In the terminology of operating systems, a process is a space in which a program can execute. Emacs runs in a process. Emacs Lisp programs can invoke other programs in processes of their own. These are called subprocesses or child processes of the Emacs process, which is their parent process.
A subprocess of Emacs may be synchronous or asynchronous, depending on how it is created. When you create a synchronous subprocess, the Lisp program waits for the subprocess to terminate before continuing execution. When you create an asynchronous subprocess, it can run in parallel with the Lisp program. This kind of subprocess is represented within Emacs by a Lisp object which is also called a "process". Lisp programs can use this object to communicate with the subprocess or to control it. For example, you can send signals, obtain status information, receive output from the process, or send input to it.
t if object is a process,
nil otherwise.
There are three functions that create a new subprocess in which to run
a program. One of them, start-process, creates an asynchronous
process and returns a process object (see section Creating an Asynchronous Process).
The other two, call-process and call-process-region,
create a synchronous process and do not return a process object
(see section Creating a Synchronous Process).
Synchronous and asynchronous processes are explained in following sections. Since the three functions are all called in a similar fashion, their common arguments are described here.
In all cases, the function's program argument specifies the
program to be run. An error is signaled if the file is not found or
cannot be executed. If the file name is relative, the variable
exec-path contains a list of directories to search. Emacs
initializes exec-path when it starts up, based on the value of
the environment variable PATH. The standard file name
constructs, `~', `.', and `..', are interpreted as usual
in exec-path, but environment variable substitutions
(`$HOME', etc.) are not recognized; use
substitute-in-file-name to perform them (see section Functions that Expand Filenames).
Each of the subprocess-creating functions has a buffer-or-name
argument which specifies where the standard output from the program will
go. If buffer-or-name is nil, that says to discard the
output unless a filter function handles it. (See section Process Filter Functions,
and section Reading and Printing Lisp Objects.) Normally, you should avoid having multiple
processes send output to the same buffer because their output would be
intermixed randomly.
All three of the subprocess-creating functions have a &rest
argument, args. The args must all be strings, and they are
supplied to program as separate command line arguments. Wildcard
characters and other shell constructs are not allowed in these strings,
since they are passed directly to the specified program.
Please note: the argument program contains only the name of the program; it may not contain any command-line arguments. You must use args to provide those.
The subprocess gets its current directory from the value of
default-directory (see section Functions that Expand Filenames).
The subprocess inherits its environment from Emacs; but you can
specify overrides for it with process-environment. See section Operating System Environment.
wakeup is an example of such a program;
the display-time command uses it to get a reminder once per
minute.
nil, which stands for the default
directory (which is the value of default-directory).
The value of exec-path is used by call-process and
start-process when the program argument is not an absolute
file name.
After a synchronous process is created, Emacs waits for the
process to terminate before continuing. Starting Dired is an example of
this: it runs ls in a synchronous process, then modifies the
output slightly. Because the process is synchronous, the entire
directory listing arrives in the buffer before Emacs tries to do
anything with it.
While Emacs waits for the synchronous subprocess to terminate, the
user can quit by typing C-g. The first C-g tries to kill
the subprocess with a SIGINT signal; but it waits until the
subprocess actually terminates before quitting. If during that time the
user types another C-g, that kills the subprocess instantly with
SIGKILL and quits immediately. See section Quitting.
The synchronous subprocess functions returned nil in version
18. In version 19, they return an indication of how the process
terminated.
The standard input for the process comes from file infile if
infile is not nil and from `/dev/null' otherwise. The
process output gets inserted in buffer buffer-or-name before point,
if that argument names a buffer. If buffer-or-name is t,
output is sent to the current buffer; if buffer-or-name is
nil, output is discarded.
If buffer-or-name is the integer 0, call-process returns
nil immediately and discards any output. In this case, the
process is not truly synchronous, since it can run in parallel with
Emacs; but you can think of it as synchronous in that Emacs is
essentially finished with the subprocess as soon as this function
returns.
