There are three ways to investigate a problem in an Emacs Lisp program, depending on what you are doing with the program when the problem appears.
Another useful debugging tool is the dribble file. When a dribble file is open, Emacs copies all keyboard input characters to that file. Afterward, you can examine the file to find out what input was used. See section Terminal Input.
For debugging problems in terminal descriptions, the
open-termscript
function can be useful. See section Terminal Output.
The ordinary Lisp debugger provides the ability to suspend evaluation of a form. While evaluation is suspended (a state that is commonly known as a break), you may examine the run time stack, examine the values of local or global variables, or change those values. Since a break is a recursive edit, all the usual editing facilities of Emacs are available; you can even run programs that will enter the debugger recursively. See section Recursive Editing.
The most important time to enter the debugger is when a Lisp error happens. This allows you to investigate the immediate causes of the error.
However, entry to the debugger is not a normal consequence of an
error. Many commands frequently cause Lisp errors when invoked
inappropriately (such as C-f at the end of the buffer),
and during ordinary editing it would be very inconvenient to enter
the debugger each time this happens. So if you want errors to enter
the debugger, set the variable debug-on-error
to
non-nil
. (The command
toggle-debug-on-error
provides an easy way to do
this.)
debug-on-error
is t
, all
kinds of errors call the debugger (except those listed in
debug-ignored-errors
). If it is nil
, none
call the debugger. The value can also be a list of error conditions that should
call the debugger. For example, if you set it to the list
(void-variable)
, then only errors about a variable
that has no value invoke the debugger.
When this variable is non-nil
, Emacs does not
create an error handler around process filter functions and
sentinels. Therefore, errors in these functions also invoke the
debugger. See section Processes.
debug-on-error
. The normal value of this variable lists several errors that
happen often during editing but rarely result from bugs in Lisp
programs. However, "rarely" is not "never"; if your program fails
with an error that matches this list, you will need to change this
list in order to debug the error. The easiest way is usually to set
debug-ignored-errors
to nil
.
condition-case
never run the debugger, even
if debug-on-error
is non-nil
. In other
words, condition-case
gets a chance to handle the
error before the debugger gets a chance. If you set debug-on-signal
to a
non-nil
value, then the debugger gets the first chance
at every error; an error will invoke the debugger regardless of any
condition-case
, if it fits the criteria specified by
the values of debug-on-error
and
debug-ignored-errors
.
Warning: This variable is strong medicine!
Various parts of Emacs handle errors in the normal course of
affairs, and you may not even realize that errors happen there. If
you set debug-on-signal
to a non-nil
value, those errors will enter the debugger.
Warning: debug-on-signal
has no
effect when debug-on-error
is nil
.
To debug an error that happens during loading of the
`.emacs' file, use the option `--debug-init',
which binds debug-on-error
to t
while
loading `.emacs', and bypasses the
condition-case
which normally catches errors in the
init file.
If your `.emacs' file sets debug-on-error
,
the effect may not last past the end of loading `.emacs'.
(This is an undesirable byproduct of the code that implements the
`--debug-init' command line option.) The best way to
make `.emacs' set debug-on-error
permanently
is with after-init-hook
, like this:
(add-hook 'after-init-hook '(lambda () (setq debug-on-error t)))
When a program loops infinitely and fails to return, your first problem is to stop the loop. On most operating systems, you can do this with C-g, which causes quit.
Ordinary quitting gives no information about why the program was
looping. To get more information, you can set the variable
debug-on-quit
to non-nil
. Quitting with
C-g is not considered an error, and
debug-on-error
has no effect on the handling of
C-g. Likewise, debug-on-quit
has no effect
on errors.
Once you have the debugger running in the middle of the infinite loop, you can proceed from the debugger using the stepping commands. If you step through the entire loop, you will probably get enough information to solve the problem.
quit
is signaled
and not handled. If debug-on-quit
is
non-nil
, then the debugger is called whenever you quit
(that is, type C-g). If debug-on-quit
is
nil
, then the debugger is not called when you quit.
See section Quitting.
To investigate a problem that happens in the middle of a program, one useful technique is to enter the debugger whenever a certain function is called. You can do this to the function in which the problem occurs, and then step through the function, or you can do this to a function called shortly before the problem, step quickly over the call to that function, and then step through its caller.
(debug 'debug)
into the function definition as the first form. Any function defined as Lisp code may be set to break on entry, regardless of whether it is interpreted code or compiled code. If the function is a command, it will enter the debugger when called from Lisp and when called interactively (after the reading of the arguments). You can't debug primitive functions (i.e., those written in C) this way.
When debug-on-entry
is called interactively, it
prompts for function-name in the minibuffer. If the
function is already set up to invoke the debugger on entry,
debug-on-entry
does nothing.
debug-on-entry
always returns
function-name.
Note: if you redefine a function after using
debug-on-entry
on it, the code to enter the debugger
is discarded by the redefinition. In effect, redefining the
function cancels the break-on-entry feature for that function.
(defun fact (n) (if (zerop n) 1 (* n (fact (1- n))))) => fact (debug-on-entry 'fact) => fact (fact 3) ------ Buffer: *Backtrace* ------ Entering: * fact(3) eval-region(4870 4878 t) byte-code("...") eval-last-sexp(nil) (let ...) eval-insert-last-sexp(nil) * call-interactively(eval-insert-last-sexp) ------ Buffer: *Backtrace* ------ (symbol-function 'fact) => (lambda (n) (debug (quote debug)) (if (zerop n) 1 (* n (fact (1- n)))))
debug-on-entry
on function-name.
When called interactively, it prompts for function-name
in the minibuffer. If function-name is nil
or the empty string, it cancels break-on-entry for all functions.
Calling cancel-debug-on-entry
does nothing to a
function which is not currently set up to break on entry. It always
returns function-name.
You can cause the debugger to be called at a certain point in
your program by writing the expression (debug)
at that
point. To do this, visit the source file, insert the text
`(debug)' at the proper place, and type
C-M-x. Warning: if you do this for
temporary debugging purposes, be sure to undo this insertion before
you save the file!
The place where you insert `(debug)' must be a
place where an additional form can be evaluated and its value
ignored. (If the value of (debug)
isn't ignored, it
will alter the execution of the program!) The most common suitable
places are inside a progn
or an implicit
progn
(see section Sequencing).
