| Allegro CL version 10.1 Unrevised from 10.0 to 10.1. 10.0 version | ||||||||||
Arguments: command &key input output error-output separate-streams (wait t) if-input-does-not-exist if-output-exists if-error-output-exists show-window environment directory uid gid effective initgroups-user (share-open-files t)
The :osi module (see Operating System
Interface Functionality in
os-interface.htm) has these new operators relating
to running subprocesses: the function command-output
and the macros with-command-output and with-command-io. They are
higher-level than run-shell-command and shell and are now recommended when the
interaction with the subprocess requires input or produces output that
must be captured. The new operators do not have separate description
pages. They are described in OSI process/uid/gid interface
functions in os-interface.htm.
This function, originally written and named for Unix machines, is badly misnamed for Windows, because on Windows, you can only run programs. Shell commands, such as dir, are not acceptable as a value of the command argument. See below in this entry for discussion on this point.
Also note that there is often a Lisp tool equivalent to a command. (For example, the Lisp function directory can be used in place of dir on Windows and ls on Unix. See below under the heading run-shell-command and Windows for why "dir" cannot be the value of the command argument.) It is always preferable to use Lisp tools where possible.
This is particularly true in an application delivered to others. The action of spawning a process, which might involve spawning a shell, may have consequences on another user's machine which you (the application developer) do not expect, and these consequences may interfere with the operation of your application in ways that are difficult to debug.
This is a powerful but complicated function, which has many options and modes of behavior. We have tried to provide sufficient internal links to navigate about this description. All necessary information is here somewhere, so if you scroll done you will likely find what you need.
The process spawned by this function will generally have (always on UNIX, console apps on Windows) standard input, standard output, and standard error. If you want input beyond what is part of the command argument, you need to get a handle on the standard input. If you want to capture (and possibly redirect) the output or the error output, you need to get a handle on standard output and standard error.
You can, as we describe in the discussion of
the input, output,
and error-output keyword
arguments below, set things up yourself, by
specifying streams or file pathnames. And, when
the wait keyword argument
is nil, you can have run-shell-command set up Lisp streams to
connect to process's input, output, and error-output. In that case,
the created streams are among the multiple values returned
by run-shell-command, as we
describe next in the discussion of return values.
This function returns either one value when the
wait argument is true (it defaults to t), or three or four values when
wait
is nil. When wait is
nil, the number of values depends on the
value of the
separate-streams argument
(described below).
When commands display information (such as ls displaying file
and directory information), that information is not what is returned
by run-shell-command. When wait
is true (the default), the single return value is the exit status of
the spawned process. When
wait is nil, the
multiple return values are 2 or 3 values which can be streams
or nil and a final value which is the pid
(process id number) of the spawned process. Output from the command is
sent to whatever the standard out of the spawned process is connected
to.
Other actions when run-shell-command runs: Lisp tries to close all open file descriptors in the process spawned by run-shell-command except for stdin, stdout and stderr. It does this by cycling through the set of legal file descriptors trying to close each one. The overhead is generally insignificant given the low frequency and heavy cost of the operation as a whole. However, this action can cause problems when you are doing performance analysis, particularly involving strace. A sequence of 65000 or so system calls can confuse the analysis. You can limit the number of invalid-fd error returns accepted on closes of sequential fds during this operation. See the function invalid-fd-close-on-fork-limit for more information.
When run-shell-command is
called with the wait keyword argument is true (or unspecified since
the default is t), then the single return
value is the exit status of the spawned process. You must set up
handling of the spawned process's input, output, and error-output
using the input, output,
and error-output keyword arguments as
described below. Here are two examples
showing what we mean. These examples use printenv and so do not
work on Windows, but they do illustrate thr return value.
;; The command is 'printenv DISPLAY'. First we call it with ;; :wait t (just for emphasis, :wait defaults to t): cl-user(131): (run-shell-command "printenv DISPLAY" :wait t) 192.132.95.16:12.0 0 ;; The DISPLAY value is printed BUT NOT RETURNED. What is returned ;; is the single value 0, the exit status of the call to printenv. ;; This is made clear by the following: cl-user(133): (multiple-value-list (run-shell-command "printenv DISPLAY")) 192.132.95.16:12.0 (0) ;; Here we let :wait simply default to t. Again the DISPLAY value is ;; printed, but, as trhe singleton list of return values -- (0) -- shows, ;; not returned. The output is being sent to the standard out of the Lisp ;; process, which the spawned process inherits. ;; In our final example, we send the process output to a stream open to ;; a file: cl-user(136): (with-open-file (s "test.txt" :direction :output :if-exists :supersede) (run-shell-command "printenv DISPLAY" :output s)) 0 ;; And here are the contents of the file: cl-user(137): (run-shell-command "cat test.txt") 192.132.95.16:12.0 0 cl-user(138):
If you specify :wait nil, be sure to also
see More on the :wait argument below for
information on reaping the process when it completes.
