Go to the previous, next section.

Compiling and Testing Programs

The previous chapter discusses the Emacs commands that are useful for making changes in programs. This chapter deals with commands that assist in the larger process of developing and maintaining programs.

Running `make', or Compilers Generally

Emacs can run compilers for noninteractive languages such as C and Fortran as inferior processes, feeding the error log into an Emacs buffer. It can also parse the error messages and show you the source lines where compilation errors occurred.

M-x compile
Run a compiler asynchronously under Emacs, with error messages to `*compilation*' buffer.
M-x grep
Run grep asynchronously under Emacs, with matching lines listed in the buffer named `*grep*'.
M-x kill-compilation
M-x kill-grep
Kill the running compilation or grep subprocess.
C-x `
Visit the locus of the next compiler error message or grep match.

To run make or another compilation command, do M-x compile. This command reads a shell command line using the minibuffer, and then executes the command in an inferior shell, putting output in the buffer named `*compilation*'. The current buffer's default directory is used as the working directory for the execution of the command; normally, therefore, the compilation happens in this directory.

When the shell command line is read, the minibuffer appears containing a default command line, which is the command you used the last time you did M-x compile. If you type just RET, the same command line is used again. For the first M-x compile, the default is `make -k'.

The default compilation command comes from the variable compile-command; if the appropriate compilation command for a file is something other than `make -k', it can be useful for the file to specify a local value for compile-command (see section Local Variables in Files).

Starting a compilation displays the buffer `*compilation*' in another window but does not select it. The buffer's mode line tells you whether compilation is finished, with the word `run' or `exit' inside the parentheses. You do not have to keep this buffer visible; compilation continues in any case. While a compilation is going on, the string `Compiling' appears in the mode lines of all windows. When this string disappears, the compilation is finished.

To kill the compilation process, do M-x kill-compilation. When the compiler process terminates, the mode line of the `*compilation*' buffer changes to say `signal' instead of `run'. Starting a new compilation also kills any running compilation, as only one can exist at any time. However, M-x compile asks for confirmation before actually killing a compilation that is running.

The `*compilation*' buffer uses a special major mode, Compilation mode. This mode provides the keys SPC and DEL to scroll by screenfuls, and M-n and M-p to move to the next or previous error message. You can also use M-{ and M-} to move up or down to an error message for a different source file. You can visit the source for any particular error message by moving point in `*compilation*' to that error message and typing C-c C-c (compile-goto-error).

To parse the compiler error messages sequentially, type C-x ` (next-error). The character following the C-x is the backquote or "grave accent," not the single-quote. This command is available in all buffers, not just in `*compilation*'. It displays the next error message at the top of one window and source location of the error in another window.

The first time C-x ` is used after the start of a compilation, it moves to the first error's location. Subsequent uses of C-x ` advance down the data set up by the first use. When the preparsed error messages are exhausted, the next C-x ` checks for any more error messages that have come in; this is useful if you start editing the compilation errors while the compilation is still going on. If no more error messages have come in, C-x ` reports an error.

C-u C-x ` discards the preparsed error message data and parses the `*compilation*' buffer over again, then displaying the first error. This way, you can process the same set of errors again.

Instead of running a compiler, you can run grep and see the lines on which matches were found. To do this, type M-x grep with an argument line that contains the same arguments you would give grep when running it normally: a grep-style regexp (usually in single-quotes to quote the shell's special characters) followed by file names which may use wildcards. The output from grep goes in the `*grep*' buffer and the lines that matched can be found with C-x ` as if they were compilation errors.

Note: a shell is used to run the compile command, but the shell is told that it should be noninteractive. This means in particular that the shell starts up with no prompt. If you find your usual shell prompt making an unsightly appearance in the `*compilation*' buffer, it means you have made a mistake in your shell's init file (`.cshrc' or `.shrc' or ...) by setting the prompt unconditionally. The shell init file should set the prompt only if there already is a prompt. In csh, here is how to do it:

if ($?prompt) set prompt = ...

