like environment to allow operations on archive members. See section File Archives, for more about these features.

If the file name you specify contains shell-style wildcard characters, Emacs visits all the files that match it. (On case-insensitive filesystems, Emacs matches the wildcards disregarding the letter case.) Wildcards include `?', `*', and `[…]' sequences. To enter the wild card `?' in a file name in the minibuffer, you need to type C-q ?. See section Quoted File Names, for information on how to visit a file whose name actually contains wildcard characters. You can disable the wildcard feature by customizing find-file-wildcards.

If you visit a file that the operating system won't let you modify, or that is marked read-only, Emacs makes the buffer read-only too, so that you won't go ahead and make changes that you'll have trouble saving afterward. You can make the buffer writable with C-x C-q (toggle-read-only). See section Miscellaneous Buffer Operations.

If you want to visit a file as read-only in order to protect yourself from entering changes accidentally, visit it with the command C-x C-r (find-file-read-only) instead of C-x C-f.

If you visit a nonexistent file unintentionally (because you typed the wrong file name), use the C-x C-v command (find-alternate-file) to visit the file you really wanted. C-x C-v is similar to C-x C-f, but it kills the current buffer (after first offering to save it if it is modified). When C-x C-v reads the file name to visit, it inserts the entire default file name in the buffer, with point just after the directory part; this is convenient if you made a slight error in typing the name.

C-x 4 f (find-file-other-window) is like C-x C-f except that the buffer containing the specified file is selected in another window. The window that was selected before C-x 4 f continues to show the same buffer it was already showing. If this command is used when only one window is being displayed, that window is split in two, with one window showing the same buffer as before, and the other one showing the newly requested file. See section Multiple Windows.

C-x 5 f (find-file-other-frame) is similar, but opens a new frame, or makes visible any existing frame showing the file you seek. This feature is available only when you are using a window system. See section Frames and Graphical Displays.

If you wish to edit a file as a sequence of ASCII characters with no special encoding or conversion, use the M-x find-file-literally command. It visits a file, like C-x C-f, but does not do format conversion (see section Editing Formatted Text), character code conversion (see section Coding Systems), or automatic uncompression (see section Accessing Compressed Files), and does not add a final newline because of require-final-newline. If you already have visited the same file in the usual (non-literal) manner, this command asks you whether to visit it literally instead.

Two special hook variables allow extensions to modify the operation of visiting files. Visiting a file that does not exist runs the functions in the list find-file-not-found-functions; this variable holds a list of functions, and the functions are called one by one (with no arguments) until one of them returns non-nil. This is not a normal hook, and the name ends in `-functions' rather than `-hook' to indicate that fact.

Successful visiting of any file, whether existing or not, calls the functions in the list find-file-hook, with no arguments. This variable is a normal hook. In the case of a nonexistent file, the find-file-not-found-functions are run first. See section Hooks.

There are several ways to specify automatically the major mode for editing the file (see section How Major Modes are Chosen), and to specify local variables defined for that file (see section Local Variables in Files).

23.3 Saving Files

Saving a buffer in Emacs means writing its contents back into the file that was visited in the buffer.

23.3.1 Commands for Saving Files

These are the commands that relate to saving and writing files.

C-x C-s

Save the current buffer in its visited file on disk (save-buffer).

C-x s

Save any or all buffers in their visited files (save-some-buffers).

M-~

Forget that the current buffer has been changed (not-modified). With prefix argument (C-u), mark the current buffer as changed.

C-x C-w

Save the current buffer with a specified file name (write-file).

M-x set-visited-file-name

Change the file name under which the current buffer will be saved.

When you wish to save the file and make your changes permanent, type C-x C-s (save-buffer). After saving is finished, C-x C-s displays a message like this:

Wrote /u/rms/gnu/gnu.tasks

If the selected buffer is not modified (no changes have been made in it since the buffer was created or last saved), saving is not really done, because it would have no effect. Instead, C-x C-s displays a message like this in the echo area:

(No changes need to be saved)

The command C-x s (save-some-buffers) offers to save any or all modified buffers. It asks you what to do with each buffer. The possible responses are analogous to those of query-replace:

y

Save this buffer and ask about the rest of the buffers.

n

Don't save this buffer, but ask about the rest of the buffers.

!

Save this buffer and all the rest with no more questions.

RET

Terminate save-some-buffers without any more saving.

.

Save this buffer, then exit save-some-buffers without even asking about other buffers.

C-r

View the buffer that you are currently being asked about. When you exit View mode, you get back to save-some-buffers, which asks the question again.

d

Diff the buffer against its corresponding file, so you can see what changes you would be saving.

C-h

Display a help message about these options.

C-x C-c, the key sequence to exit Emacs, invokes save-some-buffers and therefore asks the same questions.

If you have changed a buffer but you do not want to save the changes, you should take some action to prevent it. Otherwise, each time you use C-x s or C-x C-c, you are liable to save this buffer by mistake. One thing you can do is type M-~ (not-modified), which clears out the indication that the buffer is modified. If you do this, none of the save commands will believe that the buffer needs to be saved. (`~' is often used as a mathematical symbol for `not'; thus M-~ is `not', metafied.) You could also use set-visited-file-name (see below) to mark the buffer as visiting a different file name, one which is not in use for anything important. Alternatively, you can cancel all the changes made since the file was visited or saved, by reading the text from the file again. This is called reverting. See section Reverting a Buffer. (You could also undo all the changes by repeating the undo command C-x u until you have undone all the changes; but reverting is easier.) You can also kill the buffer.

M-x set-visited-file-name alters the name of the file that the current buffer is visiting. It reads the new file name using the minibuffer. Then it marks the buffer as visiting that file name, and changes the buffer name correspondingly. set-visited-file-name does not save the buffer in the newly visited file; it just alters the records inside Emacs in case you do save later. It also marks the buffer as "modified" so that C-x C-s in that buffer will save.

If you wish to mark the buffer as visiting a different file and save it right away, use C-x C-w (write-file). It is equivalent to set-visited-file-name followed by C-x C-s (except that C-x C-w asks for confirmation if the file exists). C-x C-s used on a buffer that is not visiting a file has the same effect as C-x C-w; that is, it reads a file name, marks the buffer as visiting that file, and saves it there. The default file name in a buffer that is not visiting a file is made by combining the buffer name with the buffer's default directory (see section File Names).

If the new file name implies a major mode, then C-x C-w switches to that major mode, in most cases. The command set-visited-file-name also does this. See section How Major Modes are Chosen.

If Emacs is about to save a file and sees that the date of the latest version on disk does not match what Emacs last read or wrote, Emacs notifies you of this fact, because it probably indicates a problem caused by simultaneous editing and requires your immediate attention. See section Simultaneous Editing.

23.3.2 Backup Files

On most operating systems, rewriting a file automatically destroys all record of what the file used to contain. Thus, saving a file from Emacs throws away the old contents of the file--or it would, except that Emacs carefully copies the old contents to another file, called the backup file, before actually saving.

For most files, the variable make-backup-files determines whether to make backup files. On most operating systems, its default value is t, so that Emacs does write backup files.

For files managed by a version control system (see section Version Control), the variable vc-make-backup-files determines whether to make backup files. By default it is nil, since backup files are redundant when you store all the previous versions in a version control system. See section General Options.

At your option, Emacs can keep either a single backup for each file, or make a series of numbered backup files for each file that you edit.

The default value of the backup-enable-predicate variable prevents backup files being written for files in the directories used for temporary files, specified by temporary-file-directory or small-temporary-file-directory.

Emacs makes a backup for a file only the first time the file is saved from one buffer. No matter how many times you save a file, its backup file continues to contain the contents from before the file was visited. Normally this means that the backup file contains the contents from before the current editing session; however, if you kill the buffer and then visit the file again, a new backup file will be made by the next save.

You can also explicitly request making another backup file from a buffer even though it has already been saved at least once. If you save the buffer with C-u C-x C-s, the version thus saved will be made into a backup file if you save the buffer again. C-u C-u C-x C-s saves the buffer, but first makes the previous file contents into a new backup file. C-u C-u C-u C-x C-s does both things: it makes a backup from the previous contents, and arranges to make another from the newly saved contents if you save again.

23.3.2.1 Numbered Backups

The choice of single backup file or multiple numbered backup files is controlled by the variable version-control. Its possible values are:

t

Make numbered backups.

nil

Make numbered backups for files that have numbered backups already. Otherwise, make single backups.

never

Never make numbered backups; always make single backups.

The usual way to set this variable is globally, through your `.emacs' file or the customization buffer. However, you can set version-control locally in an individual buffer to control the making of backups for that buffer's file. For example, Rmail mode locally sets version-control to never to make sure that there is only one backup for an Rmail file. See section Local Variables.

If you set the environment variable VERSION_CONTROL, to tell various GNU utilities what to do with backup files, Emacs also obeys the environment variable by setting the Lisp variable version-control accordingly at startup. If the environment variable's value is `t' or `numbered', then version-control becomes t; if the value is `nil' or `existing', then version-control becomes nil; if it is `never' or `simple', then version-control becomes never.

23.3.2.2 Single or Numbered Backups

When Emacs makes a single backup file, its name is normally constructed by appending `~' to the file name being edited; thus, the backup file for `eval.c' would be `eval.c~'.

You can change this behavior by defining the variable make-backup-file-name-function to a suitable function. Alternatively you can customize the variable backup-directory-alist to specify that files matching certain patterns should be backed up in specific directories.

A typical use is to add an element ("." . dir) to make all backups in the directory with absolute name dir; Emacs modifies the backup file names to avoid clashes between files with the same names originating in different directories. Alternatively, adding, say, ("." . ".~") would make backups in the invisible subdirectory `.~' of the original file's directory. Emacs creates the directory, if necessary, to make the backup.

If access control stops Emacs from writing backup files under the usual names, it writes the backup file as `%backup%~' in your home directory. Only one such file can exist, so only the most recently made such backup is available.

If you choose to have a series of numbered backup files, backup file names contain `.~', the number, and another `~' after the original file name. Thus, the backup files of `eval.c' would be called `eval.c.~1~', `eval.c.~2~', and so on, all the way through names like `eval.c.~259~' and beyond. The variable backup-directory-alist applies to numbered backups just as usual.

23.3.2.3 Automatic Deletion of Backups

To prevent excessive consumption of disk space, Emacs can delete numbered backup versions automatically. Generally Emacs keeps the first few backups and the latest few backups, deleting any in between. This happens every time a new backup is made.

The two variables kept-old-versions and kept-new-versions control this deletion. Their values are, respectively, the number of oldest (lowest-numbered) backups to keep and the number of newest (highest-numbered) ones to keep, each time a new backup is made. The backups in the middle (excluding those oldest and newest) are the excess middle versions--those backups are deleted. These variables' values are used when it is time to delete excess versions, just after a new backup version is made; the newly made backup is included in the count in kept-new-versions. By default, both variables are 2.

If delete-old-versions is t, Emacs deletes the excess backup files silently. If it is nil, the default, Emacs asks you whether it should delete the excess backup versions. If it has any other value, then Emacs never automatically deletes backups.

Dired's . (Period) command can also be used to delete old versions. See section Deleting Files with Dired.

23.3.2.4 Copying vs. Renaming

Backup files can be made by copying the old file or by renaming it. This makes a difference when the old file has multiple names (hard links). If the old file is renamed into the backup file, then the alternate names become names for the backup file. If the old file is copied instead, then the alternate names remain names for the file that you are editing, and the contents accessed by those names will be the new contents.

The method of making a backup file may also affect the file's owner and group. If copying is used, these do not change. If renaming is used, you become the file's owner, and the file's group becomes the default (different operating systems have different defaults for the group).

Having the owner change is usually a good idea, because then the owner always shows who last edited the file. Also, the owners of the backups show who produced those versions. Occasionally there is a file whose owner should not change; it is a good idea for such files to contain local variable lists to set backup-by-copying-when-mismatch locally (see section Local Variables in Files).

The choice of renaming or copying is controlled by four variables. Renaming is the default choice. If the variable backup-by-copying is non-nil, copying is used. Otherwise, if the variable backup-by-copying-when-linked is non-nil, then copying is used for files that have multiple names, but renaming may still be used when the file being edited has only one name. If the variable backup-by-copying-when-mismatch is non-nil, then copying is used if renaming would cause the file's owner or group to change. backup-by-copying-when-mismatch is t by default if you start Emacs as the superuser. The fourth variable, backup-by-copying-when-privileged-mismatch, gives the highest numeric user-id for which backup-by-copying-when-mismatch will be forced on. This is useful when low-numbered user-ids are assigned to special system users, such as root, bin, daemon, etc., which must maintain ownership of files.