When the debugger is entered, it displays the previously selected buffer in one window and a buffer named `*Backtrace*' in another window. The backtrace buffer contains one line for each level of Lisp function execution currently going on. At the beginning of this buffer is a message describing the reason that the debugger was invoked (such as the error message and associated data, if it was invoked due to an error).
The backtrace buffer is read-only and uses a special major mode, Debugger mode, in which letters are defined as debugger commands. The usual Emacs editing commands are available; thus, you can switch windows to examine the buffer that was being edited at the time of the error, switch buffers, visit files, or do any other sort of editing. However, the debugger is a recursive editing level (see section Recursive Editing) and it is wise to go back to the backtrace buffer and exit the debugger (with the q command) when you are finished with it. Exiting the debugger gets out of the recursive edit and kills the backtrace buffer.
The backtrace buffer shows you the functions that are executing and their argument values. It also allows you to specify a stack frame by moving point to the line describing that frame. (A stack frame is the place where the Lisp interpreter records information about a particular invocation of a function.) The frame whose line point is on is considered the current frame. Some of the debugger commands operate on the current frame.
The debugger itself must be run byte-compiled, since it makes assumptions about how many stack frames are used for the debugger itself. These assumptions are false if the debugger is running interpreted.
Inside the debugger (in Debugger mode), these special commands are available in addition to the usual cursor motion commands. (Keep in mind that all the usual facilities of Emacs, such as switching windows or buffers, are still available.)
The most important use of debugger commands is for stepping through code, so that you can see how control flows. The debugger can step through the control structures of an interpreted function, but cannot do so in a byte-compiled function. If you would like to step through a byte-compiled function, replace it with an interpreted definition of the same function. (To do this, visit the source for the function and type C-M-x on its definition.)
Here is a list of Debugger mode commands:
debug
and use its return
value. Otherwise, r has the same effect as c,
and the specified return value does not matter. You can't use
r when the debugger was entered due to an error.
Here we describe in full detail the function debug
that is used to invoke the debugger.
The Debugger mode c and r commands exit
the recursive edit; then debug
switches back to the
previous buffer and returns to whatever called debug
.
This is the only way the function debug
can return to
its caller.
The use of the debugger-args is that
debug
displays the rest of its arguments at the top of
the `*Backtrace*' buffer, so that the user can see
them. Except as described below, this is the only way
these arguments are used.
However, certain values for first argument to debug
have a special significance. (Normally, these values are used only
by the internals of Emacs, and not by programmers calling
debug
.) Here is a table of these special values:
lambda
lambda
means debug
was called because of
entry to a function when debug-on-next-call
was
non-nil
. The debugger displays
`Entering:' as a line of text at the top of the
buffer.
debug
debug
as first argument indicates a call to
debug
because of entry to a function that was set to
debug on entry. The debugger displays `Entering:',
just as in the lambda
case. It also marks the stack
frame for that function so that it will invoke the debugger when
exited.
t
t
, this indicates a
call to debug
due to evaluation of a list form when
debug-on-next-call
is non-nil
. The
debugger displays the following as the top line in the buffer:
Beginning evaluation of function call form:
exit
exit
, it indicates the
exit of a stack frame previously marked to invoke the debugger on
exit. The second argument given to debug
in this case
is the value being returned from the frame. The debugger displays
`Return value:' in the top line of the buffer,
followed by the value being returned.
error
error
, the debugger indicates that it is being entered
because an error or quit
was signaled and not handled,
by displaying `Signaling:' followed by the error
signaled and any arguments to signal
. For example,
(let ((debug-on-error t)) (/ 1 0)) ------ Buffer: *Backtrace* ------ Signaling: (arith-error) /(1 0) ... ------ Buffer: *Backtrace* ------If an error was signaled, presumably the variable
debug-on-error
is non-nil
. If
quit
was signaled, then presumably the variable
debug-on-quit
is non-nil
.
nil
nil
as the first of the
debugger-args when you want to enter the debugger
explicitly. The rest of the debugger-args are printed on
the top line of the buffer. You can use this feature to display
messages--for example, to remind yourself of the conditions under
which debug
is called.
This section describes functions and variables used internally by the debugger.
debug
.
The first argument that Lisp hands to the function indicates why
it was called. The convention for arguments is detailed in the
description of debug
.
debug
to fill up the
`*Backtrace*' buffer. It is written in C, since it
must have access to the stack to determine which function calls are
active. The return value is always nil
. In the following example, a Lisp expression calls
backtrace
explicitly. This prints the backtrace to the
stream standard-output
: in this case, to the buffer
`backtrace-output'. Each line of the backtrace
represents one function call. The line shows the values of the
function's arguments if they are all known. If they are still being
computed, the line says so. The arguments of special forms are
elided.
(with-output-to-temp-buffer "backtrace-output" (let ((var 1)) (save-excursion (setq var (eval '(progn (1+ var) (list 'testing (backtrace)))))))) => nil ----------- Buffer: backtrace-output ------------ backtrace() (list ...computing arguments...) (progn ...) eval((progn (1+ var) (list (quote testing) (backtrace)))) (setq ...) (save-excursion ...) (let ...) (with-output-to-temp-buffer ...) eval-region(1973 2142 #<buffer *scratch*>) byte-code("... for eval-print-last-sexp ...") eval-print-last-sexp(nil) * call-interactively(eval-print-last-sexp) ----------- Buffer: backtrace-output ------------
The character `*' indicates a frame whose debug-on-exit flag is set.
nil
, it
says to call the debugger before the next eval
,
apply
or funcall
. Entering the debugger
sets debug-on-next-call
to nil
. The d command in the debugger works by setting this variable.
nil
, this will cause the debugger to be entered
when that frame later exits. Even a nonlocal exit through that
frame will enter the debugger. This function is used only by the debugger.
nil
. The debugger can set this variable to leave
information for future debugger invocations during the same command
invocation. The advantage, for the debugger, of using this variable rather than an ordinary global variable is that the data will never carry over to a subsequent command invocation.
backtrace-frame
is intended for use in Lisp debuggers.
It returns information about what computation is happening in the
stack frame frame-number levels down. If that frame has not evaluated the arguments yet (or is a
special form), the value is (nil function
arg-forms...)
.
If that frame has evaluated its arguments and called its
function already, the value is (t function
arg-values...)