Compilation Minor mode lets you parse error messages in any buffer, not just a normal compilation output buffer. Type M-x compilation-minor-mode to enable the minor mode. This defines the key C-c C-c to find the source code for the error message at point, just as in the Compilation major mode. In an rlogin buffer (see section Remote Host), this automatically accesses remote source files by FTP (see section File Names).

The features of Compilation mode are also available in a minor mode called Compilation Minor mode.

Running Debuggers Under Emacs

The GUD (Grand Unified Debugger) library provides an interface to various symbolic debuggers from within Emacs. We recommend the debugger GDB, which is free software, but you can also run DBX, SDB or XDB if you have them.

Starting GUD

There are four commands for starting a debugger, each corresponding to a particular debugger program.

M-x gdb RET file RET
Run GDB as a subprocess of Emacs. This command creates a buffer for input and output to GDB, and switches to it. If a GDB buffer already exists, it just switches to that buffer.

M-x dbx RET file RET
Similar, but run DBX instead of GDB.

M-x xdb RET file RET
Similar, but run XDB instead of GDB. Use the variable gud-xdb-directories to specify directories to search for source files.

M-x sdb RET file RET
Similar, but run SDB instead of GDB.

Some versions of SDB do not mention source file names in their messages. When you use them, you need to have a valid tags table (see section Tag Tables) in order for GUD to find functions in the source code. If you have not visited a tags table or the tags table doesn't list one of the functions, you get a message saying `The sdb support requires a valid tags table to work'. If this happens, generate a valid tags table in the working directory and try again.

You can only run one debugger process at a time.

Each of these commands takes one argument: a command line to invoke the debugger. In the simplest case, specify just the name of the executable file you want to debug. You may also use options that the debugger supports. However, shell wild cards and variables are not allowed. GUD assumes that the first argument not preceded by a `-' is the executable file name.

Debugger Operation

When you run a debugger with GUD, the debugger uses an Emacs buffer for its ordinary input and output. This is called the GUD buffer. The debugger uses other Emacs buffers to display the source files of the program. An arrow (`=>') in one of these buffers indicates the current execution line. Moving point in this buffer does not move the arrow.

You can start editing these source files at any time in the buffers that were made to display them. The arrow is not part of the file's text; it appears only on the screen. If you do modify a source file, keep in mind that inserting or deleting lines will throw off the arrow's positioning; GUD has no way of figuring out which line corresponded before your changes to the line number in a debugger message. Also, you'll typically have to recompile and restart the program for your changes to be reflected in the debugger's tables.

If you wish, you can control your debugger process entirely through the debugger buffer, which uses a variant of Shell mode. All the usual commands for your debugger are available, and you can use the Shell mode history commands to repeat them.

Commands of GUD

The GUD interaction buffer uses a variant of Shell mode, so the commands of Shell mode are available (see section Shell Mode).

GUD provides a command available in all buffers for setting breakpoints. This command is defined globally because you need to use it in the source files' buffers.

C-x SPC
Set a breakpoint on the line that point is on.

Here are the other special commands provided by GUD. The keys starting with C-c are available only in the GUD interaction buffer. The bindings that start with C-x C-a are available in the GUD buffer and also in source files.

TAB
With GDB, complete a symbol name (gud-gdb-complete-command). This key is available only in the GUD interaction buffer.

C-c C-l
C-x C-a C-l
Display in another window the last line referred to in the GUD buffer (that is, the line indicated in the last location message). This runs the command gud-refresh.

C-c C-s
C-x C-a C-s
Execute a single line of code (gud-step). If the code contains a function call, execution stops after entering the called function.

C-c C-n
C-x C-a C-n
Execute a single line of code, stepping across entire function calls at full speed (gud-next).

C-c C-i
C-x C-a C-i
Execute a single machine instruction (gud-stepi).