When a file is managed with a version control system (see section Version Control), Emacs does not normally make backups in the usual way for that file. But check-in and check-out are similar in some ways to making backups. One unfortunate similarity is that these operations typically break hard links, disconnecting the file name you visited from any alternate names for the same file. This has nothing to do with Emacs--the version control system does it.

23.3.3 Customizing Saving of Files

If the value of the variable require-final-newline is t, saving or writing a file silently puts a newline at the end if there isn't already one there. If the value is visit, Emacs adds a newline at the end of any file that doesn't have one, just after it visits the file. (This marks the buffer as modified, and you can undo it.) If the value is visit-save, that means to add newlines both on visiting and on saving. If the value is nil, Emacs leaves the end of the file unchanged; if it's neither nil nor t, Emacs asks you whether to add a newline. The default is nil.

Many major modes are designed for specific kinds of files that are always supposed to end in newlines. These major modes set the variable require-final-newline according to mode-require-final-newline. By setting the latter variable, you can control how these modes handle final newlines.

When Emacs saves a file, it invokes the fsync system call to force the data immediately out to disk. This is important for safety if the system crashes or in case of power outage. However, it can be disruptive on laptops using power saving, because it requires the disk to spin up each time you save a file. Setting write-region-inhibit-fsync to a non-nil value disables this synchronization. Be careful--this means increased risk of data loss.

23.3.4 Protection against Simultaneous Editing

Simultaneous editing occurs when two users visit the same file, both make changes, and then both save them. If nobody were informed that this was happening, whichever user saved first would later find that his changes were lost.

On some systems, Emacs notices immediately when the second user starts to change the file, and issues an immediate warning. On all systems, Emacs checks when you save the file, and warns if you are about to overwrite another user's changes. You can prevent loss of the other user's work by taking the proper corrective action instead of saving the file.

When you make the first modification in an Emacs buffer that is visiting a file, Emacs records that the file is locked by you. (It does this by creating a symbolic link in the same directory with a different name.) Emacs removes the lock when you save the changes. The idea is that the file is locked whenever an Emacs buffer visiting it has unsaved changes.

If you begin to modify the buffer while the visited file is locked by someone else, this constitutes a collision. When Emacs detects a collision, it asks you what to do, by calling the Lisp function ask-user-about-lock. You can redefine this function for the sake of customization. The standard definition of this function asks you a question and accepts three possible answers:

s

Steal the lock. Whoever was already changing the file loses the lock, and you gain the lock.

p

Proceed. Go ahead and edit the file despite its being locked by someone else.

q

Quit. This causes an error (file-locked), and the buffer contents remain unchanged--the modification you were trying to make does not actually take place.

Note that locking works on the basis of a file name; if a file has multiple names, Emacs does not realize that the two names are the same file and cannot prevent two users from editing it simultaneously under different names. However, basing locking on names means that Emacs can interlock the editing of new files that will not really exist until they are saved.

Some systems are not configured to allow Emacs to make locks, and there are cases where lock files cannot be written. In these cases, Emacs cannot detect trouble in advance, but it still can detect the collision when you try to save a file and overwrite someone else's changes.

If Emacs or the operating system crashes, this may leave behind lock files which are stale, so you may occasionally get warnings about spurious collisions. When you determine that the collision is spurious, just use p to tell Emacs to go ahead anyway.

Every time Emacs saves a buffer, it first checks the last-modification date of the existing file on disk to verify that it has not changed since the file was last visited or saved. If the date does not match, it implies that changes were made in the file in some other way, and these changes are about to be lost if Emacs actually does save. To prevent this, Emacs displays a warning message and asks for confirmation before saving. Occasionally you will know why the file was changed and know that it does not matter; then you can answer yes and proceed. Otherwise, you should cancel the save with C-g and investigate the situation.

The first thing you should do when notified that simultaneous editing has already taken place is to list the directory with C-u C-x C-d (see section File Directories). This shows the file's current author. You should attempt to contact him to warn him not to continue editing. Often the next step is to save the contents of your Emacs buffer under a different name, and use diff to compare the two files.

23.3.5 Shadowing Files

M-x shadow-initialize

Set up file shadowing.

M-x shadow-define-literal-group

Declare a single file to be shared between sites.

M-x shadow-define-regexp-group

Make all files that match each of a group of files be shared between hosts.

M-x shadow-define-cluster RET name RET

Define a shadow file cluster name.

M-x shadow-copy-files

Copy all pending shadow files.

M-x shadow-cancel

Cancel the instruction to shadow some files.

You can arrange to keep identical shadow copies of certain files in more than one place--possibly on different machines. To do this, first you must set up a shadow file group, which is a set of identically-named files shared between a list of sites. The file group is permanent and applies to further Emacs sessions as well as the current one. Once the group is set up, every time you exit Emacs, it will copy the file you edited to the other files in its group. You can also do the copying without exiting Emacs, by typing M-x shadow-copy-files.

To set up a shadow file group, use M-x shadow-define-literal-group or M-x shadow-define-regexp-group. See their documentation strings for further information.

Before copying a file to its shadows, Emacs asks for confirmation. You can answer "no" to bypass copying of this file, this time. If you want to cancel the shadowing permanently for a certain file, use M-x shadow-cancel to eliminate or change the shadow file group.

A shadow cluster is a group of hosts that share directories, so that copying to or from one of them is sufficient to update the file on all of them. Each shadow cluster has a name, and specifies the network address of a primary host (the one we copy files to), and a regular expression that matches the host names of all the other hosts in the cluster. You can define a shadow cluster with M-x shadow-define-cluster.

23.3.6 Updating Time Stamps Automatically

You can arrange to put a time stamp in a file, so that it will be updated automatically each time you edit and save the file. The time stamp has to be in the first eight lines of the file, and you should insert it like this:

Time-stamp: <>

or like this:

Time-stamp: " "

Then add the hook function time-stamp to the hook before-save-hook; that hook function will automatically update the time stamp, inserting the current date and time when you save the file. You can also use the command M-x time-stamp to update the time stamp manually. For other customizations, see the Custom group time-stamp. Note that non-numeric fields in the time stamp are formatted according to your locale setting (see section Environment Variables).

23.4 Reverting a Buffer

If you have made extensive changes to a file and then change your mind about them, you can get rid of them by reading in the previous version of the file. To do this, use M-x revert-buffer, which operates on the current buffer. Since reverting a buffer unintentionally could lose a lot of work, you must confirm this command with yes.

revert-buffer tries to position point in such a way that, if the file was edited only slightly, you will be at approximately the same piece of text after reverting as before. However, if you have made drastic changes, point may wind up in a totally different piece of text.

Reverting marks the buffer as "not modified" until another change is made.

Some kinds of buffers whose contents reflect data bases other than files, such as Dired buffers, can also be reverted. For them, reverting means recalculating their contents from the appropriate data base. Buffers created explicitly with C-x b cannot be reverted; revert-buffer reports an error when asked to do so.

When you edit a file that changes automatically and frequently--for example, a log of output from a process that continues to run--it may be useful for Emacs to revert the file without querying you, whenever you visit the file again with C-x C-f.

To request this behavior, set the variable revert-without-query to a list of regular expressions. When a file name matches one of these regular expressions, find-file and revert-buffer will revert it automatically if it has changed--provided the buffer itself is not modified. (If you have edited the text, it would be wrong to discard your changes.)

You may find it useful to have Emacs revert files automatically when they change. Three minor modes are available to do this.

M-x global-auto-revert-mode enables Global Auto-Revert mode, which periodically checks all file buffers and reverts when the corresponding file has changed. M-x auto-revert-mode enables a local version, Auto-Revert mode, which applies only to the current buffer.

You can use Auto-Revert mode to "tail" a file such as a system log, so that changes made to that file by other programs are continuously displayed. To do this, just move the point to the end of the buffer, and it will stay there as the file contents change. However, if you are sure that the file will only change by growing at the end, use Auto-Revert Tail mode instead (auto-revert-tail-mode). It is more efficient for this.

The variable auto-revert-interval controls how often to check for a changed file. Since checking a remote file is too slow, these modes do not check or revert remote files.

See section Version Control and the Mode Line, for Auto Revert peculiarities in buffers that visit files under version control.

23.5 Auto Reverting non-file Buffers

Normally Global Auto Revert Mode only reverts file buffers. There are two ways to auto-revert certain non-file buffers: enabling Auto Revert Mode in those buffers (using M-x auto-revert-mode) and setting global-auto-revert-non-file-buffers to t. The latter enables Auto Reverting for all types of buffers for which it is implemented, that is, for the types of buffers listed in the menu below.

Like file buffers, non-file buffers should normally not revert while you are working on them, or while they contain information that might get lost after reverting. Therefore, they do not revert if they are "modified". This can get tricky, because deciding when a non-file buffer should be marked modified is usually more difficult than for file buffers.

Another tricky detail is that, for efficiency reasons, Auto Revert often does not try to detect all possible changes in the buffer, only changes that are "major" or easy to detect. Hence, enabling auto-reverting for a non-file buffer does not always guarantee that all information in the buffer is up to date and does not necessarily make manual reverts useless.

At the other extreme, certain buffers automatically auto-revert every auto-revert-interval seconds. (This currently only applies to the Buffer Menu.) In this case, Auto Revert does not print any messages while reverting, even when auto-revert-verbose is non-nil.

The details depend on the particular types of buffers and are explained in the corresponding sections.

23.5.1 Auto Reverting the Buffer Menu

If auto-reverting of non-file buffers is enabled, the Buffer Menu automatically reverts every auto-revert-interval seconds, whether there is a need for it or not. (It would probably take longer to check whether there is a need than to actually revert.)

If the Buffer Menu inappropriately gets marked modified, just revert it manually using g and auto-reverting will resume. However, if you marked certain buffers to get deleted or to be displayed, you have to be careful, because reverting erases all marks. The fact that adding marks sets the buffer's modified flag prevents Auto Revert from automatically erasing the marks.

23.5.2 Auto Reverting Dired buffers

Auto-reverting Dired buffers currently works on GNU or Unix style operating systems. It may not work satisfactorily on some other systems.

Dired buffers only auto-revert when the file list of the buffer's main directory changes. They do not auto-revert when information about a particular file changes or when inserted subdirectories change. To be sure that all listed information is up to date, you have to manually revert using g, even if auto-reverting is enabled in the Dired buffer. Sometimes, you might get the impression that modifying or saving files listed in the main directory actually does cause auto-reverting. This is because making changes to a file, or saving it, very often causes changes in the directory itself, for instance, through backup files or auto-save files. However, this is not guaranteed.

If the Dired buffer is marked modified and there are no changes you want to protect, then most of the time you can make auto-reverting resume by manually reverting the buffer using g. There is one exception. If you flag or mark files, you can safely revert the buffer. This will not erase the flags or marks (unless the marked file has been deleted, of course). However, the buffer will stay modified, even after reverting, and auto-reverting will not resume. This is because, if you flag or mark files, you may be working on the buffer and you might not want the buffer to change without warning. If you want auto-reverting to resume in the presence of marks and flags, mark the buffer non-modified using M-~. However, adding, deleting or changing marks or flags will mark it modified again.

Remote Dired buffers are not auto-reverted. Neither are Dired buffers for which you used shell wildcards or file arguments to list only some of the files. `*Find*' and `*Locate*' buffers do not auto-revert either.

23.5.3 Adding Support for Auto-Reverting additional Buffers.

This section is intended for Elisp programmers who would like to add support for auto-reverting new types of buffers.

To support auto-reverting the buffer must first of all have a revert-buffer-function. See (elisp)Definition of revert-buffer-function section `Reverting' in the Emacs Lisp Reference Manual.

In addition, it must have a buffer-stale-function.

Variable: buffer-stale-function

The value of this variable is a function to check whether a non-file buffer needs reverting. This should be a function with one optional argument noconfirm. The function should return non-nil if the buffer should be reverted. The buffer is current when this function is called.

While this function is mainly intended for use in auto-reverting, it could be used for other purposes as well. For instance, if auto-reverting is not enabled, it could be used to warn the user that the buffer needs reverting. The idea behind the noconfirm argument is that it should be t if the buffer is going to be reverted without asking the user and nil if the function is just going to be used to warn the user that the buffer is out of date. In particular, for use in auto-reverting, noconfirm is t. If the function is only going to be used for auto-reverting, you can ignore the noconfirm argument.

If you just want to automatically auto-revert every auto-revert-interval seconds, use:

(set (make-local-variable 'buffer-stale-function) #'(lambda (&optional noconfirm) 'fast))

in the buffer's mode function.