If display is non-nil, then call-process redisplays
the buffer as output is inserted. Otherwise the function does no
redisplay, and the results become visible on the screen only when Emacs
redisplays that buffer in the normal course of events.
The remaining arguments, args, are strings that specify command line arguments for the program.
The value returned by call-process (unless you told it not to
wait) indicates the reason for process termination. A number gives the
exit status of the subprocess; 0 means success, and any other value
means failure. If the process terminated with a signal,
call-process returns a string describing the signal.
In the examples below, the buffer `foo' is current.
(call-process "pwd" nil t)
=> nil
---------- Buffer: foo ----------
/usr/user/lewis/manual
---------- Buffer: foo ----------
(call-process "grep" nil "bar" nil "lewis" "/etc/passwd")
=> nil
---------- Buffer: bar ----------
lewis:5LTsHm66CSWKg:398:21:Bil Lewis:/user/lewis:/bin/csh
---------- Buffer: bar ----------
The insert-directory function contains a good example of the use
of call-process:
(call-process insert-directory-program nil t nil switches
(if full-directory-p
(concat (file-name-as-directory file) ".")
file))
nil; this is useful when buffer
is t, to insert the output in the current buffer.
The arguments buffer-or-name and display control what to do
with the output from the subprocess, and whether to update the display
as it comes in. For details, see the description of
call-process, above. If buffer-or-name is the integer 0,
call-process-region discards the output and returns nil
immediately, without waiting for the subprocess to finish.
The remaining arguments, args, are strings that specify command line arguments for the program.
The return value of call-process-region is just like that of
call-process: nil if you told it to return without
waiting; otherwise, a number or string which indicates how the
subprocess terminated.
In the following example, we use call-process-region to run the
cat utility, with standard input being the first five characters
in buffer `foo' (the word `input'). cat copies its
standard input into its standard output. Since the argument
buffer-or-name is t, this output is inserted in the current
buffer.
---------- Buffer: foo ----------
input-!-
---------- Buffer: foo ----------
(call-process-region 1 6 "cat" nil t)
=> nil
---------- Buffer: foo ----------
inputinput-!-
---------- Buffer: foo ----------
The shell-command-on-region command uses
call-process-region like this:
(call-process-region
start end
shell-file-name ; Name of program.
nil ; Do not delete region.
buffer ; Send output to buffer.
nil ; No redisplay during output.
"-c" command) ; Arguments for the shell.
On MS-DOS, you must indicate whether the data going to and from a synchronous subprocess are text or binary. Text data requires translation between the end-of-line convention used within Emacs (a single newline character) and the convention used outside Emacs (the two-character sequence, CRLF).
The variable binary-process-input applies to input sent to the
subprocess, and binary-process-output applies to output received
from it. A non-nil value means the data is non-text; nil
means the data is text, and calls for conversion.
nil, convert newlines to CRLF sequences in
the input to a synchronous subprocess.
nil, convert CRLF sequences to newlines in
the output from a synchronous subprocess.
See section Files and MS-DOS, for related information.
After an asynchronous process is created, Emacs and the Lisp
program both continue running immediately. The process may thereafter
run in parallel with Emacs, and the two may communicate with each other
using the functions described in following sections. Here we describe
how to create an asynchronous process with start-process.
The remaining arguments, args, are strings that specify command line arguments for the program.
In the example below, the first process is started and runs (rather, sleeps) for 100 seconds. Meanwhile, the second process is started, and given the name `my-process<1>' for the sake of uniqueness. It inserts the directory listing at the end of the buffer `foo', before the first process finishes. Then it finishes, and a message to that effect is inserted in the buffer. Much later, the first process finishes, and another message is inserted in the buffer for it.