.
In the return value, function is whatever was
supplied as the CAR of the evaluated list, or a lambda
expression in the case of a macro call. If the function has a
&rest
argument, that is represented as the tail of
the list arg-values.
If frame-number is out of range,
backtrace-frame
returns nil
.
Edebug is a source-level debugger for Emacs Lisp programs with which you can:
The first three sections below should tell you enough about Edebug to enable you to use it.
To debug a Lisp program with Edebug, you must first
instrument the Lisp code that you want to debug. A simple
way to do this is to first move point into the definition of a
function or macro and then do C-u C-M-x
(eval-defun
with a prefix argument). See section Instrumenting for Edebug, for
alternative ways to instrument code.
Once a function is instrumented, any call to the function activates Edebug. Activating Edebug may stop execution and let you step through the function, or it may update the display and continue execution while checking for debugging commands, depending on which Edebug execution mode you have selected. The default execution mode is step, which does stop execution. See section Edebug Execution Modes.
Within Edebug, you normally view an Emacs buffer showing the source of the Lisp code you are debugging. This is referred to as the source code buffer. This buffer is temporarily read-only.
An arrow at the left margin indicates the line where the function is executing. Point initially shows where within the line the function is executing, but this ceases to be true if you move point yourself.
If you instrument the definition of fac
(shown
below) and then execute (fac 3)
, here is what you
normally see. Point is at the open-parenthesis before
if
.
(defun fac (n) =>-!-(if (< 0 n) (* n (fac (1- n))) 1))
The places within a function
where Edebug can stop execution are called stop points.
These occur both before and after each subexpression that is a
list, and also after each variable reference. Here we show with
periods the stop points found in the function fac
:
(defun fac (n) .(if .(< 0 n.). .(* n. .(fac (1- n.).).). 1).)
The special commands of Edebug are available in the source code
buffer in addition to the commands of Emacs Lisp mode. For example,
you can type the Edebug command SPC to execute until the
next stop point. If you type SPC once after entry to
fac
, here is the display you will see:
(defun fac (n) =>(if -!-(< 0 n) (* n (fac (1- n))) 1))
When Edebug stops execution after an expression, it displays the expression's value in the echo area.
Other frequently used commands are b to set a breakpoint at a stop point, g to execute until a breakpoint is reached, and q to exit Edebug and return to the top-level command loop. Type ? to display a list of all Edebug commands.
In order to use Edebug to debug Lisp code, you must first instrument the code. Instrumenting code inserts additional code into it, to invoke Edebug at the proper places.
Once you have loaded Edebug, the command
C-M-x (eval-defun
) is redefined so that
when invoked with a prefix argument on a definition, it instruments
the definition before evaluating it. (The source code itself is not
modified.) If the variable edebug-all-defs
is
non-nil
, that inverts the meaning of the prefix
argument: then C-M-x instruments the definition
unless it has a prefix argument. The default value of
edebug-all-defs
is nil
. The command
M-x edebug-all-defs toggles the value of the variable
edebug-all-defs
.
If edebug-all-defs
is
non-nil
, then the commands eval-region
,
eval-current-buffer
, and eval-buffer
also
instrument any definitions they evaluate. Similarly,
edebug-all-forms
controls whether
eval-region
should instrument any form, even
non-defining forms. This doesn't apply to loading or evaluations in
the minibuffer. The command M-x edebug-all-forms toggles
this option.
Another command, M-x
edebug-eval-top-level-form, is available to instrument any
top-level form regardless of the values of
edebug-all-defs
and edebug-all-forms
.
While Edebug is active, the command I
(edebug-instrument-callee
) instruments the definition
of the function or macro called by the list form after point, if is
not already instrumented. This is possible only if Edebug knows
where to find the source for that function; after loading Edebug,
eval-region
records the position of every definition
it evaluates, even if not instrumenting it. See also the
i command (see section Jumping), which steps into the call
after instrumenting the function.
Edebug knows how to instrument all the standard
special forms, interactive
forms with an expression
argument, anonymous lambda expressions, and other defining forms.
Edebug cannot know what a user-defined macro will do with the
arguments of a macro call, so you must tell it; see section Instrumenting Macro Calls, for
details.
When Edebug is about to instrument code for the first time in a
session, it runs the hook edebug-setup-hook
, then sets
it to nil
. You can use this to arrange to load Edebug
specifications (see section Instrumenting Macro Calls) associated
with a package you are using, but actually load them only if you
use Edebug.
To remove instrumentation from
a definition, simply re-evaluate its definition in a way that does
not instrument. There are two ways of evaluating forms that never
instrument them: from a file with load
, and from the
minibuffer with eval-expression
(M-:).
If Edebug detects a syntax error while instrumenting, it leaves
point at the erroneous code and signals an
invalid-read-syntax
error.
See section Evaluation, for other evaluation functions available inside of Edebug.
Edebug supports several execution modes for running the program you are debugging. We call these alternatives Edebug execution modes; do not confuse them with major or minor modes. The current Edebug execution mode determines how far Edebug continues execution before stopping--whether it stops at each stop point, or continues to the next breakpoint, for example--and how much Edebug displays the progress of the evaluation before it stops.
Normally, you specify the Edebug execution mode by typing a command to continue the program in a certain mode. Here is a table of these commands. All except for S resume execution of the program, at least for a certain distance.
edebug-stop
).
edebug-step-mode
).
edebug-next-mode
). Also see
edebug-forward-sexp
in section Miscellaneous Edebug Commands.
edebug-trace-mode
).
edebug-Trace-fast-mode
).
edebug-go-mode
). See section Breakpoints.
edebug-continue-mode
).
edebug-Continue-fast-mode
).
edebug-Go-nonstop-mode
). You can still stop the
program by typing S, or any editing command.
In general, the execution modes earlier in the above list run the program more slowly or stop sooner than the modes later in the list.
While executing or tracing, you can interrupt the execution by typing any Edebug command. Edebug stops the program at the next stop point and then executes the command you typed. For example, typing t during execution switches to trace mode at the next stop point. You can use S to stop execution without doing anything else.
If your function happens to read input, a character you type intending to interrupt execution may be read by the function instead. You can avoid such unintended results by paying attention to when your program wants input.
Keyboard macros containing the
commands in this section do not completely work: exiting from
Edebug, to resume the program, loses track of the keyboard macro.