The special return value `fast' tells the caller that the need for reverting was not checked, but that reverting the buffer is fast. It also tells Auto Revert not to print any revert messages, even if auto-revert-verbose is non-nil. This is important, as getting revert messages every auto-revert-interval seconds can be very annoying. The information provided by this return value could also be useful if the function is consulted for purposes other than auto-reverting.

Once the buffer has a revert-buffer-function and a buffer-stale-function, several problems usually remain.

The buffer will only auto-revert if it is marked unmodified. Hence, you will have to make sure that various functions mark the buffer modified if and only if either the buffer contains information that might be lost by reverting or there is reason to believe that the user might be inconvenienced by auto-reverting, because he is actively working on the buffer. The user can always override this by manually adjusting the modified status of the buffer. To support this, calling the revert-buffer-function on a buffer that is marked unmodified should always keep the buffer marked unmodified.

It is important to assure that point does not continuously jump around as a consequence of auto-reverting. Of course, moving point might be inevitable if the buffer radically changes.

You should make sure that the revert-buffer-function does not print messages that unnecessarily duplicate Auto Revert's own messages if auto-revert-verbose is t and effectively override a nil value for auto-revert-verbose. Hence, adapting a mode for auto-reverting often involves getting rid of such messages. This is especially important for buffers that automatically auto-revert every auto-revert-interval seconds.

Also, you may want to update the documentation string of global-auto-revert-non-file-buffers.

Finally, you should add a section to this chapter. This section should at the very least make clear whether enabling auto-reverting for the buffer reliably assures that all information in the buffer is completely up to date (or will be after auto-revert-interval seconds).

23.6 Auto-Saving: Protection Against Disasters

Emacs saves all the visited files from time to time (based on counting your keystrokes) without being asked. This is called auto-saving. It prevents you from losing more than a limited amount of work if the system crashes.

When Emacs determines that it is time for auto-saving, it considers each buffer, and each is auto-saved if auto-saving is enabled for it and it has been changed since the last time it was auto-saved. The message `Auto-saving...' is displayed in the echo area during auto-saving, if any files are actually auto-saved. Errors occurring during auto-saving are caught so that they do not interfere with the execution of commands you have been typing.

23.6.1 Auto-Save Files

Auto-saving does not normally save in the files that you visited, because it can be very undesirable to save a program that is in an inconsistent state when you have made half of a planned change. Instead, auto-saving is done in a different file called the auto-save file, and the visited file is changed only when you request saving explicitly (such as with C-x C-s).

Normally, the auto-save file name is made by appending `#' to the front and rear of the visited file name. Thus, a buffer visiting file `foo.c' is auto-saved in a file `#foo.c#'. Most buffers that are not visiting files are auto-saved only if you request it explicitly; when they are auto-saved, the auto-save file name is made by appending `#' to the front and rear of buffer name, then adding digits and letters at the end for uniqueness. For example, the `*mail*' buffer in which you compose messages to be sent might be auto-saved in a file named `#*mail*#704juu'. Auto-save file names are made this way unless you reprogram parts of Emacs to do something different (the functions make-auto-save-file-name and auto-save-file-name-p). The file name to be used for auto-saving in a buffer is calculated when auto-saving is turned on in that buffer.

The variable auto-save-file-name-transforms allows a degree of control over the auto-save file name. It lets you specify a series of regular expressions and replacements to transform the auto save file name. The default value puts the auto-save files for remote files (see section Remote Files) into the temporary file directory on the local machine.

When you delete a substantial part of the text in a large buffer, auto save turns off temporarily in that buffer. This is because if you deleted the text unintentionally, you might find the auto-save file more useful if it contains the deleted text. To reenable auto-saving after this happens, save the buffer with C-x C-s, or use C-u 1 M-x auto-save-mode.

If you want auto-saving to be done in the visited file rather than in a separate auto-save file, set the variable auto-save-visited-file-name to a non-nil value. In this mode, there is no real difference between auto-saving and explicit saving.

A buffer's auto-save file is deleted when you save the buffer in its visited file. (You can inhibit this by setting the variable delete-auto-save-files to nil.) Changing the visited file name with C-x C-w or set-visited-file-name renames any auto-save file to go with the new visited name.

23.6.2 Controlling Auto-Saving

Each time you visit a file, auto-saving is turned on for that file's buffer if the variable auto-save-default is non-nil (but not in batch mode; see section Entering and Exiting Emacs). The default for this variable is t, so auto-saving is the usual practice for file-visiting buffers. Auto-saving can be turned on or off for any existing buffer with the command M-x auto-save-mode. Like other minor mode commands, M-x auto-save-mode turns auto-saving on with a positive argument, off with a zero or negative argument; with no argument, it toggles.

Emacs does auto-saving periodically based on counting how many characters you have typed since the last time auto-saving was done. The variable auto-save-interval specifies how many characters there are between auto-saves. By default, it is 300. Emacs doesn't accept values that are too small: if you customize auto-save-interval to a value less than 20, Emacs will behave as if the value is 20.

Auto-saving also takes place when you stop typing for a while. The variable auto-save-timeout says how many seconds Emacs should wait before it does an auto save (and perhaps also a garbage collection). (The actual time period is longer if the current buffer is long; this is a heuristic which aims to keep out of your way when you are editing long buffers, in which auto-save takes an appreciable amount of time.) Auto-saving during idle periods accomplishes two things: first, it makes sure all your work is saved if you go away from the terminal for a while; second, it may avoid some auto-saving while you are actually typing.

Emacs also does auto-saving whenever it gets a fatal error. This includes killing the Emacs job with a shell command such as `kill %emacs', or disconnecting a phone line or network connection.

You can request an auto-save explicitly with the command M-x do-auto-save.

23.6.3 Recovering Data from Auto-Saves

You can use the contents of an auto-save file to recover from a loss of data with the command M-x recover-file RET file RET. This visits file and then (after your confirmation) restores the contents from its auto-save file `#file#'. You can then save with C-x C-s to put the recovered text into file itself. For example, to recover file `foo.c' from its auto-save file `#foo.c#', do:

M-x recover-file RET foo.c RET
yes RET
C-x C-s

Before asking for confirmation, M-x recover-file displays a directory listing describing the specified file and the auto-save file, so you can compare their sizes and dates. If the auto-save file is older, M-x recover-file does not offer to read it.

If Emacs or the computer crashes, you can recover all the files you were editing from their auto save files with the command M-x recover-session. This first shows you a list of recorded interrupted sessions. Move point to the one you choose, and type C-c C-c.

Then recover-session asks about each of the files that were being edited during that session, asking whether to recover that file. If you answer y, it calls recover-file, which works in its normal fashion. It shows the dates of the original file and its auto-save file, and asks once again whether to recover that file.

When recover-session is done, the files you've chosen to recover are present in Emacs buffers. You should then save them. Only this--saving them--updates the files themselves.

Emacs records information about interrupted sessions for later recovery in files named `~/.emacs.d/auto-save-list/.saves-pid-hostname'. All of this name except the `pid-hostname' part comes from the value of auto-save-list-file-prefix. You can record sessions in a different place by customizing that variable. If you set auto-save-list-file-prefix to nil in your `.emacs' file, sessions are not recorded for recovery.

23.7 File Name Aliases

Symbolic links and hard links both make it possible for several file names to refer to the same file. Hard links are alternate names that refer directly to the file; all the names are equally valid, and no one of them is preferred. By contrast, a symbolic link is a kind of defined alias: when `foo' is a symbolic link to `bar', you can use either name to refer to the file, but `bar' is the real name, while `foo' is just an alias. More complex cases occur when symbolic links point to directories.

Normally, if you visit a file which Emacs is already visiting under a different name, Emacs displays a message in the echo area and uses the existing buffer visiting that file. This can happen on systems that support hard or symbolic links, or if you use a long file name on a system that truncates long file names, or on a case-insensitive file system. You can suppress the message by setting the variable find-file-suppress-same-file-warnings to a non-nil value. You can disable this feature entirely by setting the variable find-file-existing-other-name to nil: then if you visit the same file under two different names, you get a separate buffer for each file name.

If the variable find-file-visit-truename is non-nil, then the file name recorded for a buffer is the file's truename (made by replacing all symbolic links with their target names), rather than the name you specify. Setting find-file-visit-truename also implies the effect of find-file-existing-other-name.

23.8 Version Control

Version control systems are packages that can record multiple versions of a source file, usually storing the unchanged parts of the file just once. Version control systems also record history information such as the creation time of each version, who created it, and a description of what was changed in that version.

The Emacs version control interface is called VC. Its commands work with different version control systems--currently, it supports CVS, GNU Arch, RCS, Meta-CVS, Subversion, and SCCS. Of these, the GNU project distributes CVS, GNU Arch, and RCS; we recommend that you use either CVS or GNU Arch for your projects, and RCS for individual files. We also have free software to replace SCCS, known as CSSC; if you are using SCCS and don't want to make the incompatible change to RCS or CVS, you can switch to CSSC.

VC is enabled by default in Emacs. To disable it, set the customizable variable vc-handled-backends to nil (see section Customizing VC).

23.8.1 Introduction to Version Control

VC allows you to use a version control system from within Emacs, integrating the version control operations smoothly with editing. VC provides a uniform interface to version control, so that regardless of which version control system is in use, you can use it the same way.

This section provides a general overview of version control, and describes the version control systems that VC supports. You can skip this section if you are already familiar with the version control system you want to use.

23.8.1.1 Supported Version Control Systems

VC currently works with six different version control systems or "back ends": CVS, GNU Arch, RCS, Meta-CVS, Subversion, and SCCS.

CVS is a free version control system that is used for the majority of free software projects today. It allows concurrent multi-user development either locally or over the network. Some of its shortcomings, corrected by newer systems such as GNU Arch, are that it lacks atomic commits or support for renaming files. VC supports all basic editing operations under CVS, but for some less common tasks you still need to call CVS from the command line. Note also that before using CVS you must set up a repository, which is a subject too complex to treat here.

GNU Arch is a new version control system that is designed for distributed work. It differs in many ways from old well-known systems, such as CVS and RCS. It supports different transports for interoperating between users, offline operations, and it has good branching and merging features. It also supports atomic commits, and history of file renaming and moving. VC does not support all operations provided by GNU Arch, so you must sometimes invoke it from the command line, or use a specialized module.

RCS is the free version control system around which VC was initially built. The VC commands are therefore conceptually closest to RCS. Almost everything you can do with RCS can be done through VC. You cannot use RCS over the network though, and it only works at the level of individual files, rather than projects. You should use it if you want a simple, yet reliable tool for handling individual files.

Subversion is a free version control system designed to be similar to CVS but without CVS's problems. Subversion supports atomic commits, and versions directories, symbolic links, meta-data, renames, copies, and deletes. It can be used via http or via its own protocol.

Meta-CVS is another attempt to solve problems arising in CVS. It supports directory structure versioning, improved branching and merging, and use of symbolic links and meta-data in repositories.

SCCS is a proprietary but widely used version control system. In terms of capabilities, it is the weakest of the six that VC supports. VC compensates for certain features missing in SCCS (snapshots, for example) by implementing them itself, but some other VC features, such as multiple branches, are not available with SCCS. Since SCCS is non-free, not respecting its users freedom, you should not use it; use its free replacement CSSC instead. But you should use CSSC only if for some reason you cannot use RCS, or one of the higher-level systems such as CVS or GNU Arch.

In the following, we discuss mainly RCS, SCCS and CVS. Nearly everything said about CVS applies to GNU Arch, Subversion and Meta-CVS as well.

23.8.1.2 Concepts of Version Control

When a file is under version control, we also say that it is registered in the version control system. Each registered file has a corresponding master file which represents the file's present state plus its change history--enough to reconstruct the current version or any earlier version. Usually the master file also records a log entry for each version, describing in words what was changed in that version.

The file that is maintained under version control is sometimes called the work file corresponding to its master file. You edit the work file and make changes in it, as you would with an ordinary file. (With SCCS and RCS, you must lock the file before you start to edit it.) After you are done with a set of changes, you check the file in, which records the changes in the master file, along with a log entry for them.

With CVS, there are usually multiple work files corresponding to a single master file--often each user has his own copy. It is also possible to use RCS in this way, but this is not the usual way to use RCS.

A version control system typically has some mechanism to coordinate between users who want to change the same file. One method is locking (analogous to the locking that Emacs uses to detect simultaneous editing of a file, but distinct from it). The other method is to merge your changes with other people's changes when you check them in.

With version control locking, work files are normally read-only so that you cannot change them. You ask the version control system to make a work file writable for you by locking it; only one user can do this at any given time. When you check in your changes, that unlocks the file, making the work file read-only again. This allows other users to lock the file to make further changes. SCCS always uses locking, and RCS normally does.

The other alternative for RCS is to let each user modify the work file at any time. In this mode, locking is not required, but it is permitted; check-in is still the way to record a new version.