(start-process "my-process" "foo" "sleep" "100")
=> #<process my-process>
(start-process "my-process" "foo" "ls" "-l" "/user/lewis/bin")
=> #<process my-process<1>>
---------- Buffer: foo ----------
total 2
lrwxrwxrwx 1 lewis 14 Jul 22 10:12 gnuemacs --> /emacs
-rwxrwxrwx 1 lewis 19 Jul 30 21:02 lemon
Process my-process<1> finished
Process my-process finished
---------- Buffer: foo ----------
start-process except that it uses a shell
to execute the specified command. The argument command is a shell
command name, and command-args are the arguments for the shell
command.
nil, then pipes are used.
If it is t, then PTYs are used (or pipes if PTYs are
not supported).
PTYs are usually preferable for processes visible to the user, as in Shell mode, because they allow job control (C-c, C-z, etc.) to work between the process and its children whereas pipes do not. For subprocesses used for internal purposes by programs, it is often better to use a pipe, because they are more efficient. In addition, the total number of PTYs is limited on many systems and it is good not to waste them.
The value process-connection-type is used when
start-process is called. So you can specify how to communicate
with one subprocess by binding the variable around the call to
start-process.
(let ((process-connection-type nil)) ; Use a pipe. (start-process ...))
Deleting a process disconnects Emacs immediately from the subprocess, and removes it from the list of active processes. It sends a signal to the subprocess to make the subprocess terminate, but this is not guaranteed to happen immediately. The process object itself continues to exist as long as other Lisp objects point to it.
You can delete a process explicitly at any time. Processes are deleted automatically after they terminate, but not necessarily right away. If you delete a terminated process explicitly before it is deleted automatically, no harm results.
exit or to a signal). If it is
nil, then they continue to exist until the user runs
list-processes. Otherwise, they are deleted immediately after
they exit.
SIGHUP signal. The argument name may be a process,
the name of a process, a buffer, or the name of a buffer.
(delete-process "*shell*")
=> nil
t.
(process-kill-without-query (get-process "shell"))
=> t
Several functions return information about processes.
list-processes is provided for interactive use.
nil.
(process-list)
=> (#<process display-time> #<process shell>)
nil if
there is none. An error is signaled if name is not a string.
(get-process "shell")
=> #<process shell>
(process-command (get-process "shell"))
=> ("/bin/csh" "-i")
The possible values for an actual subprocess are:
run
stop
exit
signal
open
closed
nil
(process-status "shell")
=> run
(process-status (get-buffer "*shell*"))
=> run
x
=> #<process xx<1>>
(process-status x)
=> exit
For a network connection, process-status returns one of the symbols
open or closed. The latter means that the other side
closed the connection, or Emacs did delete-process.
In earlier Emacs versions (prior to version 19), the status of a network
connection was run if open, and exit if closed.
process-status to
determine which of those it is.) If process has not yet
terminated, the value is 0.
Asynchronous subprocesses receive input when it is sent to them by Emacs, which is done with the functions in this section. You must specify the process to send input to, and the input data to send. The data appears on the "standard input" of the subprocess.
Some operating systems have limited space for buffered input in a PTY. On these systems, Emacs sends an EOF periodically amidst the other characters, to force them through. For most programs, these EOFs do no harm.
nil, the current buffer's
process is used.
The function returns nil.
(process-send-string "shell<1>" "ls\n")
=> nil
---------- Buffer: *shell* ----------
...
introduction.texi syntax-tables.texi~
introduction.texi~ text.texi
introduction.txt text.texi~
...
---------- Buffer: *shell* ----------
nil, the current buffer's process is
used.)
An error is signaled unless both start and end are integers or markers that indicate positions in the current buffer. (It is unimportant which number is larger.)
If process-name is not supplied, or if it is nil, then
this function sends the EOF to the current buffer's process. An
error is signaled if the current buffer has no process.
The function returns process-name.
(process-send-eof "shell")
=> "shell"
Sending a signal to a subprocess is a way of interrupting its
activities. There are several different signals, each with its own
meaning. The set of signals and their names is defined by the operating
system. For example, the signal SIGINT means that the user has
typed C-c, or that some analogous thing has happened.