This is not easy to fix. Also, defining or executing a keyboard
macro outside of Edebug does not affect commands inside Edebug.
This is usually an advantage. But see the
edebug-continue-kbd-macro
option (see section Edebug Options).
When you enter a new Edebug level, the initial execution mode
comes from the value of the variable
edebug-initial-mode
. By default, this specifies step
mode. Note that you may reenter the same Edebug level several times
if, for example, an instrumented function is called several times
from one command.
The commands described in this section execute until they reach a specified location. All except i make a temporary breakpoint to establish the place to stop, then switch to go mode. Any other breakpoint reached before the intended stop point will also stop execution. See section Breakpoints, for the details on breakpoints.
These commands may fail to work as expected in case of nonlocal exit, because a nonlocal exit can bypass the temporary breakpoint where you expected the program to stop.
edebug-goto-here
).
edebug-forward-sexp
).
The h command proceeds to the stop point near the current location of point, using a temporary breakpoint. See section Breakpoints, for more information about breakpoints.
The f command runs the program forward over one expression. More precisely, it sets a temporary breakpoint at the position that C-M-f would reach, then executes in go mode so that the program will stop at breakpoints.
With a prefix argument n, the temporary breakpoint is placed n sexps beyond point. If the containing list ends before n more elements, then the place to stop is after the containing expression.
Be careful that the position C-M-f finds is a place
that the program will really get to; this may not be true in a
cond
, for example.
The f command does forward-sexp
starting
at point, rather than at the stop point, for flexibility. If you
want to execute one expression from the current stop
point, type w first, to move point there, and then
type f.
The o command continues "out of" an expression. It places a temporary breakpoint at the end of the sexp containing point. If the containing sexp is a function definition itself, o continues until just before the last sexp in the definition. If that is where you are now, it returns from the function and then stops. In other words, this command does not exit the currently executing function unless you are positioned after the last sexp.
The i command steps into the function or macro called by the list form after point, and stops at its first stop point. Note that the form need not be the one about to be evaluated. But if the form is a function call about to be evaluated, remember to use this command before any of the arguments are evaluated, since otherwise it will be too late.
The i command instruments the function or macro it's supposed to step into, if it isn't instrumented already. This is convenient, but keep in mind that the function or macro remains instrumented unless you explicitly arrange to deinstrument it.
Some miscellaneous Edebug commands are described here.
edebug-help
).
abort-recursive-edit
).
top-level
). This exits all recursive editing levels,
including all levels of Edebug activity. However, instrumented code
protected with unwind-protect
or
condition-case
forms may resume debugging.
top-level-nonstop
).
edebug-previous-result
).
edebug-backtrace
). You cannot use debugger
commands in the backtrace buffer in Edebug as you would in the
standard debugger. The backtrace buffer is killed automatically
when you continue execution.
From the Edebug recursive edit, you may invoke commands that activate Edebug again recursively. Any time Edebug is active, you can quit to the top level with q or abort one recursive edit level with C-]. You can display a backtrace of all the pending evaluations with d.
Edebug's step mode stops execution at the next stop point reached. There are three other ways to stop Edebug execution once it has started: breakpoints, the global break condition, and source breakpoints.
While using Edebug, you can specify breakpoints in the program you are testing: points where execution should stop. You can set a breakpoint at any stop point, as defined in section Using Edebug. For setting and unsetting breakpoints, the stop point that is affected is the first one at or after point in the source code buffer. Here are the Edebug commands for breakpoints:
edebug-set-breakpoint
). If you use a prefix argument,
the breakpoint is temporary (it turns off the first time it stops
the program).
edebug-unset-breakpoint
).
nil
value
(edebug-set-conditional-breakpoint
). With a prefix
argument, the breakpoint is temporary.
edebug-next-breakpoint
).
While in Edebug, you can set a breakpoint with b and unset one with u. First move point to the Edebug stop point of your choice, then type b or u to set or unset a breakpoint there. Unsetting a breakpoint where none has been set has no effect.
Re-evaluating or reinstrumenting a definition forgets all its breakpoints.
A conditional breakpoint tests a condition each time
the program gets there. Any errors that occur as a result of
evaluating the condition are ignored, as if the result were
nil
. To set a conditional breakpoint, use
x, and specify the condition expression in the
minibuffer. Setting a conditional breakpoint at a stop point that
has a previously established conditional breakpoint puts the
previous condition expression in the minibuffer so you can edit
it.
You can make a conditional or unconditional breakpoint temporary by using a prefix argument with the command to set the breakpoint. When a temporary breakpoint stops the program, it is automatically unset.
Edebug always stops or pauses at a breakpoint except when the Edebug mode is Go-nonstop. In that mode, it ignores breakpoints entirely.
To find out where your breakpoints are, use the B command, which moves point to the next breakpoint following point, within the same function, or to the first breakpoint if there are no following breakpoints. This command does not continue execution--it just moves point in the buffer.
A global break condition stops
execution when a specified condition is satisfied, no matter where
that may occur. Edebug evaluates the global break condition at
every stop point. If it evaluates to a non-nil
value,
then execution stops or pauses depending on the execution mode, as
if a breakpoint had been hit. If evaluating the condition gets an
error, execution does not stop.
The condition expression is
stored in edebug-global-break-condition
. You can
specify a new expression using the X command
(edebug-set-global-break-condition
).
The global break condition is the simplest way to find where in
your code some event occurs, but it makes code run much more
slowly. So you should reset the condition to nil
when
not using it.
All breakpoints in a definition are forgotten
each time you reinstrument it. To make a breakpoint that won't be
forgotten, you can write a source breakpoint, which is
simply a call to the function edebug
in your source
code. You can, of course, make such a call conditional. For
example, in the fac
function, insert the first line as
shown below to stop when the argument reaches zero:
(defun fac (n) (if (= n 0) (edebug)) (if (< 0 n) (* n (fac (1- n))) 1))
When the fac
definition is instrumented and the
function is called, the call to edebug
acts as a
breakpoint. Depending on the execution mode, Edebug stops or pauses
there.
If no instrumented code is being executed when
edebug
is called, that function calls
debug
.
Emacs normally displays an error message when an error is
signaled and not handled with condition-case
. While
Edebug is active and executing instrumented code, it normally
responds to all unhandled errors. You can customize this with the
options edebug-on-error
and
edebug-on-quit
; see section Edebug Options.