CVS normally allows each user to modify his own copy of the work file at any time, but requires merging with changes from other users at check-in time. However, CVS can also be set up to require locking. (see section Options specific for CVS).

23.8.1.3 Types of Log File

Projects that use a revision control system can have two types of log for changes. One is the per-file log maintained by the revision control system: each time you check in a change, you must fill out a log entry for the change (see section Features of the Log Entry Buffer). This kind of log is called the version control log, also the revision control log, RCS log, or CVS log.

The other kind of log is the file `ChangeLog' (see section Change Logs). It provides a chronological record of all changes to a large portion of a program--typically one directory and its subdirectories. A small program would use one `ChangeLog' file; a large program may well merit a `ChangeLog' file in each major directory. See section Change Logs.

A project maintained with version control can use just the per-file log, or it can use both kinds of logs. It can handle some files one way and some files the other way. Each project has its policy, which you should follow.

When the policy is to use both, you typically want to write an entry for each change just once, then put it into both logs. You can write the entry in `ChangeLog', then copy it to the log buffer when you check in the change. Or you can write the entry in the log buffer while checking in the change, and later use the C-x v a command to copy it to `ChangeLog' (see section Change Logs and VC).

23.8.2 Version Control and the Mode Line

When you visit a file that is under version control, Emacs indicates this on the mode line. For example, `RCS-1.3' says that RCS is used for that file, and the current version is 1.3.

The character between the back-end name and the version number indicates the version control status of the file. `-' means that the work file is not locked (if locking is in use), or not modified (if locking is not in use). `:' indicates that the file is locked, or that it is modified. If the file is locked by some other user (for instance, `jim'), that is displayed as `RCS:jim:1.3'.

When Auto Revert mode (see section Reverting a Buffer) reverts a buffer that is under version control, it updates the version control information in the mode line. However, Auto Revert mode may not properly update this information if the version control status changes without changes to the work file, from outside the current Emacs session. If you set auto-revert-check-vc-info to t, Auto Revert mode updates the version control status information every auto-revert-interval seconds, even if the work file itself is unchanged. The resulting CPU usage depends on the version control system, but is usually not excessive.

23.8.3 Basic Editing under Version Control

The principal VC command is an all-purpose command that performs either locking or check-in, depending on the situation.

C-x v v

Perform the next logical version control operation on this file.

The precise action of this command depends on the state of the file, and whether the version control system uses locking or not. SCCS and RCS normally use locking; CVS normally does not use locking.

As a special convenience that is particularly useful for files with locking, you can let Emacs check a file in or out whenever you change its read-only flag. This means, for example, that you cannot accidentally edit a file without properly checking it out first. To achieve this, bind the key C-x C-q to vc-toggle-read-only in your `~/.emacs' file. (See section Rebinding Keys in Your Init File.)

23.8.3.1 Basic Version Control with Locking

If locking is used for the file (as with SCCS, and RCS in its default mode), C-x v v can either lock a file or check it in:

• If the file is not locked, C-x v v locks it, and makes it writable so that you can change it.

• If the file is locked by you, and contains changes, C-x v v checks in the changes. In order to do this, it first reads the log entry for the new version. See section Features of the Log Entry Buffer.

• If the file is locked by you, but you have not changed it since you locked it, C-x v v releases the lock and makes the file read-only again.

• If the file is locked by some other user, C-x v v asks you whether you want to "steal the lock" from that user. If you say yes, the file becomes locked by you, but a message is sent to the person who had formerly locked the file, to inform him of what has happened.

These rules also apply when you use CVS in locking mode, except that there is no such thing as stealing a lock.

23.8.3.2 Basic Version Control without Locking

When there is no locking--the default for CVS--work files are always writable; you do not need to do anything before you begin to edit a file. The status indicator on the mode line is `-' if the file is unmodified; it flips to `:' as soon as you save any changes in the work file.

Here is what C-x v v does when using CVS:

• If some other user has checked in changes into the master file, Emacs asks you whether you want to merge those changes into your own work file. You must do this before you can check in your own changes. (To pick up any recent changes from the master file without trying to commit your own changes, type C-x v m RET.) See section Merging Branches.

• If there are no new changes in the master file, but you have made modifications in your work file, C-x v v checks in your changes. In order to do this, it first reads the log entry for the new version. See section Features of the Log Entry Buffer.

• If the file is not modified, the C-x v v does nothing.

These rules also apply when you use RCS in the mode that does not require locking, except that automatic merging of changes from the master file is not implemented. Unfortunately, this means that nothing informs you if another user has checked in changes in the same file since you began editing it, and when this happens, his changes will be effectively removed when you check in your version (though they will remain in the master file, so they will not be entirely lost). You must therefore verify that the current version is unchanged, before you check in your changes. We hope to eliminate this risk and provide automatic merging with RCS in a future Emacs version.

In addition, locking is possible with RCS even in this mode, although it is not required; C-x v v with an unmodified file locks the file, just as it does with RCS in its normal (locking) mode.

23.8.3.3 Advanced Control in C-x v v

When you give a prefix argument to vc-next-action (C-u C-x v v), it still performs the next logical version control operation, but accepts additional arguments to specify precisely how to do the operation.

• If the file is modified (or locked), you can specify the version number to use for the new version that you check in. This is one way to create a new branch (see section Multiple Branches of a File).

• If the file is not modified (and unlocked), you can specify the version to select; this lets you start working from an older version, or on another branch. If you do not enter any version, that takes you to the highest version on the current branch; therefore C-u C-x v v RET is a convenient way to get the latest version of a file from the repository.

• Instead of the version number, you can also specify the name of a version control system. This is useful when one file is being managed with two version control systems at the same time (see section Local Version Control).

23.8.3.4 Features of the Log Entry Buffer

When you check in changes, C-x v v first reads a log entry. It pops up a buffer called `*VC-Log*' for you to enter the log entry.

Sometimes the `*VC-Log*' buffer contains default text when you enter it, typically the last log message entered. If it does, mark and point are set around the entire contents of the buffer so that it is easy to kill the contents of the buffer with C-w.

If you work by writing entries in the `ChangeLog' (see section Change Logs) and then commit the change under revision control, you can generate the Log Edit text from the ChangeLog using C-c C-a (log-edit-insert-changelog). This looks for entries for the file(s) concerned in the top entry in the ChangeLog and uses those paragraphs as the log text. This text is only inserted if the top entry was made under your user name on the current date. See section Change Logs and VC, for the opposite way of working--generating ChangeLog entries from the revision control log.

In the `*VC-Log*' buffer, C-c C-f (M-x log-edit-show-files) shows the list of files to be committed in case you need to check that. (This can be a list of more than one file if you use VC Dired mode or PCL-CVS. See section Dired under VC, and (pcl-cvs)Top section `About PCL-CVS' in PCL-CVS -- The Emacs Front-End to CVS.)

When you have finished editing the log message, type C-c C-c to exit the buffer and commit the change.

To abort check-in, just don't type C-c C-c in that buffer. You can switch buffers and do other editing. As long as you don't try to check in another file, the entry you were editing remains in the `*VC-Log*' buffer, and you can go back to that buffer at any time to complete the check-in.

If you change several source files for the same reason, it is often convenient to specify the same log entry for many of the files. To do this, use the history of previous log entries. The commands M-n, M-p, M-s and M-r for doing this work just like the minibuffer history commands (except that these versions are used outside the minibuffer).

Each time you check in a file, the log entry buffer is put into VC Log mode, which involves running two hooks: text-mode-hook and vc-log-mode-hook. See section Hooks.

23.8.4 Examining And Comparing Old Versions

One of the convenient features of version control is the ability to examine any version of a file, or compare two versions.

C-x v ~ version RET

Examine version version of the visited file, in a buffer of its own.

C-x v =

Compare the current buffer contents with the master version from which you started editing.

C-u C-x v = file RET oldvers RET newvers RET

Compare the specified two versions of file.

C-x v g

Display the file with per-line version information and using colors.

To examine an old version in its entirety, visit the file and then type C-x v ~ version RET (vc-version-other-window). This puts the text of version version in a file named `filename.~version~', and visits it in its own buffer in a separate window. (In RCS, you can also select an old version and create a branch from it. See section Multiple Branches of a File.)

It is usually more convenient to compare two versions of the file, with the command C-x v = (vc-diff). Plain C-x v = compares the current buffer contents (saving them in the file if necessary) with the master version from which you started editing the file (this is not necessarily the latest version of the file). C-u C-x v =, with a numeric argument, reads a file name and two version numbers, then compares those versions of the specified file. Both forms display the output in a special buffer in another window.

You can specify a checked-in version by its number; an empty input specifies the current contents of the work file (which may be different from all the checked-in versions). You can also specify a snapshot name (see section Snapshots) instead of one or both version numbers.

If you supply a directory name instead of the name of a registered file, this command compares the two specified versions of all registered files in that directory and its subdirectories.

C-x v = works by running a variant of the diff utility designed to work with the version control system in use. When you invoke diff this way, in addition to the options specified by diff-switches (see section Comparing Files), it receives those specified by vc-diff-switches, plus those specified for the specific back end by vc-backend-diff-switches. For instance, when the version control back end is RCS, diff uses the options in vc-rcs-diff-switches. The `vc…diff-switches' variables are nil by default.

The buffer produced by C-x v = supports the commands of Compilation mode (see section Compilation Mode), such as C-x ` and C-c C-c, in both the "old" and "new" text, and they always find the corresponding locations in the current work file. (Older versions are not, in general, present as files on your disk.)

For some back ends, you can display the file annotated with per-line version information and using colors to enhance the visual appearance, with the command M-x vc-annotate. It creates a new buffer (the "annotate buffer") displaying the file's text, with each part colored to show how old it is. Text colored red is new, blue means old, and intermediate colors indicate intermediate ages. By default, the color is scaled over the full range of ages, such that the oldest changes are blue, and the newest changes are red.

When you give a prefix argument to this command, it uses the minibuffer to read two arguments: which version number to display and annotate (instead of the current file contents), and the time span in days the color range should cover.

From the annotate buffer, these and other color scaling options are available from the `VC-Annotate' menu. In this buffer, you can also use the following keys to browse the annotations of past revisions, view diffs, or view log entries:

P

Annotate the previous revision, that is to say, the revision before the one currently annotated. A numeric prefix argument is a repeat count, so C-u 10 P would take you back 10 revisions.

N

Annotate the next revision--the one after the revision currently annotated. A numeric prefix argument is a repeat count.

J

Annotate the revision indicated by the current line.

A

Annotate the revision before the one indicated by the current line. This is useful to see the state the file was in before the change on the current line was made.

D

Display the diff between the current line's revision and the previous revision. This is useful to see what the current line's revision actually changed in the file.

L

Show the log of the current line's revision. This is useful to see the author's description of the changes in the revision on the current line.

W

Annotate the workfile version-the one you are editing. If you used P and N to browse to other revisions, use this key to return to your current version.

23.8.5 The Secondary Commands of VC

This section explains the secondary commands of VC; those that you might use once a day.

23.8.5.1 Registering a File for Version Control

You can put any file under version control by simply visiting it, and then typing C-x v i (vc-register).

C-x v i

Register the visited file for version control.

To register the file, Emacs must choose which version control system to use for it. If the file's directory already contains files registered in a version control system, Emacs uses that system. If there is more than one system in use for a directory, Emacs uses the one that appears first in vc-handled-backends (see section Customizing VC). On the other hand, if there are no files already registered, Emacs uses the first system from vc-handled-backends that could register the file (for example, you cannot register a file under CVS if its directory is not already part of a CVS tree); with the default value of vc-handled-backends, this means that Emacs uses RCS in this situation.

If locking is in use, C-x v i leaves the file unlocked and read-only. Type C-x v v if you wish to start editing it. After registering a file with CVS, you must subsequently commit the initial version by typing C-x v v. Until you do that, the version appears as `@@' in the mode line.

The initial version number for a newly registered file is 1.1, by default. You can specify a different default by setting the variable vc-default-init-version, or you can give C-x v i a numeric argument; then it reads the initial version number for this particular file using the minibuffer.

If vc-initial-comment is non-nil, C-x v i reads an initial comment to describe the purpose of this source file. Reading the initial comment works like reading a log entry (see section Features of the Log Entry Buffer).

23.8.5.2 VC Status Commands

C-x v l

Display version control state and change history.

To view the detailed version control status and history of a file, type C-x v l (vc-print-log). It displays the history of changes to the current file, including the text of the log entries. The output appears in a separate window. The point is centered at the revision of the file that is currently being visited.