Each signal has a standard effect on the subprocess. Most signals kill the subprocess, but some stop or resume execution instead. Most signals can optionally be handled by programs; if the program handles the signal, then we can say nothing in general about its effects.
You can send signals explicitly by calling the functions in this
section. Emacs also sends signals automatically at certain times:
killing a buffer sends a SIGHUP signal to all its associated
processes; killing Emacs sends a SIGHUP signal to all remaining
processes. (SIGHUP is a signal that usually indicates that the
user hung up the phone.)
Each of the signal-sending functions takes two optional arguments: process-name and current-group.
The argument process-name must be either a process, the name of
one, or nil. If it is nil, the process defaults to the
process associated with the current buffer. An error is signaled if
process-name does not identify a process.
The argument current-group is a flag that makes a difference
when you are running a job-control shell as an Emacs subprocess. If it
is non-nil, then the signal is sent to the current process-group
of the terminal which Emacs uses to communicate with the subprocess. If
the process is a job-control shell, this means the shell's current
subjob. If it is nil, the signal is sent to the process group of
the immediate subprocess of Emacs. If the subprocess is a job-control
shell, this is the shell itself.
The flag current-group has no effect when a pipe is used to
communicate with the subprocess, because the operating system does not
support the distinction in the case of pipes. For the same reason,
job-control shells won't work when a pipe is used. See
process-connection-type in section Creating an Asynchronous Process.
SIGINT. Outside of Emacs, typing the "interrupt
character" (normally C-c on some systems, and DEL on
others) sends this signal. When the argument current-group is
non-nil, you can think of this function as "typing C-c"
on the terminal by which Emacs talks to the subprocess.
SIGKILL. This signal kills the subprocess immediately,
and cannot be handled by the subprocess.
SIGQUIT to the process
process-name. This signal is the one sent by the "quit
character" (usually C-b or C-\) when you are not inside
Emacs.
SIGTSTP. Use continue-process to resume its
execution.
On systems with job control, the "stop character" (usually C-z)
sends this signal (outside of Emacs). When current-group is
non-nil, you can think of this function as "typing C-z"
on the terminal Emacs uses to communicate with the subprocess.
SIGCONT. This presumes that process-name was
stopped previously.
There are two ways to receive the output that a subprocess writes to its standard output stream. The output can be inserted in a buffer, which is called the associated buffer of the process, or a function called the filter function can be called to act on the output.
A process can (and usually does) have an associated buffer, which is an ordinary Emacs buffer that is used for two purposes: storing the output from the process, and deciding when to kill the process. You can also use the buffer to identify a process to operate on, since in normal practice only one process is associated with any given buffer. Many applications of processes also use the buffer for editing input to be sent to the process, but this is not built into Emacs Lisp.
Unless the process has a filter function (see section Process Filter Functions),
its output is inserted in the associated buffer. The position to insert
the output is determined by the process-mark (see section Process Information), which is then updated to point to the end of the text
just inserted. Usually, but not always, the process-mark is at
the end of the buffer. If the process has no buffer and no filter
function, its output is discarded.
(process-buffer (get-process "shell"))
=> #<buffer *shell*>
If process does not have a buffer, process-mark returns a
marker that points nowhere.
Insertion of process output in a buffer uses this marker to decide where to insert, and updates it to point after the inserted text. That is why successive batches of output are inserted consecutively.
Filter functions normally should use this marker in the same fashion
as is done by direct insertion of output in the buffer. A good
example of a filter function that uses process-mark is found at
the end of the following section.
When the user is expected to enter input in the process buffer for transmission to the process, the process marker is useful for distinguishing the new input from previous output.
nil, the process becomes
associated with no buffer.