When Edebug responds to an error, it shows the last stop point encountered before the error. This may be the location of a call to a function which was not instrumented, within which the error actually occurred. For an unbound variable error, the last known stop point might be quite distant from the offending variable reference. In that case you might want to display a full backtrace (see section Miscellaneous Edebug Commands).
If you change debug-on-error
or
debug-on-quit
while Edebug is active, these changes
will be forgotten when Edebug becomes inactive. Furthermore, during
Edebug's recursive edit, these variables are bound to the values
they had outside of Edebug.
These Edebug commands let you view aspects of the buffer and window status as they were before entry to Edebug. The outside window configuration is the collection of windows and contents that were in effect outside of Edebug.
edebug-view-outside
).
edebug-bounce-point
). With a
prefix argument n, pause for n seconds
instead.
edebug-where
). If you use this command in a
different window displaying the same buffer, that window will be
used instead to display the current definition in the future.
edebug-toggle-save-windows
). With a
prefix argument, W
only toggles saving and restoring
of the selected window. To specify a window that is not displaying
the source code buffer, you must use C-x X W from the
global keymap.
You can view the outside window configuration with v or just bounce to the point in the current buffer with p, even if it is not normally displayed. After moving point, you may wish to jump back to the stop point with w from a source code buffer.
Each time you use W to turn saving off, Edebug forgets the saved outside window configuration--so that even if you turn saving back on, the current window configuration remains unchanged when you next exit Edebug (by continuing the program). However, the automatic redisplay of `*edebug*' and `*edebug-trace*' may conflict with the buffers you wish to see unless you have enough windows open.
While within Edebug, you can evaluate expressions "as if" Edebug were not running. Edebug tries to be invisible to the expression's evaluation and printing. Evaluation of expressions that cause side effects will work as expected except for things that Edebug explicitly saves and restores. See section The Outside Context, for details on this process.
edebug-eval-expression
). That is, Edebug tries
to minimize its interference with the evaluation.
edebug-eval-last-sexp
).
Edebug supports evaluation of
expressions containing references to lexically bound symbols
created by the following constructs in `cl.el' (version
2.03 or later): lexical-let
, macrolet
,
and symbol-macrolet
.
You can use the evaluation list buffer, called `*edebug*', to evaluate expressions interactively. You can also set up the evaluation list of expressions to be evaluated automatically each time Edebug updates the display.
edebug-visit-eval-list
).
In the `*edebug*' buffer you can use the commands of Lisp Interaction mode (see section `Lisp Interaction' in The GNU Emacs Manual) as well as these special commands:
edebug-eval-print-last-sexp
).
edebug-eval-last-sexp
).
edebug-update-eval-list
).
edebug-delete-eval-item
).
edebug-where
).
You can evaluate expressions in the evaluation list window with C-j or C-x C-e, just as you would in `*scratch*'; but they are evaluated in the context outside of Edebug.
The expressions you enter interactively (and their results) are lost when you continue execution; but you can set up an evaluation list consisting of expressions to be evaluated each time execution stops.
To do this, write one or more evaluation list groups in the evaluation list buffer. An evaluation list group consists of one or more Lisp expressions. Groups are separated by comment lines.
The command C-c C-u
(edebug-update-eval-list
) rebuilds the evaluation
list, scanning the buffer and using the first expression of each
group. (The idea is that the second expression of the group is the
value previously computed and displayed.)
Each entry to Edebug redisplays the evaluation list by inserting each expression in the buffer, followed by its current value. It also inserts comment lines so that each expression becomes its own group. Thus, if you type C-c C-u again without changing the buffer text, the evaluation list is effectively unchanged.
If an error occurs during an evaluation from the evaluation list, the error message is displayed in a string as if it were the result. Therefore, expressions that use variables not currently valid do not interrupt your debugging.
Here is an example of what the evaluation list window looks like after several expressions have been added to it:
(current-buffer) #<buffer *scratch*> ;--------------------------------------------------------------- (selected-window) #<window 16 on *scratch*> ;--------------------------------------------------------------- (point) 196 ;--------------------------------------------------------------- bad-var "Symbol's value as variable is void: bad-var" ;--------------------------------------------------------------- (recursion-depth) 0 ;--------------------------------------------------------------- this-command eval-last-sexp ;---------------------------------------------------------------
To delete a group, move point into it and type C-c C-d, or simply delete the text for the group and update the evaluation list with C-c C-u. To add a new expression to the evaluation list, insert the expression at a suitable place, and insert a new comment line. (You need not insert dashes in the comment line--its contents don't matter.) Then type C-c C-u.
After selecting `*edebug*', you can return to the source code buffer with C-c C-w. The `*edebug*' buffer is killed when you continue execution, and recreated next time it is needed.
If an expression in your program produces a value containing circular list structure, you may get an error when Edebug attempts to print it.
One way to cope with circular structure is to set
print-length
or print-level
to truncate
the printing. Edebug does this for you; it binds
print-length
and print-level
to 50 if
they were nil
. (Actually, the variables
edebug-print-length
and
edebug-print-level
specify the values to use within
Edebug.) See section Variables
Affecting Output.
nil
, bind
print-length
to this while printing results in Edebug.
The default value is 50
.
nil
, bind
print-level
to this while printing results in Edebug.
The default value is 50
.
You can also print circular structures and structures that share elements more informatively by using the `cust-print' package.
To load `cust-print' and activate custom printing only for Edebug, simply use the command M-x edebug-install-custom-print. To restore the standard print functions, use M-x edebug-uninstall-custom-print.
Here is an example of code that creates a circular structure:
(setq a '(x y)) (setcar a a)
Custom printing prints this as `Result: #1=(#1# y)'. The `#1=' notation labels the structure that follows it with the label `1', and the `#1#' notation references the previously labeled structure. This notation is used for any shared elements of lists or vectors.
nil
, bind
print-circle
to this while printing results in Edebug.
The default value is nil
.
Other programs can also use custom printing; see `cust-print.el' for details.
Edebug can record an execution trace, storing it in a buffer
named `*edebug-trace*'. This is a log of function
calls and returns, showing the function names and their arguments
and values. To enable trace recording, set
edebug-trace
to a non-nil
value.