In the change log buffer, you can use the following keys to move between the logs of revisions and of files, to view past revisions, and to view diffs:

p

Move to the previous revision-item in the buffer. (Revision entries in the log buffer are usually in reverse-chronological order, so the previous revision-item usually corresponds to a newer revision.) A numeric prefix argument is a repeat count.

n

Move to the next revision-item (which most often corresponds to the previous revision of the file). A numeric prefix argument is a repeat count.

P

Move to the log of the previous file, when the logs of multiple files are in the log buffer (see section Dired under VC). Otherwise, just move to the beginning of the log. A numeric prefix argument is a repeat count, so C-u 10 P would move backward 10 files.

N

Move to the log of the next file, when the logs of multiple files are in the log buffer (see section Dired under VC). It also takes a numeric prefix argument as a repeat count.

f

Visit the revision indicated at the current line, like typing C-x v ~ and specifying this revision's number (see section Examining And Comparing Old Versions).

d

Display the diff (see section Comparing Files) between the revision indicated at the current line and the next earlier revision. This is useful to see what actually changed when the revision indicated on the current line was committed.

23.8.5.3 Undoing Version Control Actions

C-x v u

Revert the buffer and the file to the version from which you started editing the file.

C-x v c

Remove the last-entered change from the master for the visited file. This undoes your last check-in.

If you want to discard your current set of changes and revert to the version from which you started editing the file, use C-x v u (vc-revert-buffer). This leaves the file unlocked; if locking is in use, you must first lock the file again before you change it again. C-x v u requires confirmation, unless it sees that you haven't made any changes with respect to the master version.

C-x v u is also the command to unlock a file if you lock it and then decide not to change it.

To cancel a change that you already checked in, use C-x v c (vc-cancel-version). This command discards all record of the most recent checked-in version, but only if your work file corresponds to that version--you cannot use C-x v c to cancel a version that is not the latest on its branch. C-x v c also offers to revert your work file and buffer to the previous version (the one that precedes the version that is deleted).

If you answer no, VC keeps your changes in the buffer, and locks the file. The no-revert option is useful when you have checked in a change and then discover a trivial error in it; you can cancel the erroneous check-in, fix the error, and check the file in again.

When C-x v c does not revert the buffer, it unexpands all version control headers in the buffer instead (see section Inserting Version Control Headers). This is because the buffer no longer corresponds to any existing version. If you check it in again, the check-in process will expand the headers properly for the new version number.

However, it is impossible to unexpand the RCS `$Log$' header automatically. If you use that header feature, you have to unexpand it by hand--by deleting the entry for the version that you just canceled.

Be careful when invoking C-x v c, as it is easy to lose a lot of work with it. To help you be careful, this command always requires confirmation with yes. Note also that this command is disabled under CVS, because canceling versions is very dangerous and discouraged with CVS.

23.8.5.4 Dired under VC

The VC Dired Mode described here works with all the version control systems that VC supports. Another more powerful facility, designed specifically for CVS, is called PCL-CVS. See (pcl-cvs)Top section `About PCL-CVS' in PCL-CVS -- The Emacs Front-End to CVS.

When you are working on a large program, it is often useful to find out which files have changed within an entire directory tree, or to view the status of all files under version control at once, and to perform version control operations on collections of files. You can use the command C-x v d (vc-directory) to make a directory listing that includes only files relevant for version control.

C-x v d creates a buffer which uses VC Dired Mode. This looks much like an ordinary Dired buffer (see section Dired, the Directory Editor); however, normally it shows only the noteworthy files (those locked or not up-to-date). This is called terse display. If you set the variable vc-dired-terse-display to nil, then VC Dired shows all relevant files--those managed under version control, plus all subdirectories (full display). The command v t in a VC Dired buffer toggles between terse display and full display (see section VC Dired Commands).

By default, VC Dired produces a recursive listing of noteworthy or relevant files at or below the given directory. You can change this by setting the variable vc-dired-recurse to nil; then VC Dired shows only the files in the given directory.

The line for an individual file shows the version control state in the place of the hard link count, owner, group, and size of the file. If the file is unmodified, in sync with the master file, the version control state shown is blank. Otherwise it consists of text in parentheses. Under RCS and SCCS, the name of the user locking the file is shown; under CVS, an abbreviated version of the `cvs status' output is used. Here is an example using RCS:

  /home/jim/project:

  -rw-r--r-- (jim)      Apr  2 23:39 file1
  -r--r--r--            Apr  5 20:21 file2

The files `file1' and `file2' are under version control, `file1' is locked by user jim, and `file2' is unlocked.

Here is an example using CVS:

  /home/joe/develop:

  -rw-r--r-- (modified) Aug  2  1997 file1.c
  -rw-r--r--            Apr  4 20:09 file2.c
  -rw-r--r-- (merge)    Sep 13  1996 file3.c

Here `file1.c' is modified with respect to the repository, and `file2.c' is not. `file3.c' is modified, but other changes have also been checked in to the repository--you need to merge them with the work file before you can check it in.

In the above, if the repository were on a remote machine, VC would only contact it when the variable vc-stay-local (or vc-cvs-stay-local) is nil (see section Options specific for CVS). This is because access to the repository may be slow, or you may be working offline and not have access to the repository at all. As a consequence, VC would not be able to tell you that `file3.c' is in the "merge" state; you would learn that only when you try to check-in your modified copy of the file, or use a command such as C-x v m.

In practice, this is not a problem because CVS handles this case consistently whenever it arises. In VC, you'll simply get prompted to merge the remote changes into your work file first. The benefits of less network communication usually outweigh the disadvantage of not seeing remote changes immediately.

When VC Dired displays subdirectories (in the "full" display mode), it omits some that should never contain any files under version control. By default, this includes Version Control subdirectories such as `RCS' and `CVS'; you can customize this by setting the variable vc-directory-exclusion-list.

You can fine-tune VC Dired's format by typing C-u C-x v d--as in ordinary Dired, that allows you to specify additional switches for the `ls' command.

23.8.5.5 VC Dired Commands

All the usual Dired commands work normally in VC Dired mode, except for v, which is redefined as the version control prefix. You can invoke VC commands such as vc-diff and vc-print-log by typing v =, or v l, and so on. Most of these commands apply to the file name on the current line.

The command v v (vc-next-action) operates on all the marked files, so that you can lock or check in several files at once. If it operates on more than one file, it handles each file according to its current state; thus, it might lock one file, but check in another file. This could be confusing; it is up to you to avoid confusing behavior by marking a set of files that are in a similar state. If no files are marked, v v operates on the file in the current line.

If any files call for check-in, v v reads a single log entry, then uses it for all the files being checked in. This is convenient for registering or checking in several files at once, as part of the same change.

You can toggle between terse display (only locked files, or files not up-to-date) and full display at any time by typing v t (vc-dired-toggle-terse-mode). There is also a special command * l (vc-dired-mark-locked), which marks all files currently locked (or, with CVS, all files not up-to-date). Thus, typing * l t k is another way to delete from the buffer all files except those currently locked.

23.8.6 Multiple Branches of a File

One use of version control is to maintain multiple "current" versions of a file. For example, you might have different versions of a program in which you are gradually adding various unfinished new features. Each such independent line of development is called a branch. VC allows you to create branches, switch between different branches, and merge changes from one branch to another. Please note, however, that branches are not supported for SCCS.

A file's main line of development is usually called the trunk. The versions on the trunk are normally numbered 1.1, 1.2, 1.3, etc. At any such version, you can start an independent branch. A branch starting at version 1.2 would have version number 1.2.1.1, and consecutive versions on this branch would have numbers 1.2.1.2, 1.2.1.3, 1.2.1.4, and so on. If there is a second branch also starting at version 1.2, it would consist of versions 1.2.2.1, 1.2.2.2, 1.2.2.3, etc.

If you omit the final component of a version number, that is called a branch number. It refers to the highest existing version on that branch--the head version of that branch. The branches in the example above have branch numbers 1.2.1 and 1.2.2.

23.8.6.1 Switching between Branches

To switch between branches, type C-u C-x v v and specify the version number you want to select. This version is then visited unlocked (write-protected), so you can examine it before locking it. Switching branches in this way is allowed only when the file is not locked.

You can omit the minor version number, thus giving only the branch number; this takes you to the head version on the chosen branch. If you only type RET, Emacs goes to the highest version on the trunk.

After you have switched to any branch (including the main branch), you stay on it for subsequent VC commands, until you explicitly select some other branch.

23.8.6.2 Creating New Branches

To create a new branch from a head version (one that is the latest in the branch that contains it), first select that version if necessary, lock it with C-x v v, and make whatever changes you want. Then, when you check in the changes, use C-u C-x v v. This lets you specify the version number for the new version. You should specify a suitable branch number for a branch starting at the current version. For example, if the current version is 2.5, the branch number should be 2.5.1, 2.5.2, and so on, depending on the number of existing branches at that point.

To create a new branch at an older version (one that is no longer the head of a branch), first select that version (see section Switching between Branches), then lock it with C-x v v. You'll be asked to confirm, when you lock the old version, that you really mean to create a new branch--if you say no, you'll be offered a chance to lock the latest version instead.

Then make your changes and type C-x v v again to check in a new version. This automatically creates a new branch starting from the selected version. You need not specially request a new branch, because that's the only way to add a new version at a point that is not the head of a branch.

After the branch is created, you "stay" on it. That means that subsequent check-ins create new versions on that branch. To leave the branch, you must explicitly select a different version with C-u C-x v v. To transfer changes from one branch to another, use the merge command, described in the next section.

23.8.6.3 Merging Branches

When you have finished the changes on a certain branch, you will often want to incorporate them into the file's main line of development (the trunk). This is not a trivial operation, because development might also have proceeded on the trunk, so that you must merge the changes into a file that has already been changed otherwise. VC allows you to do this (and other things) with the vc-merge command.

C-x v m (vc-merge)

Merge changes into the work file.

C-x v m (vc-merge) takes a set of changes and merges it into the current version of the work file. It firsts asks you in the minibuffer where the changes should come from. If you just type RET, Emacs merges any changes that were made on the same branch since you checked the file out (we call this merging the news). This is the common way to pick up recent changes from the repository, regardless of whether you have already changed the file yourself.

You can also enter a branch number or a pair of version numbers in the minibuffer. Then C-x v m finds the changes from that branch, or the differences between the two versions you specified, and merges them into the current version of the current file.

As an example, suppose that you have finished a certain feature on branch 1.3.1. In the meantime, development on the trunk has proceeded to version 1.5. To merge the changes from the branch to the trunk, first go to the head version of the trunk, by typing C-u C-x v v RET. Version 1.5 is now current. If locking is used for the file, type C-x v v to lock version 1.5 so that you can change it. Next, type C-x v m 1.3.1 RET. This takes the entire set of changes on branch 1.3.1 (relative to version 1.3, where the branch started, up to the last version on the branch) and merges it into the current version of the work file. You can now check in the changed file, thus creating version 1.6 containing the changes from the branch.

It is possible to do further editing after merging the branch, before the next check-in. But it is usually wiser to check in the merged version, then lock it and make the further changes. This will keep a better record of the history of changes.

When you merge changes into a file that has itself been modified, the changes might overlap. We call this situation a conflict, and reconciling the conflicting changes is called resolving a conflict.

Whenever conflicts occur during merging, VC detects them, tells you about them in the echo area, and asks whether you want help in merging. If you say yes, it starts an Ediff session (see Ediff: (ediff)Top section `Ediff' in The Ediff Manual).

If you say no, the conflicting changes are both inserted into the file, surrounded by conflict markers. The example below shows how a conflict region looks; the file is called `name' and the current master file version with user B's changes in it is 1.11.

<<<<<<< name
  User A's version
=======
  User B's version
>>>>>>> 1.11

Then you can resolve the conflicts by editing the file manually. Or you can type M-x vc-resolve-conflicts after visiting the file. This starts an Ediff session, as described above. Don't forget to check in the merged version afterwards.

23.8.6.4 Multi-User Branching

It is often useful for multiple developers to work simultaneously on different branches of a file. CVS allows this by default; for RCS, it is possible if you create multiple source directories. Each source directory should have a link named `RCS' which points to a common directory of RCS master files. Then each source directory can have its own choice of selected versions, but all share the same common RCS records.

This technique works reliably and automatically, provided that the source files contain RCS version headers (see section Inserting Version Control Headers). The headers enable Emacs to be sure, at all times, which version number is present in the work file.

If the files do not have version headers, you must instead tell Emacs explicitly in each session which branch you are working on. To do this, first find the file, then type C-u C-x v v and specify the correct branch number. This ensures that Emacs knows which branch it is using during this particular editing session.