When wait is
specified nil, run-shell-command returns three or four
values: three when the separate-streams keyword
argument is nil, four
when separate-streams is true. The
separate-streams argument is discussed in this
section and in its own section below).
When the value of the wait keyword argument is
specified nil, then you can have
run-shell-command create
streams for any or all of the spawned process's standard input, output,
and error-output. Any created streams must be made available, and this
is done by returning them. (run-shell-command returns immediately when
wait is nil, and so any returned streams are available at once.)
The interface is complex and admittedly non-intuitive (but it has been
around so long that it cannot be easily changed). Just keep in mind
that multiple values are returned, that the last return value is the
pid of the spawned process's pid, and that the earlier return values
are either streams set up by run-shell-command
or nil.
The streams that run-shell-command might set up are individual streams for the spawned process's standard in and out, a bidirectional stream for both standard in and out, and/or an individual stream for the spawned process's error output.
run-shell-command will set
up a stream for standard input when input is
:stream, for standard output
when output is :stream, and
for error output when error-output
is :stream. If both input and output
are :stream, a single bidirection stream is set up
for both unless the separate-streams keyword
argument is specified true, in which case different streams are set up
for input and output. run-shell-command will not itself set up a
stream for input, output, or error output when,
respectively, input, output,
and error-output have any value other than
:stream. The other allowable values for those
arguments are discussed below.
Here are some sample calling templates to illustrate the range of possible return values. Unspecified keyword arguments (indicated by suspension points -- ...) do not affect the indicated return values.
;; First a reminder about calls with :wait t.
;; When the :wait argument is true (as it is by default),
;; a single value is returned. Here are two examples with :wait t.
;; The second signals an error since none of :input, :output, or
;; :error-output can be :stream when :wait is true.
(run-shell-command command :wait t ...)
RETURNS as a single value the exit status of the spawned
process which executes command (note that the default value
of :wait is T so ':wait t' need not be specified)
(run-shell-command command :wait t
:input :stream
:output :stream
:error-output :stream
...)
ERROR: none of :input, :output, or :error-output can be :stream
when :wait is t.
In all remaining examples, wait
is nil and multiple values are
returned. input, output, and
error can each have the
value :stream or other values
described below. For showing what is
returned, it only matters whether each of the
arguments input, output, and
error is :stream or not, so we
show the value as :stream, or [some value other than :stream], or
[any value including :stream].
run-shell-command,
when wait is nil,
returns three values when the separate-streams
keyword argument is nil (the default), and
returns four values when separate-streams is
true. separate-streams refers to streams for
input and output: when nil, a single stream
is set up for input, or output, or both (a bidirectional stream in
that case; when true, separate streams are set up.
;; separate-streams is nil in the first set of examples:
(run-shell-command command
:input [some value other than :stream]
:output [some value other than :stream]
:error-output [some value other than :stream]
:separate-streams nil :wait nil ...)
RETURNS three values
nil
nil
the process id of the spawned process
(run-shell-command command
:input :stream
:output [some value other than :stream]
:error-output [some value other than :stream]
:separate-streams nil :wait nil ...)
RETURNS three values
a stream (which acts as standard input to the spawned process)
nil
the process id of the spawned process
(run-shell-command command
:input [some value other than :stream]
:output :stream
:error-output [some value other than :stream]
:separate-streams nil :wait nil ...)
RETURNS three values
a stream (which acts as standard output from the spawned process)
nil
the process id of the spawned process
(run-shell-command command
:input :stream
:output :stream
:error-output [some value other than :stream]
:separate-streams nil :wait nil ...)
RETURNS three values
bi-directional-stream (standard in and standard out of spawned process)
nil
the process id of the spawned process
(run-shell-command command
:input [any value including :stream]
:output [any value including :stream]
:error-output :stream
:separate-streams nil :wait nil ...)
RETURNS three values:
stream-or-nil (stream if either :input or :output is :stream)
stream (for standard error of the spawned process)
the process id of the spawned process
;; In the next example, separate-streams is
;; t and four values are returned.
(run-shell-command command :separate-streams t :wait nil ...)