Making a trace buffer is not the same thing as using trace execution mode (see section Edebug Execution Modes).
When trace recording is enabled, each function entry and exit adds lines to the trace buffer. A function entry record looks like `::::{' followed by the function name and argument values. A function exit record looks like `::::}' followed by the function name and result of the function.
The number of `:'s in an entry shows its recursion depth. You can use the braces in the trace buffer to find the matching beginning or end of function calls.
You can customize trace recording for function
entry and exit by redefining the functions
edebug-print-trace-before
and
edebug-print-trace-after
.
edebug-tracing
returns the value of the last form in
body.
(apply 'format
format-string format-args)
. It also
appends a newline to separate entries.
edebug-tracing
and edebug-trace
insert
lines in the trace buffer whenever they are called, even if Edebug
is not active. Adding text to the trace buffer also scrolls its
window to show the last lines inserted.
Edebug provides rudimentary coverage testing and display of execution frequency.
Coverage testing works by comparing the result of each expression with the previous result; each form in the program is considered "covered" if it has returned two different values since you began testing coverage in the current Emacs session. Thus, to do coverage testing on your program, execute it under various conditions and note whether it behaves correctly; Edebug will tell you when you have tried enough different conditions that each form has returned two different values.
Coverage testing makes execution slower, so it is only done if
edebug-test-coverage
is non-nil
.
Frequency counting is performed for all execution of an
instrumented function, even if the execution mode is Go-nonstop,
and regardless of whether coverage testing is enabled.
Use M-x edebug-display-freq-count to display both the coverage information and the frequency counts for a definition.
The frequency counts appear as comment lines after each line of
code, and you can undo all insertions with one undo
command. The counts appear under the `(' before an
expression or the `)' after an expression, or on the
last character of a variable. To simplify the display, a count is
not shown if it is equal to the count of an earlier expression on
the same line.
The character `=' following the count for an expression says that the expression has returned the same value each time it was evaluated. In other words, it is not yet "covered" for coverage testing purposes.
To clear the frequency count and coverage data for a definition,
simply reinstrument it with eval-defun
.
For example, after evaluating (fac 5)
with a source
breakpoint, and setting edebug-test-coverage
to
t
, when the breakpoint is reached, the frequency data
looks like this:
(defun fac (n) (if (= n 0) (edebug)) ;#6 1 0 =5 (if (< 0 n) ;#5 = (* n (fac (1- n))) ;# 5 0 1)) ;# 0
The comment lines show that fac
was called 6 times.
The first if
statement returned 5 times with the same
result each time; the same is true of the condition on the second
if
. The recursive call of fac
did not
return at all.
Edebug tries to be transparent to the program you are debugging, but it does not succeed completely. Edebug also tries to be transparent when you evaluate expressions with e or with the evaluation list buffer, by temporarily restoring the outside context. This section explains precisely what context Edebug restores, and how Edebug fails to be completely transparent.
Whenever Edebug is entered, it needs to save and restore certain data before even deciding whether to make trace information or stop the program.
max-lisp-eval-depth
and
max-specpdl-size
are both incremented once to reduce
Edebug's impact on the stack. You could, however, still run out of
stack space when using Edebug.
executing-macro
is bound to
edebug-continue-kbd-macro
.
When Edebug needs to display something (e.g., in trace mode), it saves the current window configuration from "outside" Edebug (see section Window Configurations). When you exit Edebug (by continuing the program), it restores the previous window configuration.
Emacs redisplays only when it pauses. Usually, when you continue execution, the program comes back into Edebug at a breakpoint or after stepping without pausing or reading input in between. In such cases, Emacs never gets a chance to redisplay the "outside" configuration. What you see is the same window configuration as the last time Edebug was active, with no interruption.
Entry to Edebug for displaying something also saves and restores the following data, but some of these are deliberately not restored if an error or quit signal occurs.
edebug-save-windows
is non-nil
(see
section Edebug Display Update). The
window configuration is not restored on error or quit, but the
outside selected window is reselected even on error or
quit in case a save-excursion
is active. If the value
of edebug-save-windows
is a list, only the listed
windows are saved and restored. The window start and horizontal
scrolling of the source code buffer are not restored, however, so
that the display remains coherent within Edebug.
edebug-save-displayed-buffer-points
is
non-nil
.
overlay-arrow-position
and
overlay-arrow-string
are saved and restored. So you
can safely invoke Edebug from the recursive edit elsewhere in the
same buffer.
cursor-in-echo-area
is locally bound to
nil
so that the cursor shows up in the window.
When Edebug is entered and actually reads commands from the user, it saves (and later restores) these additional data:
last-command
, this-command
,
last-command-char
, last-input-char
,
last-input-event
, last-command-event
,
last-event-frame
, last-nonmenu-event
, and
track-mouse
. Commands used within Edebug do not affect
these variables outside of Edebug. The key sequence returned by
this-command-keys
is changed by executing commands
within Edebug and there is no way to reset the key sequence from
Lisp. Edebug cannot save and restore the value of
unread-command-events
. Entering Edebug while this
variable has a nontrivial value can interfere with execution of the
program you are debugging.
command-history
. In rare cases this can alter
execution.
standard-output
and standard-input
are bound to nil
by the recursive-edit
,
but Edebug temporarily restores them during evaluations.
defining-kbd-macro
is bound to
edebug-continue-kbd-macro
.
When Edebug instruments an expression that calls a Lisp macro, it needs additional information about the macro to do the job properly. This is because there is no a-priori way to tell which subexpressions of the macro call are forms to be evaluated. (Evaluation may occur explicitly in the macro body, or when the resulting expansion is evaluated, or any time later.)
Therefore, you must define an Edebug specification for each
macro that Edebug will encounter, to explain the format of calls to
that macro. To do this, use def-edebug-spec
.
The macro argument can actually be any symbol, not just a macro name.
Here is a simple example that defines the specification for the
for
example macro (see section Evaluating Macro Arguments
Repeatedly), followed by an alternative, equivalent
specification.
(def-edebug-spec for (symbolp "from" form "to" form "do" &rest form)) (def-edebug-spec for (symbolp ['from form] ['to form] ['do body]))
Here is a table of the possibilities for specification and how each directs processing of arguments.
t
0
A specification list
is required for an Edebug specification if some arguments of a
macro call are evaluated while others are not. Some elements in a
specification list match one or more arguments, but others modify
the processing of all following elements. The latter, called
specification keywords, are symbols beginning with
`&' (such as &optional
).