23.8.7 Remote Repositories

A common way of using CVS is to set up a central CVS repository on some Internet host, then have each developer check out a personal working copy of the files on his local machine. Committing changes to the repository, and picking up changes from other users into one's own working area, then works by direct interactions with the CVS server.

One difficulty is that access to the CVS server is often slow, and that developers might need to work off-line as well. VC is designed to reduce the amount of network interaction necessary.

23.8.7.1 Version Backups

When VC sees that the CVS repository for a file is on a remote machine, it automatically makes local backups of unmodified versions of the file--automatic version backups. This means that you can compare the file to the repository version (C-x v =), or revert to that version (C-x v u), without any network interactions.

The local copy of the unmodified file is called a version backup to indicate that it corresponds exactly to a version that is stored in the repository. Note that version backups are not the same as ordinary Emacs backup files (see section Backup Files). But they follow a similar naming convention.

For a file that comes from a remote CVS repository, VC makes a version backup whenever you save the first changes to the file, and removes it after you have committed your modified version to the repository. You can disable the making of automatic version backups by setting vc-cvs-stay-local to nil (see section Options specific for CVS).

The name of the automatic version backup for version version of file file is file.~version.~. This is almost the same as the name used by C-x v ~ (see section Examining And Comparing Old Versions), the only difference being the additional dot (`.') after the version number. This similarity is intentional, because both kinds of files store the same kind of information. The file made by C-x v ~ acts as a manual version backup.

All the VC commands that operate on old versions of a file can use both kinds of version backups. For instance, C-x v ~ uses either an automatic or a manual version backup, if possible, to get the contents of the version you request. Likewise, C-x v = and C-x v u use either an automatic or a manual version backup, if one of them exists, to get the contents of a version to compare or revert to. If you changed a file outside of Emacs, so that no automatic version backup was created for the previous text, you can create a manual backup of that version using C-x v ~, and thus obtain the benefit of the local copy for Emacs commands.

The only difference in Emacs's handling of manual and automatic version backups, once they exist, is that Emacs deletes automatic version backups when you commit to the repository. By contrast, manual version backups remain until you delete them.

23.8.7.2 Local Version Control

When you make many changes to a file that comes from a remote repository, it can be convenient to have version control on your local machine as well. You can then record intermediate versions, revert to a previous state, etc., before you actually commit your changes to the remote server.

VC lets you do this by putting a file under a second, local version control system, so that the file is effectively registered in two systems at the same time. For the description here, we will assume that the remote system is CVS, and you use RCS locally, although the mechanism works with any combination of version control systems (back ends).

To make it work with other back ends, you must make sure that the "more local" back end comes before the "more remote" back end in the setting of vc-handled-backends (see section Customizing VC). By default, this variable is set up so that you can use remote CVS and local RCS as described here.

To start using local RCS for a file that comes from a remote CVS server, you must register the file in RCS, by typing C-u C-x v v rcs RET. (In other words, use vc-next-action with a prefix argument, and specify RCS as the back end.)

You can do this at any time; it does not matter whether you have already modified the file with respect to the version in the CVS repository. If possible, VC tries to make the RCS master start with the unmodified repository version, then checks in any local changes as a new version. This works if you have not made any changes yet, or if the unmodified repository version exists locally as a version backup (see section Version Backups). If the unmodified version is not available locally, the RCS master starts with the modified version; the only drawback to this is that you cannot compare your changes locally to what is stored in the repository.

The version number of the RCS master is derived from the current CVS version, starting a branch from it. For example, if the current CVS version is 1.23, the local RCS branch will be 1.23.1. Version 1.23 in the RCS master will be identical to version 1.23 under CVS; your first changes are checked in as 1.23.1.1. (If the unmodified file is not available locally, VC will check in the modified file twice, both as 1.23 and 1.23.1.1, to make the revision numbers consistent.)

If you do not use locking under CVS (the default), locking is also disabled for RCS, so that editing under RCS works exactly as under CVS.

When you are done with local editing, you can commit the final version back to the CVS repository by typing C-u C-x v v cvs RET. This initializes the log entry buffer (see section Features of the Log Entry Buffer) to contain all the log entries you have recorded in the RCS master; you can edit them as you wish, and then commit in CVS by typing C-c C-c. If the commit is successful, VC removes the RCS master, so that the file is once again registered under CVS only. (The RCS master is not actually deleted, just renamed by appending `~' to the name, so that you can refer to it later if you wish.)

While using local RCS, you can pick up recent changes from the CVS repository into your local file, or commit some of your changes back to CVS, without terminating local RCS version control. To do this, switch to the CVS back end temporarily, with the C-x v b command:

C-x v b

Switch to another back end that the current file is registered under (vc-switch-backend).

C-u C-x v b backend RET

Switch to backend for the current file.

C-x v b does not change the buffer contents, or any files; it only changes VC's perspective on how to handle the file. Any subsequent VC commands for that file will operate on the back end that is currently selected.

If the current file is registered in more than one back end, typing C-x v b "cycles" through all of these back ends. With a prefix argument, it asks for the back end to use in the minibuffer.

Thus, if you are using local RCS, and you want to pick up some recent changes in the file from remote CVS, first visit the file, then type C-x v b to switch to CVS, and finally use C-x v m RET to merge the news (see section Merging Branches). You can then switch back to RCS by typing C-x v b again, and continue to edit locally.

But if you do this, the revision numbers in the RCS master no longer correspond to those of CVS. Technically, this is not a problem, but it can become difficult to keep track of what is in the CVS repository and what is not. So we suggest that you return from time to time to CVS-only operation, by committing your local changes back to the repository using C-u C-x v v cvs RET.

23.8.8 Snapshots

A snapshot is a named set of file versions (one for each registered file) that you can treat as a unit. One important kind of snapshot is a release, a (theoretically) stable version of the system that is ready for distribution to users.

23.8.8.1 Making and Using Snapshots

There are two basic commands for snapshots; one makes a snapshot with a given name, the other retrieves a named snapshot.

C-x v s name RET

Define the last saved versions of every registered file in or under the current directory as a snapshot named name (vc-create-snapshot).

C-x v r name RET

For all registered files at or below the current directory level, select whatever versions correspond to the snapshot name (vc-retrieve-snapshot).

This command reports an error if any files are locked at or below the current directory, without changing anything; this is to avoid overwriting work in progress.

A snapshot uses a very small amount of resources--just enough to record the list of file names and which version belongs to the snapshot. Thus, you need not hesitate to create snapshots whenever they are useful.

You can give a snapshot name as an argument to C-x v = or C-x v ~ (see section Examining And Comparing Old Versions). Thus, you can use it to compare a snapshot against the current files, or two snapshots against each other, or a snapshot against a named version.

23.8.8.2 Snapshot Caveats

VC's snapshot facilities are modeled on RCS's named-configuration support. They use RCS's native facilities for this, so snapshots made using RCS through VC are visible even when you bypass VC.

With CVS, Meta-CVS, and Subversion, VC also uses the native mechanism provided by that back end to make snapshots and retrieve them (tags for CVS and Meta-CVS, copies for Subversion).

For SCCS, VC implements snapshots itself. The files it uses contain name/file/version-number triples. These snapshots are visible only through VC.

There is no support for VC snapshots using GNU Arch yet.

A snapshot is a set of checked-in versions. So make sure that all the files are checked in and not locked when you make a snapshot.

File renaming and deletion can create some difficulties with snapshots. This is not a VC-specific problem, but a general design issue in version control systems that no one has solved very well yet.

If you rename a registered file, you need to rename its master along with it (the command vc-rename-file does this automatically). If you are using SCCS, you must also update the records of the snapshot, to mention the file by its new name (vc-rename-file does this, too). An old snapshot that refers to a master file that no longer exists under the recorded name is invalid; VC can no longer retrieve it. It would be beyond the scope of this manual to explain enough about RCS and SCCS to explain how to update the snapshots by hand.

Using vc-rename-file makes the snapshot remain valid for retrieval, but it does not solve all problems. For example, some of the files in your program probably refer to others by name. At the very least, the makefile probably mentions the file that you renamed. If you retrieve an old snapshot, the renamed file is retrieved under its new name, which is not the name that the makefile expects. So the program won't really work as retrieved.

23.8.9 Miscellaneous Commands and Features of VC

This section explains the less-frequently-used features of VC.

23.8.9.1 Change Logs and VC

If you use RCS or CVS for a program and also maintain a change log file for it (see section Change Logs), you can generate change log entries automatically from the version control log entries:

C-x v a

Visit the current directory's change log file and, for registered files in that directory, create new entries for versions checked in since the most recent entry in the change log file. (vc-update-change-log).

This command works with RCS or CVS only, not with any of the other back ends.

C-u C-x v a

As above, but only find entries for the current buffer's file.

M-1 C-x v a

As above, but find entries for all the currently visited files that are maintained with version control. This works only with RCS, and it puts all entries in the log for the default directory, which may not be appropriate.

For example, suppose the first line of `ChangeLog' is dated 1999-04-10, and that the only check-in since then was by Nathaniel Bowditch to `rcs2log' on 1999-05-22 with log text `Ignore log messages that start with `#'.'. Then C-x v a visits `ChangeLog' and inserts text like this:

1999-05-22  Nathaniel Bowditch  <nat@apn.org>

        * rcs2log: Ignore log messages that start with `#'.

You can then edit the new change log entry further as you wish.

Some of the new change log entries may duplicate what's already in ChangeLog. You will have to remove these duplicates by hand.

Normally, the log entry for file `foo' is displayed as `* foo: text of log entry'. The `:' after `foo' is omitted if the text of the log entry starts with `(functionname): '. For example, if the log entry for `vc.el' is `(vc-do-command): Check call-process status.', then the text in `ChangeLog' looks like this:

1999-05-06  Nathaniel Bowditch  <nat@apn.org>

        * vc.el (vc-do-command): Check call-process status.

When C-x v a adds several change log entries at once, it groups related log entries together if they all are checked in by the same author at nearly the same time. If the log entries for several such files all have the same text, it coalesces them into a single entry. For example, suppose the most recent check-ins have the following log entries:

• For `vc.texinfo': `Fix expansion typos.' • For `vc.el': `Don't call expand-file-name.' • For `vc-hooks.el': `Don't call expand-file-name.'

They appear like this in `ChangeLog':

1999-04-01  Nathaniel Bowditch  <nat@apn.org>

        * vc.texinfo: Fix expansion typos.

        * vc.el, vc-hooks.el: Don't call expand-file-name.

Normally, C-x v a separates log entries by a blank line, but you can mark several related log entries to be clumped together (without an intervening blank line) by starting the text of each related log entry with a label of the form `{clumpname} '. The label itself is not copied to `ChangeLog'. For example, suppose the log entries are:

• For `vc.texinfo': `{expand} Fix expansion typos.' • For `vc.el': `{expand} Don't call expand-file-name.' • For `vc-hooks.el': `{expand} Don't call expand-file-name.'

Then the text in `ChangeLog' looks like this:

1999-04-01  Nathaniel Bowditch  <nat@apn.org>

        * vc.texinfo: Fix expansion typos.
        * vc.el, vc-hooks.el: Don't call expand-file-name.

A log entry whose text begins with `#' is not copied to `ChangeLog'. For example, if you merely fix some misspellings in comments, you can log the change with an entry beginning with `#' to avoid putting such trivia into `ChangeLog'.

23.8.9.2 Renaming VC Work Files and Master Files

When you rename a registered file, you must also rename its master file correspondingly to get proper results. Use vc-rename-file to rename the source file as you specify, and rename its master file accordingly. It also updates any snapshots (see section Snapshots) that mention the file, so that they use the new name; despite this, the snapshot thus modified may not completely work (see section Snapshot Caveats).

Some back ends do not provide an explicit rename operation to their repositories. After issuing vc-rename-file, use C-x v v on the original and renamed buffers and provide the necessary edit log.

You cannot use vc-rename-file on a file that is locked by someone else.

23.8.9.3 Inserting Version Control Headers

Sometimes it is convenient to put version identification strings directly into working files. Certain special strings called version headers are replaced in each successive version by the number of that version, the name of the user who created it, and other relevant information. All of the back ends that VC supports have such a mechanism, except GNU Arch.

VC does not normally use the information contained in these headers. The exception is RCS--with RCS, version headers are sometimes more reliable than the master file to determine which version of the file you are editing. Note that in a multi-branch environment, version headers are necessary to make VC behave correctly (see section Multi-User Branching).

Searching for RCS version headers is controlled by the variable vc-consult-headers. If it is non-nil (the default), Emacs searches for headers to determine the version number you are editing. Setting it to nil disables this feature.

Note that although CVS uses the same kind of version headers as RCS does, VC never searches for these headers if you are using CVS, regardless of the above setting.

You can use the C-x v h command (vc-insert-headers) to insert a suitable header string.