C-c C-r
C-x C-a C-r
Continue execution until the next breakpoint, or other event that would normally stop the program (gud-cont).

C-c C-d
C-x C-a C-d
Delete the breakpoint(s) on the current source line, if any (gud-remove). If you use this command in the GUD interaction buffer, it applies to the line where the program last stopped.

C-c C-t
C-x C-a C-t
Set a temporary breakpoint on the current source line, if any. If you use this command in the GUD interaction buffer, it applies to the line where the program last stopped.

The above commands are common to all supported debuggers. If you are using GDB or (some versions of) DBX, these additional commands are available:

C-c <
C-x C-a <
Select the next enclosing stack frame (gud-up). This is equivalent to the `up' command.

C-c >
C-x C-a >
Select the next inner stack frame (gud-down). This is equivalent to the `down' command.

If you are using GDB, two additional keybindings are available:

C-c C-f
C-x C-a C-f
Run the program until the selected stack frame returns (or until it stops for some other reason).

These commands interpret a prefix argument as a repeat count, when that makes sense.

GUD Customization

On startup, GUD runs one of the following hooks: gdb-mode-hook, if you are using GDB; dbx-mode-hook, if you are using DBX; sdb-mode-hook, if you are using SDB; and xdb-mode-hook, if you are using XDB. You can use these hooks to define custom keybindings for the debugger interaction buffer. See section Hooks.

Here is a convenient way to define a command that sends a particular command string to the debugger, and set up a key binding for it in the debugger interaction buffer:

(gud-def function cmdstring binding docstring)

This defines a command named function which sends cmdstring to the debugger process, with documentation string docstring. You can use the command thus defined in any buffer. If binding is non-nil, gud-def also binds the command to C-c binding in the GUD buffer's mode and to C-x C-a binding generally.

The command string cmdstring may contain certain escape sequences that are filled in with varying data at the time function is called:

`%f'
Name of the current source file. If the current buffer is the GUD buffer, then the "current source file" is the file that the program stopped in.

`%l'
Number of the current source line. If the current buffer is the GUD buffer, then the "current source line" is the line that the program stopped in.

`%e'
Text of the C lvalue or function-call expression surrounding point.

`%a'
Text of the hexadecimal address surrounding point.

`%p'
Prefix (numeric) argument of the called function (if any) as a number. (If you don't use `%p' in the command string, the command you define ignores any prefix argument.)

Executing Lisp Expressions

Emacs has several different major modes for Lisp and Scheme. They are the same in terms of editing commands, but differ in the commands for executing Lisp expressions.

Emacs-Lisp mode
The mode for editing source files of programs to run in Emacs Lisp. This mode defines C-M-x to evaluate the current defun. See section Libraries of Lisp Code for Emacs.
Lisp Interaction mode
The mode for an interactive session with Emacs Lisp. It defines LFD to evaluate the sexp before point and insert its value in the buffer. See section Lisp Interaction Buffers.
Lisp mode
The mode for editing source files of programs that run in Lisps other than Emacs Lisp. This mode defines C-M-x to send the current defun to an inferior Lisp process. See section Running an External Lisp.
Inferior Lisp mode
The mode for an interactive session with an inferior Lisp process. This mode combines the special features of Lisp mode and Shell mode (see section Shell Mode).
Scheme mode
Like Lisp mode but for Scheme programs.
Inferior Scheme mode
The mode for an interactive session with an inferior Scheme process.

The editing commands for working with Lisp programs are in fact available globally. See section Editing Programs.

Libraries of Lisp Code for Emacs

Lisp code for Emacs editing commands is stored in files whose names conventionally end in `.el'. This ending tells Emacs to edit them in Emacs-Lisp mode (see section Executing Lisp Expressions).

To execute a file of Emacs Lisp code, use M-x load-file. This command reads a file name using the minibuffer and then executes the contents of that file as Lisp code. It is not necessary to visit the file first; in any case, this command reads the file as found on disk, not text in an Emacs buffer.