A specification list may contain sublists which match arguments that are themselves lists, or it may contain vectors used for grouping. Sublists and groups thus subdivide the specification list into a hierarchy of levels. Specification keywords apply only to the remainder of the sublist or group they are contained in.
When a specification list involves alternatives or repetition, matching it against an actual macro call may require backtracking. See section Backtracking in Specifications, for more details.
Edebug specifications provide the power of regular expression matching, plus some context-free grammar constructs: the matching of sublists with balanced parentheses, recursive processing of forms, and recursion via indirect specifications.
Here's a table of the possible elements of a specification list, with their meanings:
sexp
form
place
setf
construct.
body
&rest form
. See
&rest
below.
function-form
quote
rather than function
, since it instruments the body of
the lambda expression either way.
lambda-expr
&optional
[&optional specs...]
. To specify that
several elements must all match or none, use &optional
[specs...]
. See the defun
example
below.
&rest
[&rest
specs...]
. To specify several elements that must
all match on every repetition, use &rest
[specs...]
.
&or
&or
specification
fails. Each list element following &or
is a single
alternative. To group two or more list elements as a single
alternative, enclose them in [...]
.
¬
&or
, but if any of them match, the specification
fails. If none of them match, nothing is matched, but the
¬
specification succeeds.
&define
&define
keyword should be the
first element in a list specification.
nil
gate
let
example below.
other-symbol
def-edebug-spec
just as for macros. See the
defun
example below. Otherwise, the symbol should be a
predicate. The predicate is called with the argument and the
specification fails if the predicate returns nil
. In
either case, that argument is not instrumented. Some suitable
predicates include symbolp
, integerp
,
stringp
, vectorp
, and
atom
.
[elements...]
"string"
'symbol
, where the name of
symbol is the string, but the string form is
preferred.
(vector elements...)
(elements...)
(spec .
[(more specs...)])
) whose elements match the non-dotted list
arguments. This is useful in recursive specifications such as in
the backquote example below. Also see the description of a
nil
specification above for terminating such
recursion. Note that a sublist specification written as
(specs . nil)
is equivalent to (specs)
,
and (specs . (sublist-elements...))
is equivalent to
(specs sublist-elements...)
.
Here is a list of additional specifications that may appear only
after &define
. See the defun
example
below.
name
:name
:name
should be a symbol; it is used as an
additional name component for the definition. You can use this to
add a unique, static component to the name of the definition. It
may be used more than once.
arg
lambda-list
def-body
body
, described above, but a definition body must be
instrumented with a different Edebug call that looks up information
associated with the definition. Use def-body
for the
highest level list of forms within the definition.
def-form
def-body
, except use this to match a
single form rather than a list of forms. As a special case,
def-form
also means that tracing information is not
output when the form is executed. See the interactive
example below.
If a specification fails to match at some point,
this does not necessarily mean a syntax error will be signaled;
instead, backtracking will take place until all
alternatives have been exhausted. Eventually every element of the
argument list must be matched by some element in the specification,
and every required element in the specification must match some
argument. When a syntax error is detected, it might not be reported
until much later after higher-level alternatives have been
exhausted, and with the point positioned further from the real
error. But if backtracking is disabled when an error occurs, it can
be reported immediately. Note that backtracking is also reenabled
automatically in several situations; it is reenabled when a new
alternative is established by &optional
,
&rest
, or &or
, or at the start of
processing a sublist, group, or indirect specification. The effect
of enabling or disabling backtracking is limited to the remainder
of the level currently being processed and lower levels.
Backtracking is disabled while matching any of the form
specifications (that is, form
, body
,
def-form
, and def-body
). These
specifications will match any form so any error must be in the form
itself rather than at a higher level.
Backtracking is also disabled after successfully matching a
quoted symbol or string specification, since this usually indicates
a recognized construct. But if you have a set of alternative
constructs that all begin with the same symbol, you can usually
work around this constraint by factoring the symbol out of the
alternatives, e.g., ["foo" &or [first case] [second case]
...]
.
Most needs are satisfied by these two ways that bactracking is
automatically disabled, but occasionally it is useful to explicitly
disable backtracking by using the gate
specification.
This is useful when you know that no higher alternatives could
apply. See the example of the let
specification.
It may be easier to understand Edebug specifications by studying the examples provided here.
A let
special form has a sequence of bindings and a
body. Each of the bindings is either a symbol or a sublist with a
symbol and optional expression. In the specification below, notice
the gate
inside of the sublist to prevent backtracking
once a sublist is found.
(def-edebug-spec let ((&rest &or symbolp (gate symbolp &optional form)) body))
Edebug uses the following specifications for defun
and defmacro
and the associated argument list and
interactive
specifications. It is necessary to handle
interactive forms specially since an expression argument it is
actually evaluated outside of the function body.
(def-edebug-spec defmacro defun) ; Indirect ref todefun
spec. (def-edebug-spec defun (&define name lambda-list [&optional stringp] ; Match the doc string, if present. [&optional ("interactive" interactive)] def-body)) (def-edebug-spec lambda-list (([&rest arg] [&optional ["&optional" arg &rest arg]] &optional ["&rest" arg] ))) (def-edebug-spec interactive (&optional &or stringp def-form)) ; Notice:def-form
The specification for backquote below illustrates how to match
dotted lists and use nil
to terminate recursion. It
also illustrates how components of a vector may be matched. (The
actual specification defined by Edebug does not support dotted
lists because doing so causes very deep recursion that could
fail.)
(def-edebug-spec ` (backquote-form)) ; Alias just for clarity. (def-edebug-spec backquote-form (&or ([&or "," ",@"] &or ("quote" backquote-form) form) (backquote-form . [&or nil backquote-form]) (vector &rest backquote-form) sexp))
These options affect the behavior of Edebug:
edebug-setup-hook
is reset to nil
. You could use this to load up Edebug
specifications associated with a package you are using but only
when you also use Edebug. See section Instrumenting for Edebug.
nil
, normal evaluation of defining forms such as
defun
and defmacro
instruments them for
Edebug. This applies to eval-defun
,
eval-region
, eval-buffer
, and
eval-current-buffer
. Use the command M-x edebug-all-defs to toggle the value of this option. See section Instrumenting for Edebug.
nil
, the commands eval-defun
,
eval-region
, eval-buffer
, and
eval-current-buffer
instrument all forms, even those
that don't define anything. This doesn't apply to loading or
evaluations in the minibuffer. Use the command M-x edebug-all-forms to toggle the value of this option. See section Instrumenting for Edebug.
nil
, Edebug saves and restores the window
configuration. That takes some time, so if your program does not
care what happens to the window configurations, it is better to set
this variable to nil
. If the value is a list, only the listed windows are saved and restored.