(get-buffer-process "*shell*")
=> #<process shell>
Killing the process's buffer deletes the process, which kills the
subprocess with a SIGHUP signal (see section Sending Signals to Processes).
A process filter function is a function that receives the standard output from the associated process. If a process has a filter, then all output from that process, that would otherwise have been in a buffer, is passed to the filter. The process buffer is used directly for output from the process only when there is no filter.
A filter function must accept two arguments: the associated process and a string, which is the output. The function is then free to do whatever it chooses with the output.
A filter function runs only while Emacs is waiting (e.g., for terminal
input, or for time to elapse, or for process output). This avoids the
timing errors that could result from running filters at random places in
the middle of other Lisp programs. You may explicitly cause Emacs to
wait, so that filter functions will run, by calling sit-for,
sleep-for or accept-process-output (see section Accepting Output from Processes). Emacs is also waiting when the command loop is reading input.
Quitting is normally inhibited within a filter function--otherwise,
the effect of typing C-g at command level or to quit a user
command would be unpredictable. If you want to permit quitting inside a
filter function, bind inhibit-quit to nil.
See section Quitting.
Many filter functions sometimes or always insert the text in the
process's buffer, mimicking the actions of Emacs when there is no
filter. Such filter functions need to use set-buffer in order to
be sure to insert in that buffer. To avoid setting the current buffer
semipermanently, these filter functions must use unwind-protect
to make sure to restore the previous current buffer. They should also
update the process marker, and in some cases update the value of point.
Here is how to do these things:
(defun ordinary-insertion-filter (proc string)
(let ((old-buffer (current-buffer)))
(unwind-protect
(let (moving)
(set-buffer (process-buffer proc))
(setq moving (= (point) (process-mark proc)))
(save-excursion
;; Insert the text, moving the process-marker.
(goto-char (process-mark proc))
(insert string)
(set-marker (process-mark proc) (point)))
(if moving (goto-char (process-mark proc))))
(set-buffer old-buffer))))
The reason to use an explicit unwind-protect rather than letting
save-excursion restore the current buffer is so as to preserve
the change in point made by goto-char.
To make the filter force the process buffer to be visible whenever new
text arrives, insert the following line just before the
unwind-protect:
(display-buffer (process-buffer proc))
To force point to move to the end of the new output no matter where
it was previously, eliminate the variable moving and call
goto-char unconditionally.
All filter functions that do regexp searching or matching should save
and restore the match data. Otherwise, a filter function that runs
during a call to sit-for might clobber the match data of the
program that called sit-for. See section The Match Data.
A filter function that writes the output into the buffer of the
process should check whether the process is still alive. If it tries to
insert into a dead buffer, it will get an error. If the buffer is dead,
(buffer-name (process-buffer process)) returns nil.
The output to the function may come in chunks of any size. A program that produces the same output twice in a row may send it as one batch of 200 characters one time, and five batches of 40 characters the next.
nil, it gives the process no filter.
nil
if it has none.
Here is an example of use of a filter function:
(defun keep-output (process output)
(setq kept (cons output kept)))
=> keep-output
(setq kept nil)
=> nil
(set-process-filter (get-process "shell") 'keep-output)
=> keep-output
(process-send-string "shell" "ls ~/other\n")
=> nil
kept
=> ("lewis@slug[8] % "
"FINAL-W87-SHORT.MSS backup.otl kolstad.mss~
address.txt backup.psf kolstad.psf
backup.bib~ david.mss resume-Dec-86.mss~
backup.err david.psf resume-Dec.psf
backup.mss dland syllabus.mss
"
"#backups.mss# backup.mss~ kolstad.mss
")
Output from asynchronous subprocesses normally arrives only while Emacs is waiting for some sort of external event, such as elapsed time or terminal input. Occasionally it is useful in a Lisp program to explicitly permit output to arrive at a specific point, or even to wait until output arrives from a process.
nil then this function does
not return until some output has been received from process.