Once a file of Lisp code is installed in the Emacs Lisp library directories, users can load it using M-x load-library. Programs can load it by calling load-library, or with load, a more primitive function that is similar but accepts some additional arguments.

M-x load-library differs from M-x load-file in that it searches a sequence of directories and tries three file names in each directory. Suppose your argument is lib; the three names are `lib.elc', `lib.el', and lastly just `lib'. If `lib.elc' exists, it is by convention the result of compiling `lib.el'; it is better to load the compiled file, since it will load and run faster.

If load-library finds that `lib.el' is newer than `lib.elc' file, it prints a warning, because it's likely that somebody made changes to the `.el' file and forgot to recompile it.

Because the argument to load-library is usually not in itself a valid file name, file name completion is not available. Indeed, when using this command, you usually do not know exactly what file name will be used.

The sequence of directories searched by M-x load-library is specified by the variable load-path, a list of strings that are directory names. The default value of the list contains the directory where the Lisp code for Emacs itself is stored. If you have libraries of your own, put them in a single directory and add that directory to load-path. nil in this list stands for the current default directory, but it is probably not a good idea to put nil in the list. If you find yourself wishing that nil were in the list, most likely what you really want to do is use M-x load-file this once.

Often you do not have to give any command to load a library, because the commands defined in the library are set up to autoload that library. Running any of those commands causes load to be called to load the library; this replaces the autoload definitions with the real ones from the library.

If autoloading a file does not finish, either because of an error or because of a C-g quit, all function definitions made by the file are undone automatically. So are any calls to provide. As a consequence, if you use one of the autoloadable commands again, the entire file will be loaded a second time. This prevents problems where the command is no longer autoloading but it works wrong because not all the file was loaded. Function definitions are undone only for autoloading; explicit calls to load do not undo anything if loading is not completed.

Emacs Lisp code can be compiled into byte-code which loads faster, takes up less space when loaded, and executes faster. See section `Byte Compilation' in the Emacs Lisp Reference Manual.

By convention, the compiled code for a library goes in a separate file whose name consists of the library source file with `c' appended. Thus, the compiled code for `foo.el' goes in `foo.elc'. That's why load-library searches for `.elc' files first.

Evaluating Emacs-Lisp Expressions

Lisp programs intended to be run in Emacs should be edited in Emacs-Lisp mode; this happens automatically for file names ending in `.el'. By contrast, Lisp mode itself is used for editing Lisp programs intended for other Lisp systems. To switch to Emacs-Lisp mode explicitly, use the command M-x emacs-lisp-mode.

For testing of Lisp programs to run in Emacs, it is often useful to evaluate part of the program as it is found in the Emacs buffer. For example, after changing the text of a Lisp function definition, evaluating the definition installs the change for future calls to the function. Evaluation of Lisp expressions is also useful in any kind of editing, for invoking noninteractive functions (functions that are not commands).

M-ESC
Read a single Lisp expression in the minibuffer, evaluate it, and print the value in the echo area (eval-expression).
C-x C-e
Evaluate the Lisp expression before point, and print the value in the echo area (eval-last-sexp).
C-M-x
Evaluate the defun containing or after point, and print the value in the echo area (eval-defun).
M-x eval-region
Evaluate all the Lisp expressions in the region.
M-x eval-current-buffer
Evaluate all the Lisp expressions in the buffer.

M-ESC (eval-expression) is the most basic command for evaluating a Lisp expression interactively. It reads the expression using the minibuffer, so you can execute any expression on a buffer regardless of what the buffer contains. When the expression is evaluated, the current buffer is once again the buffer that was current when M-ESC was typed.

M-ESC can easily confuse users who do not understand it, especially on keyboards with autorepeat where it can result from holding down the ESC key for too long. Therefore, eval-expression is normally a disabled command. Attempting to use this command asks for confirmation and gives you the option of enabling it; once you enable the command, confirmation will no longer be required for it. See section Disabling Commands.