You can use the W command in Edebug to change this variable interactively. See section Edebug Display Update.
nil
, Edebug saves and restores point in all
displayed buffers. Saving and restoring point in other buffers is necessary if you are debugging code that changes the point of a buffer which is displayed in a non-selected window. If Edebug or the user then selects the window, point in that buffer will move to the window's value of point.
Saving and restoring point in all buffers is expensive, since it requires selecting each window twice, so enable this only if you need it. See section Edebug Display Update.
nil
, it specifies the initial execution mode for
Edebug when it is first activated. Possible values are
step
, next
, go
,
Go-nonstop
, trace
,
Trace-fast
, continue
, and
Continue-fast
. The default value is step
. See section Edebug Execution Modes.
nil
means
display a trace of function entry and exit. Tracing output is
displayed in a buffer named `*edebug-trace*', one
function entry or exit per line, indented by the recursion level.
The default value is nil
.
Also see edebug-tracing
, in section Trace Buffer.
nil
,
Edebug tests coverage of all expressions debugged. See section Coverage Testing.
nil
,
continue defining or executing any keyboard macro that is executing
outside of Edebug. Use this with caution since it is not debugged.
See section Edebug Execution
Modes.
debug-on-error
to this value, if
debug-on-error
was previously nil
. See
section Trapping Errors.
debug-on-quit
to this value, if
debug-on-quit
was previously nil
. See
section Trapping Errors.
If you change the values of edebug-on-error
or
edebug-on-quit
while Edebug is active, their values
won't be used until the next time Edebug is invoked via a
new command.
nil
, an
expression to test for at every stop point. If the result is
non-nil, then break. Errors are ignored. See section Global Break Condition.
The Lisp reader reports invalid syntax, but cannot say where the real problem is. For example, the error "End of file during parsing" in evaluating an expression indicates an excess of open parentheses (or square brackets). The reader detects this imbalance at the end of the file, but it cannot figure out where the close parenthesis should have been. Likewise, "Invalid read syntax: ")"" indicates an excess close parenthesis or missing open parenthesis, but does not say where the missing parenthesis belongs. How, then, to find what to change?
If the problem is not simply an imbalance of parentheses, a useful technique is to try C-M-e at the beginning of each defun, and see if it goes to the place where that defun appears to end. If it does not, there is a problem in that defun.
However, unmatched parentheses are the most common syntax errors in Lisp, and we can give further advice for those cases. (In addition, just moving point through the code with Show Paren mode enabled might find the mismatch.)
The first step is to find the defun that is unbalanced. If there
is an excess open parenthesis, the way to do this is to insert a
close parenthesis at the end of the file and type C-M-b
(backward-sexp
). This will move you to the beginning
of the defun that is unbalanced. (Then type C-SPC C-_
C-u C-SPC to set the mark there, undo the insertion of
the close parenthesis, and finally return to the mark.)
The next step is to determine precisely what is wrong. There is no way to be sure of this except by studying the program, but often the existing indentation is a clue to where the parentheses should have been. The easiest way to use this clue is to reindent with C-M-q and see what moves. But don't do this yet! Keep reading, first.
Before you do this, make sure the defun has enough close parentheses. Otherwise, C-M-q will get an error, or will reindent all the rest of the file until the end. So move to the end of the defun and insert a close parenthesis there. Don't use C-M-e to move there, since that too will fail to work until the defun is balanced.
Now you can go to the beginning of the defun and type C-M-q. Usually all the lines from a certain point to the end of the function will shift to the right. There is probably a missing close parenthesis, or a superfluous open parenthesis, near that point. (However, don't assume this is true; study the code to make sure.) Once you have found the discrepancy, undo the C-M-q with C-_, since the old indentation is probably appropriate to the intended parentheses.
After you think you have fixed the problem, use C-M-q again. If the old indentation actually fit the intended nesting of parentheses, and you have put back those parentheses, C-M-q should not change anything.
To deal with an excess close parenthesis, first insert an open parenthesis at the beginning of the file, back up over it, and type C-M-f to find the end of the unbalanced defun. (Then type C-SPC C-_ C-u C-SPC to set the mark there, undo the insertion of the open parenthesis, and finally return to the mark.)
Then find the actual matching close parenthesis by typing C-M-f at the beginning of that defun. This will leave you somewhere short of the place where the defun ought to end. It is possible that you will find a spurious close parenthesis in that vicinity.
If you don't see a problem at that point, the next thing to do is to type C-M-q at the beginning of the defun. A range of lines will probably shift left; if so, the missing open parenthesis or spurious close parenthesis is probably near the first of those lines. (However, don't assume this is true; study the code to make sure.) Once you have found the discrepancy, undo the C-M-q with C-_, since the old indentation is probably appropriate to the intended parentheses.
After you think you have fixed the problem, use C-M-q again. If the old indentation actually fit the intended nesting of parentheses, and you have put back those parentheses, C-M-q should not change anything.
When an error happens during byte compilation, it is normally due to invalid syntax in the program you are compiling. The compiler prints a suitable error message in the `*Compile-Log*' buffer, and then stops. The message may state a function name in which the error was found, or it may not. Either way, here is how to find out where in the file the error occurred.
What you should do is switch to the buffer ` *Compiler Input*'. (Note that the buffer name starts with a space, so it does not show up in M-x list-buffers.) This buffer contains the program being compiled, and point shows how far the byte compiler was able to read.
If the error was due to invalid Lisp syntax, point shows exactly where the invalid syntax was detected. The cause of the error is not necessarily near by! Use the techniques in the previous section to find the error.
If the error was detected while compiling a form that had been read successfully, then point is located at the end of the form. In this case, this technique can't localize the error precisely, but can still show you which function to check.