The arguments seconds and millisec let you specify timeout
periods. The former specifies a period measured in seconds and the
latter specifies one measured in milliseconds. The two time periods
thus specified are added together, and accept-process-output
returns after that much time whether or not there has been any
subprocess output.
Not all operating systems support waiting periods other than multiples of a second; on those that do not, you get an error if you specify nonzero millisec.
The function accept-process-output returns non-nil if it
did get some output, or nil if the timeout expired before output
arrived.
A process sentinel is a function that is called whenever the associated process changes status for any reason, including signals (whether sent by Emacs or caused by the process's own actions) that terminate, stop, or continue the process. The process sentinel is also called if the process exits. The sentinel receives two arguments: the process for which the event occurred, and a string describing the type of event.
The string describing the event looks like one of the following:
"finished\n".
"exited abnormally with code exitcode\n".
"name-of-signal\n".
"name-of-signal (core dumped)\n".
A sentinel runs only while Emacs is waiting (e.g., for terminal input,
or for time to elapse, or for process output). This avoids the timing
errors that could result from running them at random places in the
middle of other Lisp programs. A program can wait, so that sentinels
will run, by calling sit-for, sleep-for or
accept-process-output (see section Accepting Output from Processes). Emacs is also
waiting when the command loop is reading input.
Quitting is normally inhibited within a sentinel--otherwise, the
effect of typing C-g at command level or to quit a user command
would be unpredictable. If you want to permit quitting inside a
sentinel, bind inhibit-quit to nil. See section Quitting.
A sentinel that writes the output into the buffer of the process
should check whether the process is still alive. If it tries to insert
into a dead buffer, it will get an error. If the buffer is dead,
(buffer-name (process-buffer process)) returns nil.
All sentinels that do regexp searching or matching should save and
restore the match data. Otherwise, a sentinel that runs during a call
to sit-for might clobber the match data of the program that
called sit-for. See section The Match Data.
nil, then the process will have no sentinel.
The default behavior when there is no sentinel is to insert a message in
the process's buffer when the process status changes.
(defun msg-me (process event)
(princ
(format "Process: %s had the event `%s'" process event)))
(set-process-sentinel (get-process "shell") 'msg-me)
=> msg-me
(kill-process (get-process "shell"))
-| Process: #<process shell> had the event `killed'
=> #<process shell>
nil if it
has none.
nil if Emacs was waiting for keyboard input from the user at
the time the sentinel or filter function was called, nil if it
was not.
You can use a transaction queue for more convenient communication
with subprocesses using transactions. First use tq-create to
create a transaction queue communicating with a specified process. Then
you can call tq-enqueue to send a transaction.
The argument question is the outgoing message which starts the transaction. The argument fn is the function to call when the corresponding answer comes back; it is called with two arguments: closure, and the answer received.
The argument regexp is a regular expression that should match the
entire answer, but nothing less; that's how tq-enqueue determines
where the answer ends.
The return value of tq-enqueue itself is not meaningful.
Transaction queues are implemented by means of a filter function. See section Process Filter Functions.
Emacs Lisp programs can open TCP connections to other processes on the
same machine or other machines. A network connection is handled by Lisp
much like a subprocess, and is represented by a process object.
However, the process you are communicating with is not a child of the
Emacs process, so you can't kill it or send it signals. All you can do
is send and receive data. delete-process closes the connection,
but does not kill the process at the other end; that process must decide
what to do about closure of the connection.
You can distinguish process objects representing network connections
from those representing subprocesses with the process-status
function. See section Process Information.
The name argument specifies the name for the process object. It is modified as necessary to make it unique.
The buffer-or-name argument is the buffer to associate with the
connection. Output from the connection is inserted in the buffer,
unless you specify a filter function to handle the output. If
buffer-or-name is nil, it means that the connection is not
associated with any buffer.
The arguments host and service specify where to connect to; host is the host name (a string), and service is the name of a defined network service (a string) or a port number (an integer).