In Emacs-Lisp mode, the key C-M-x is bound to the command eval-defun, which parses the defun containing or following point as a Lisp expression and evaluates it. The value is printed in the echo area. This command is convenient for installing in the Lisp environment changes that you have just made in the text of a function definition.

The command C-x C-e (eval-last-sexp) performs a similar job but is available in all major modes, not just Emacs-Lisp mode. It finds the sexp before point, reads it as a Lisp expression, evaluates it, and prints the value in the echo area. It is sometimes useful to type in an expression and then, with point still after it, type C-x C-e.

If C-M-x or C-x C-e is given a numeric argument, it prints the value by insertion into the current buffer at point, rather than in the echo area. The argument value does not matter.

The most general command for evaluating Lisp expressions from a buffer is eval-region. M-x eval-region parses the text of the region as one or more Lisp expressions, evaluating them one by one. M-x eval-current-buffer is similar but evaluates the entire buffer. This is a reasonable way to install the contents of a file of Lisp code that you are just ready to test. After finding and fixing a bug, use C-M-x on each function that you change, to keep the Lisp world in step with the source file.

Lisp Interaction Buffers

The buffer `*scratch*' which is selected when Emacs starts up is provided for evaluating Lisp expressions interactively inside Emacs.

Thus, the way to use the `*scratch*' buffer is to insert Lisp expressions at the end, ending each one with LFD so that it will be evaluated. This command reads the Lisp expression before point, evaluates it, and inserts the value in printed representation before point. The result is a complete typescript of the expressions you have evaluated and their values.

The `*scratch*' buffer's major mode is Lisp Interaction mode, which is the same as Emacs-Lisp mode except for the binding of LFD.

The rationale for this feature is that Emacs must have a buffer when it starts up, but that buffer is not useful for editing files since a new buffer is made for every file that you visit. The Lisp interpreter typescript is the most useful thing I can think of for the initial buffer to do. Type M-x lisp-interaction-mode to put the current buffer in Lisp Interaction mode.

Running an External Lisp

Emacs has facilities for running programs in other Lisp systems. You can run a Lisp process as an inferior of Emacs, and pass expressions to it to be evaluated. You can also pass changed function definitions directly from the Emacs buffers in which you edit the Lisp programs to the inferior Lisp process.

To run an inferior Lisp process, type C-c C-z (run-lisp). This runs the program named lisp, the same program you would run by typing lisp as a shell command, with both input and output going through an Emacs buffer named `*lisp*'. That is to say, any "terminal output" from Lisp will go into the buffer, advancing point, and any "terminal input" for Lisp comes from text in the buffer. (You can change the name of the Lisp executable file by setting the variable inferior-lisp-program.)

To give input to Lisp, go to the end of the buffer and type the input, terminated by RET. The `*lisp*' buffer is in Inferior Lisp mode, which combines the special characteristics of Lisp mode with most of the features of Shell mode (see section Shell Mode).

For the source files of programs to run in external Lisps, use Lisp mode. This mode can be selected with M-x lisp-mode, and is used automatically for files whose names end in `.l', `.lsp', or `.lisp', as most Lisp systems usually expect.

When you edit a function in a Lisp program you are running, the easiest way to send the changed definition to the inferior Lisp process is the key C-M-x. In Lisp mode, this runs the function lisp-eval-defun, which finds the defun around or following point and sends it as input to the Lisp process. (Emacs can send input to any inferior process regardless of what buffer is current.)

Contrast the meanings of C-M-x in Lisp mode (for editing programs to be run in another Lisp system) and Emacs-Lisp mode (for editing Lisp programs to be run in Emacs): in both modes it has the effect of installing the function definition that point is in, but the way of doing so is different according to where the relevant Lisp environment is found. See section Executing Lisp Expressions.

Go to the previous, next section.