C-x v h

Insert headers in a file for use with your version-control system.

The default header string is `$Id$' for RCS and `%W%' for SCCS. You can specify other headers to insert by setting the variables vc-backend-header where backend is rcs or sccs.

Instead of a single string, you can specify a list of strings; then each string in the list is inserted as a separate header on a line of its own.

It may be necessary to use apparently-superfluous backslashes when writing the strings that you put in this variable. For instance, you might write "$Id\$" rather than "$Id$". The extra backslash prevents the string constant from being interpreted as a header, if the Emacs Lisp file containing it is maintained with version control.

Each header is inserted surrounded by tabs, inside comment delimiters, on a new line at point. Normally the ordinary comment start and comment end strings of the current mode are used, but for certain modes, there are special comment delimiters for this purpose; the variable vc-comment-alist specifies them. Each element of this list has the form (mode starter ender).

The variable vc-static-header-alist specifies further strings to add based on the name of the buffer. Its value should be a list of elements of the form (regexp . format). Whenever regexp matches the buffer name, format is inserted as part of the header. A header line is inserted for each element that matches the buffer name, and for each string specified by vc-backend-header. The header line is made by processing the string from vc-backend-header with the format taken from the element. The default value for vc-static-header-alist is as follows:

(("\\.c$" .
  "\n#ifndef lint\nstatic char vcid[] = \"\%s\";\n\
#endif /* lint */\n"))

It specifies insertion of text of this form:

#ifndef lint
static char vcid[] = "string";
#endif /* lint */

Note that the text above starts with a blank line.

If you use more than one version header in a file, put them close together in the file. The mechanism in revert-buffer that preserves markers may not handle markers positioned between two version headers.

23.8.10 Customizing VC

The variable vc-handled-backends determines which version control systems VC should handle. The default value is (RCS CVS SVN SCCS Arch MCVS), so it contains all six version systems that are currently supported. If you want VC to ignore one or more of these systems, exclude its name from the list. To disable VC entirely, set this variable to nil.

The order of systems in the list is significant: when you visit a file registered in more than one system (see section Local Version Control), VC uses the system that comes first in vc-handled-backends by default. The order is also significant when you register a file for the first time, see Registering a File for Version Control, for details.

23.8.10.1 General Options

Emacs normally does not save backup files for source files that are maintained with version control. If you want to make backup files even for files that use version control, set the variable vc-make-backup-files to a non-nil value.

Normally the work file exists all the time, whether it is locked or not. If you set vc-keep-workfiles to nil, then checking in a new version with C-x v v deletes the work file; but any attempt to visit the file with Emacs creates it again. (With CVS, work files are always kept.)

Editing a version-controlled file through a symbolic link can be dangerous. It bypasses the version control system--you can edit the file without locking it, and fail to check your changes in. Also, your changes might overwrite those of another user. To protect against this, VC checks each symbolic link that you visit, to see if it points to a file under version control.

The variable vc-follow-symlinks controls what to do when a symbolic link points to a version-controlled file. If it is nil, VC only displays a warning message. If it is t, VC automatically follows the link, and visits the real file instead, telling you about this in the echo area. If the value is ask (the default), VC asks you each time whether to follow the link.

If vc-suppress-confirm is non-nil, then C-x v v and C-x v i can save the current buffer without asking, and C-x v u also operates without asking for confirmation. (This variable does not affect C-x v c; that operation is so drastic that it should always ask for confirmation.)

VC mode does much of its work by running the shell commands for RCS, CVS and SCCS. If vc-command-messages is non-nil, VC displays messages to indicate which shell commands it runs, and additional messages when the commands finish.

You can specify additional directories to search for version control programs by setting the variable vc-path. These directories are searched before the usual search path. It is rarely necessary to set this variable, because VC normally finds the proper files automatically.

23.8.10.2 Options for RCS and SCCS

By default, RCS uses locking to coordinate the activities of several users, but there is a mode called non-strict locking in which you can check-in changes without locking the file first. Use `rcs -U' to switch to non-strict locking for a particular file, see the rcs manual page for details.

When deducing the version control state of an RCS file, VC first looks for an RCS version header string in the file (see section Inserting Version Control Headers). If there is no header string, VC normally looks at the file permissions of the work file; this is fast. But there might be situations when the file permissions cannot be trusted. In this case the master file has to be consulted, which is rather expensive. Also the master file can only tell you if there's any lock on the file, but not whether your work file really contains that locked version.

You can tell VC not to use version headers to determine the file status by setting vc-consult-headers to nil. VC then always uses the file permissions (if it is supposed to trust them), or else checks the master file.

You can specify the criterion for whether to trust the file permissions by setting the variable vc-mistrust-permissions. Its value can be t (always mistrust the file permissions and check the master file), nil (always trust the file permissions), or a function of one argument which makes the decision. The argument is the directory name of the `RCS' subdirectory. A non-nil value from the function says to mistrust the file permissions. If you find that the file permissions of work files are changed erroneously, set vc-mistrust-permissions to t. Then VC always checks the master file to determine the file's status.

VC determines the version control state of files under SCCS much as with RCS. It does not consider SCCS version headers, though. Thus, the variable vc-mistrust-permissions affects SCCS use, but vc-consult-headers does not.

23.8.10.3 Options specific for CVS

By default, CVS does not use locking to coordinate the activities of several users; anyone can change a work file at any time. However, there are ways to restrict this, resulting in behavior that resembles locking.

For one thing, you can set the CVSREAD environment variable (the value you use makes no difference). If this variable is defined, CVS makes your work files read-only by default. In Emacs, you must type C-x v v to make the file writable, so that editing works in fact similar as if locking was used. Note however, that no actual locking is performed, so several users can make their files writable at the same time. When setting CVSREAD for the first time, make sure to check out all your modules anew, so that the file protections are set correctly.

Another way to achieve something similar to locking is to use the watch feature of CVS. If a file is being watched, CVS makes it read-only by default, and you must also use C-x v v in Emacs to make it writable. VC calls cvs edit to make the file writable, and CVS takes care to notify other developers of the fact that you intend to change the file. See the CVS documentation for details on using the watch feature.

When a file's repository is on a remote machine, VC tries to keep network interactions to a minimum. This is controlled by the variable vc-cvs-stay-local. There is another variable, vc-stay-local, which enables the feature also for other back ends that support it, including CVS. In the following, we will talk only about vc-cvs-stay-local, but everything applies to vc-stay-local as well.

If vc-cvs-stay-local is t (the default), then VC uses only the entry in the local CVS subdirectory to determine the file's state (and possibly information returned by previous CVS commands). One consequence of this is that when you have modified a file, and somebody else has already checked in other changes to the file, you are not notified of it until you actually try to commit. (But you can try to pick up any recent changes from the repository first, using C-x v m RET, see section Merging Branches).

When vc-cvs-stay-local is t, VC also makes local version backups, so that simple diff and revert operations are completely local (see section Version Backups).

On the other hand, if you set vc-cvs-stay-local to nil, then VC queries the remote repository before it decides what to do in vc-next-action (C-x v v), just as it does for local repositories. It also does not make any version backups.

You can also set vc-cvs-stay-local to a regular expression that is matched against the repository host name; VC then stays local only for repositories from hosts that match the pattern.

You can specify additional command line options to pass to all CVS operations in the variable vc-cvs-global-switches. These switches are inserted immediately after the cvs command, before the name of the operation to invoke.

23.9 File Directories

The file system groups files into directories. A directory listing is a list of all the files in a directory. Emacs provides commands to create and delete directories, and to make directory listings in brief format (file names only) and verbose format (sizes, dates, and authors included). Emacs also includes a directory browser feature called Dired; see Dired, the Directory Editor.

C-x C-d dir-or-pattern RET

Display a brief directory listing (list-directory).

C-u C-x C-d dir-or-pattern RET

Display a verbose directory listing.

M-x make-directory RET dirname RET

Create a new directory named dirname.

M-x delete-directory RET dirname RET

Delete the directory named dirname. It must be empty, or you get an error.

The command to display a directory listing is C-x C-d (list-directory). It reads using the minibuffer a file name which is either a directory to be listed or a wildcard-containing pattern for the files to be listed. For example,

C-x C-d /u2/emacs/etc RET

lists all the files in directory `/u2/emacs/etc'. Here is an example of specifying a file name pattern:

C-x C-d /u2/emacs/src/*.c RET

Normally, C-x C-d displays a brief directory listing containing just file names. A numeric argument (regardless of value) tells it to make a verbose listing including sizes, dates, and owners (like `ls -l').

The text of a directory listing is mostly obtained by running ls in an inferior process. Two Emacs variables control the switches passed to ls: list-directory-brief-switches is a string giving the switches to use in brief listings ("-CF" by default), and list-directory-verbose-switches is a string giving the switches to use in a verbose listing ("-l" by default).

In verbose directory listings, Emacs adds information about the amount of free space on the disk that contains the directory. To do this, it runs the program specified by directory-free-space-program with arguments directory-free-space-args.

23.10 Comparing Files

The command M-x diff compares two files, displaying the differences in an Emacs buffer named `*diff*'. It works by running the diff program, using options taken from the variable diff-switches. The value of diff-switches should be a string; the default is "-c" to specify a context diff. See (diff)Top section `Diff' in Comparing and Merging Files, for more information about diff output formats.

The command M-x diff-backup compares a specified file with its most recent backup. If you specify the name of a backup file, diff-backup compares it with the source file that it is a backup of.

The command M-x compare-windows compares the text in the current window with that in the next window. (For more information about windows in Emacs, Multiple Windows.) Comparison starts at point in each window, after pushing each initial point value on the mark ring in its respective buffer. Then it moves point forward in each window, one character at a time, until it reaches characters that don't match. Then the command exits.

If point in the two windows is followed by non-matching text when the command starts, M-x compare-windows tries heuristically to advance up to matching text in the two windows, and then exits. So if you use M-x compare-windows repeatedly, each time it either skips one matching range or finds the start of another.

With a numeric argument, compare-windows ignores changes in whitespace. If the variable compare-ignore-case is non-nil, the comparison ignores differences in case as well. If the variable compare-ignore-whitespace is non-nil, compare-windows normally ignores changes in whitespace, and a prefix argument turns that off.

You can use M-x smerge-mode to turn on Smerge mode, a minor mode for editing output from the diff3 program. This is typically the result of a failed merge from a version control system "update" outside VC, due to conflicting changes to a file. Smerge mode provides commands to resolve conflicts by selecting specific changes.

See section Merging Files with Emerge, for the Emerge facility, which provides a powerful interface for merging files.

23.11 Diff Mode

Diff mode is used for the output of M-x diff; it is also useful for editing patches and comparisons produced by the diff program. To select Diff mode manually, type M-x diff-mode.

One general feature of Diff mode is that manual edits to the patch automatically correct line numbers, including those in the hunk header, so that you can actually apply the edited patch. Diff mode treats each hunk location as an "error message," so that you can use commands such as C-x ' to visit the corresponding source locations. It also provides the following commands to navigate, manipulate and apply parts of patches:

M-n

Move to the next hunk-start (diff-hunk-next).

M-p

Move to the previous hunk-start (diff-hunk-prev).

M-}

Move to the next file-start, in a multi-file patch (diff-file-next).

M-{

Move to the previous file-start, in a multi-file patch (diff-file-prev).

M-k

Kill the hunk at point (diff-hunk-kill).

M-K

In a multi-file patch, kill the current file part. (diff-file-kill).

C-c C-a

Apply this hunk to its target file (diff-apply-hunk). With a prefix argument of C-u, revert this hunk.

C-c C-c

Go to the source corresponding to this hunk (diff-goto-source).

C-c C-e

Start an Ediff session with the patch (diff-ediff-patch). See Ediff: (ediff)Top section `Ediff' in The Ediff Manual.

C-c C-n

Restrict the view to the current hunk (diff-restrict-view). See section Narrowing. With a prefix argument of C-u, restrict the view to the current patch of a multiple file patch. To widen again, use C-x n w.

C-c C-r

Reverse the direction of comparison for the entire buffer (diff-reverse-direction).

C-c C-s

Split the hunk at point (diff-split-hunk). This is for manually editing patches, and only works with the unified diff format.

C-c C-u

Convert the entire buffer to unified format (diff-context->unified). With a prefix argument, convert unified format to context format. In Transient Mark mode, when the mark is active, this command operates only on the region.

C-c C-w

Refine the current hunk so that it disregards changes in whitespace (diff-refine-hunk).

C-x 4 a in Diff mode operates on behalf of the target file, but gets the function name from the patch itself. See section Change Logs. This is useful for making log entries for functions that are deleted by the patch.

23.12 Miscellaneous File Operations

Emacs has commands for performing many other operations on files. All operate on one file; they do not accept wildcard file names.

M-x view-file allows you to scan or read a file by sequential screenfuls. It reads a file name argument using the minibuffer. After reading the file into an Emacs buffer, view-file displays the beginning. You can then type SPC to scroll forward one windowful, or DEL to scroll backward. Various other commands are provided for moving around in the file, but none for changing it; type ? while viewing for a list of them. They are mostly the same as normal Emacs cursor motion commands. To exit from viewing, type q. The commands for viewing are defined by a special minor mode called View mode.

A related command, M-x view-buffer, views a buffer already present in Emacs. See section Miscellaneous Buffer Operations.

M-x insert-file (also C-x i) inserts a copy of the contents of the specified file into the current buffer at point, leaving point unchanged before the contents and the mark after them.

M-x insert-file-literally is like M-x insert-file, except the file is inserted "literally": it is treated as a sequence of ASCII characters with no special encoding or conversion, similar to the M-x find-file-literally command (see section Visiting Files).

M-x write-region is the inverse of M-x insert-file; it copies the contents of the region into the specified file. M-x append-to-file adds the text of the region to the end of the specified file. See section Accumulating Text. The variable write-region-inhibit-fsync applies to these commands, as well as saving files; see Customizing Saving of Files.

M-x delete-file deletes the specified file, like the rm command in the shell. If you are deleting many files in one directory, it may be more convenient to use Dired (see section Dired, the Directory Editor).

M-x rename-file reads two file names old and new using the minibuffer, then renames file old as new. If the file name new already exists, you must confirm with yes or renaming is not done; this is because renaming causes the old meaning of the name new to be lost. If old and new are on different file systems, the file old is copied and deleted.

If the argument new is just a directory name, the real new name is in that directory, with the same non-directory component as old. For example, M-x rename-file RET ~/foo RET /tmp RET renames `~/foo' to `/tmp/foo'. The same rule applies to all the remaining commands in this section. All of them ask for confirmation when the new file name already exists, too.

The similar command M-x add-name-to-file is used to add an additional name to an existing file without removing its old name. The new name is created as a "hard link" to the existing file. The new name must belong on the same file system that the file is on. On MS-Windows, this command works only if the file resides in an NTFS file system. On MS-DOS, it works by copying the file.

M-x copy-file reads the file old and writes a new file named new with the same contents.

M-x make-symbolic-link reads two file names target and linkname, then creates a symbolic link named linkname, which points at target. The effect is that future attempts to open file linkname will refer to whatever file is named target at the time the opening is done, or will get an error if the name target is nonexistent at that time. This command does not expand the argument target, so that it allows you to specify a relative name as the target of the link.

Not all systems support symbolic links; on systems that don't support them, this command is not defined.

23.13 Accessing Compressed Files

Emacs automatically uncompresses compressed files when you visit them, and automatically recompresses them if you alter them and save them. Emacs recognizes compressed files by their file names. File names ending in `.gz' indicate a file compressed with gzip. Other endings indicate other compression programs.

Automatic uncompression and compression apply to all the operations in which Emacs uses the contents of a file. This includes visiting it, saving it, inserting its contents into a buffer, loading it, and byte compiling it.

To disable this feature, type the command M-x auto-compression-mode. You can disable it permanently by customizing the variable auto-compression-mode.

23.14 File Archives

A file whose name ends in `.tar' is normally an archive made by the tar program. Emacs views these files in a special mode called Tar mode which provides a Dired-like list of the contents (see section Dired, the Directory Editor). You can move around through the list just as you would in Dired, and visit the subfiles contained in the archive. However, not all Dired commands are available in Tar mode.

If Auto Compression mode is enabled (see section Accessing Compressed Files), then Tar mode is used also for compressed archives--files with extensions `.tgz', .tar.Z and .tar.gz.

The keys e, f and RET all extract a component file into its own buffer. You can edit it there, and if you save the buffer, the edited version will replace the version in the Tar buffer. v extracts a file into a buffer in View mode. o extracts the file and displays it in another window, so you could edit the file and operate on the archive simultaneously. d marks a file for deletion when you later use x, and u unmarks a file, as in Dired. C copies a file from the archive to disk and R renames a file within the archive. g reverts the buffer from the archive on disk.

The keys M, G, and O change the file's permission bits, group, and owner, respectively.

If your display supports colors and the mouse, moving the mouse pointer across a file name highlights that file name, indicating that you can click on it. Clicking Mouse-2 on the highlighted file name extracts the file into a buffer and displays that buffer.

Saving the Tar buffer writes a new version of the archive to disk with the changes you made to the components.

You don't need the tar program to use Tar mode--Emacs reads the archives directly. However, accessing compressed archives requires the appropriate uncompression program.

A separate but similar Archive mode is used for archives produced by the programs arc, jar, lzh, zip, and zoo, which have extensions corresponding to the program names. Archive mode also works for those exe files that are self-extracting executables.

The key bindings of Archive mode are similar to those in Tar mode, with the addition of the m key which marks a file for subsequent operations, and M-DEL which unmarks all the marked files. Also, the a key toggles the display of detailed file information, for those archive types where it won't fit in a single line. Operations such as renaming a subfile, or changing its mode or owner, are supported only for some of the archive formats.

Unlike Tar mode, Archive mode runs the archiving program to unpack and repack archives. Details of the program names and their options can be set in the `Archive' Customize group. However, you don't need these programs to look at the archive table of contents, only to extract or manipulate the subfiles in the archive.

23.15 Remote Files

You can refer to files on other machines using a special file name syntax:

/host:filename
/user@host:filename
/user@host#port:filename
/method:user@host:filename
/method:user@host#port:filename

To carry out this request, Emacs uses either the FTP program or a remote-login program such as ssh, rlogin, or telnet. You can always specify in the file name which method to use--for example, `/ftp:user@host:filename' uses FTP, whereas `/ssh:user@host:filename' uses ssh. When you don't specify a method in the file name, Emacs chooses the method as follows:

1. If the host name starts with `ftp.' (with dot), then Emacs uses FTP.

2. If the user name is `ftp' or `anonymous', then Emacs uses FTP.

3. Otherwise, Emacs uses ssh.

Remote file access through FTP is handled by the Ange-FTP package, which is documented in the following. Remote file access through the other methods is handled by the Tramp package, which has its own manual. See The Tramp Manual: (tramp)Top section `Top' in The Tramp Manual.

When the Ange-FTP package is used, Emacs logs in through FTP using your user name or the name user. It may ask you for a password from time to time; this is used for logging in on host. The form using port allows you to access servers running on a non-default TCP port.

If you want to disable backups for remote files, set the variable ange-ftp-make-backup-files to nil.

By default, the auto-save files (see section Auto-Save Files) for remote files are made in the temporary file directory on the local machine. This is achieved using the variable auto-save-file-name-transforms.

Normally, if you do not specify a user name in a remote file name, that means to use your own user name. But if you set the variable ange-ftp-default-user to a string, that string is used instead.

To visit files accessible by anonymous FTP, you use special user names `anonymous' or `ftp'. Passwords for these user names are handled specially. The variable ange-ftp-generate-anonymous-password controls what happens: if the value of this variable is a string, then that string is used as the password; if non-nil (the default), then the value of user-mail-address is used; if nil, then Emacs prompts you for a password as usual.

Sometimes you may be unable to access files on a remote machine because a firewall in between blocks the connection for security reasons. If you can log in on a gateway machine from which the target files are accessible, and whose FTP server supports gatewaying features, you can still use remote file names; all you have to do is specify the name of the gateway machine by setting the variable ange-ftp-gateway-host, and set ange-ftp-smart-gateway to t. Otherwise you may be able to make remote file names work, but the procedure is complex. You can read the instructions by typing M-x finder-commentary RET ange-ftp RET.

You can entirely turn off the FTP file name feature by removing the entries ange-ftp-completion-hook-function and ange-ftp-hook-function from the variable file-name-handler-alist. You can turn off the feature in individual cases by quoting the file name with `/:' (see section Quoted File Names).

23.16 Quoted File Names

You can quote an absolute file name to prevent special characters and syntax in it from having their special effects. The way to do this is to add `/:' at the beginning.

For example, you can quote a local file name which appears remote, to prevent it from being treated as a remote file name. Thus, if you have a directory named `/foo:' and a file named `bar' in it, you can refer to that file in Emacs as `/:/foo:/bar'.

`/:' can also prevent `~' from being treated as a special character for a user's home directory. For example, `/:/tmp/~hack' refers to a file whose name is `~hack' in directory `/tmp'.

Quoting with `/:' is also a way to enter in the minibuffer a file name that contains `$'. In order for this to work, the `/:' must be at the beginning of the minibuffer contents. (You can also double each `$'; see File Names with $.)

You can also quote wildcard characters with `/:', for visiting. For example, `/:/tmp/foo*bar' visits the file `/tmp/foo*bar'.

Another method of getting the same result is to enter `/tmp/foo[*]bar', which is a wildcard specification that matches only `/tmp/foo*bar'. However, in many cases there is no need to quote the wildcard characters because even unquoted they give the right result. For example, if the only file name in `/tmp' that starts with `foo' and ends with `bar' is `foo*bar', then specifying `/tmp/foo*bar' will visit only `/tmp/foo*bar'.

23.17 File Name Cache

You can use the file name cache to make it easy to locate a file by name, without having to remember exactly where it is located. When typing a file name in the minibuffer, C-tab (file-cache-minibuffer-complete) completes it using the file name cache. If you repeat C-tab, that cycles through the possible completions of what you had originally typed. (However, note that the C-tab character cannot be typed on most text-only terminals.)

The file name cache does not fill up automatically. Instead, you load file names into the cache using these commands:

M-x file-cache-add-directory RET directory RET

Add each file name in directory to the file name cache.

M-x file-cache-add-directory-using-find RET directory RET

Add each file name in directory and all of its nested subdirectories to the file name cache.

M-x file-cache-add-directory-using-locate RET directory RET

Add each file name in directory and all of its nested subdirectories to the file name cache, using locate to find them all.

M-x file-cache-add-directory-list RET variable RET

Add each file name in each directory listed in variable to the file name cache. variable should be a Lisp variable such as load-path or exec-path, whose value is a list of directory names.

M-x file-cache-clear-cache RET

Clear the cache; that is, remove all file names from it.

The file name cache is not persistent: it is kept and maintained only for the duration of the Emacs session. You can view the contents of the cache with the file-cache-display command.

23.18 Convenience Features for Finding Files

In this section, we introduce some convenient facilities for finding recently-opened files, reading file names from a buffer, and viewing image files.

If you enable Recentf mode, with M-x recentf-mode, the `File' menu includes a submenu containing a list of recently opened files. M-x recentf-save-list saves the current recent-file-list to a file, and M-x recentf-edit-list edits it.

The M-x ffap command generalizes find-file with more powerful heuristic defaults (see section Finding Files and URLs at Point), often based on the text at point. Partial Completion mode offers other features extending find-file, which can be used with ffap. See section Completion Options.

Visiting image files automatically selects Image mode. This major mode allows you to toggle between displaying the file as an image in the Emacs buffer, and displaying its underlying text representation, using the command C-c C-c (image-toggle-display). This works only when Emacs can display the specific image type.

See also the Image-Dired package (see section Viewing Image Thumbnails in Dired) for viewing images as thumbnails.

23.19 Filesets

If you regularly edit a certain group of files, you can define them as a fileset. This lets you perform certain operations, such as visiting, query-replace, and shell commands on all the files at once. To make use of filesets, you must first add the expression (filesets-init) to your `.emacs' file (see section The Init File, `~/.emacs'). This adds a `Filesets' menu to the menu bar.

The simplest way to define a fileset is by adding files to it one at a time. To add a file to fileset name, visit the file and type M-x filesets-add-buffer RET name RET. If there is no fileset name, this creates a new one, which initially creates only the current file. The command M-x filesets-remove-buffer removes the current file from a fileset.

You can also edit the list of filesets directly, with M-x filesets-edit (or by choosing `Edit Filesets' from the `Filesets' menu). The editing is performed in a Customize buffer (see section Easy Customization Interface). Filesets need not be a simple list of files--you can also define filesets using regular expression matching file names. Some examples of these more complicated filesets are shown in the Customize buffer. Remember to select `Save for future sessions' if you want to use the same filesets in future Emacs sessions.

You can use the command M-x filesets-open to visit all the files in a fileset, and M-x filesets-close to close them. Use M-x filesets-run-cmd to run a shell command on all the files in a fileset. These commands are also available from the `Filesets' menu, where each existing fileset is represented by a submenu.

This document was generated by Mark Kaminski on July, 3 2008 using texi2html 1.70.