RETURNS four values:
stream-or-nil (a stream to standard input if :input is :stream, nil otherwise)
stream-or-nil (a stream to standard output if :output is :stream, nil otherwise)
stream-or-nil (a stream to error-output if :error-output is :stream, nil otherwise)
the process id of the spawned process
;; In this next example, we create a stream ourselves and use it for
;; :output. Note that RUN-SHELL-COMMAND returns NIL for the output stream
;; value (second returned value since separate-streams is true)
;; even though output is actually a stream, since RUN-SHELL-COMMAND
;; itself set up the stream:
cl-user(139): (with-open-file (s "test.txt" :direction :output
:if-exists :supersede)
(run-shell-command "printenv DISPLAY" :output s :wait nil
:separate-streams t))
nil
nil
nil
13147
cl-user(140):
The command argument can be a string containing a shell command (in Unix) or a program (in Windows). On Unix only, command can be a simple vector (element type t).
Here are two Unix invocations, one using a vector and one a string:
(run-shell-command #("ls" "ls" "-l" "dcl"))
(run-shell-command "ls -l dcl")
Note that when command is a string, and wait
is nil, the returned process-id will be that
of the shell, not the command itself, unless the command is prefaced
with the shell `exec' built-in command. Thus
(run-shell-command "ls" :wait nil)
will return the process-id of the shell in which ls is run while
(run-shell-command "exec ls" :wait nil)
will return the process-id of ls as will
(run-shell-command #("ls" "ls") :wait nil)
The values of input, output, and error-output keyword arguments control what the spawned process will use as standard input (file descriptor 0), standard output (file descriptor 1) and standard error (file descriptor 2). The values can be
nil - inherit standard input, output or error
(respectively) from Lisp. Note that this is Unix standard input and
output inherited from the Lisp process and is unrelated to the Lisp
variables *standard-input* and *standard-output*.
:stream - create a stream and return it
to the Lisp program. Since waiting and having a stream open to a
process can cause the process to hang, :wait must
be nil; if :wait is
t, an error will result. (When
:stream is specified, the actual stream is not
available to the program until run-shell-command returns, at which time it is
really too late to be useful when :wait is t.)
error-output has an additional allowed value:
:output, which directs standard error to the same
place as standard output (file descriptor 1 rather than 2).
The default value in all cases is nil.
The directory keyword argument can be used to
specify the directory in which the command runs. It defaults to
nil, which results in the command being run
in the directory returned by current-directory. If its value is a directory
pathname or namestring, the command is run in that directory.
wait may be t or nil. If wait is t, Lisp will wait for the command to exit before
resuming. If wait is nil, Lisp will start the process and then resume
without waiting for it to finish. The default for
wait is t.
Note that if wait is specified as nil, then the process will remain in the system after
it completes until either Lisp exits or Lisp executes reap-os-subprocess to inquire
about the exit status. To prevent the system becoming clogged with
processes, a program that spawns a number of processes with :wait set
to nil must be sure to call reap-os-subprocess after each process
finishes. Further, the streams returned by run-shell-command
must be closed (the system will not close them automatically).
Also see Killing a process started with :wait nil below.
The keyword arguments if-input-does-not-exist, if-output-exists and if-error-output-exists are used when input, output or error-output are pathnames (or strings naming files).
Lisp uses open to open a stream to the file identified by the pathname (or string) and the values of if-input-does-not-exist, if-output-exists and if-error-output-exists are passed to the open function as the value of its if-does-not-exist parameter (for input) and the if-exists parameter (for output and error-output.)
The permissible values for if-does-not-exist are
:error, :create and nil.
Those for if-output-exists and
if-error-output-exists are
:error, :overwrite,
:append, :supersede and nil. The default is :error in all
cases.
The separate-streams keyword argument causes
separate streams to be created and returned (rather than a single,
bi-directional stream) when input and
output are both :stream. When
separate-streams is true, four values are
returned, for the input stream, the
output stream, and the
error-output stream.
In all cases, nil will be returned in place
of a stream if the relevant argument is not :stream
(see the call templates Again, the allowed values are
described above).
The value of separate-streams only matters when
the value of the wait is nil.
The value of environment should be an association list of names and values or a list of lists, where the name is the first element of the list and the value is the second (further elements are ignored). Names and values should be strings. Each name should be the name of an environment variable to set in the process being spawned, and the corresponding value should be the desired value for that variable. The environment variable is only set in the process being spawned, and is not set in the process executing the run-shell-command, nor is it remembered in future calls to run-shell-command.
run-shell-command adds the specified environment variables and values to the existing environment and passes that to the subprocess it is starting. Specifying an environment name that already exists will cause that name's value to be replaced with the new value specified.
4096 bytes is allocated for storing new names and values.
In the following example, we show that the DISPLAY environment
variable has a value, but that value is changed in the spawned process
when we specify :environment '(("DISPLAY"
"111.222.33.444:0")), and that the change is not remembered.
cl-user(6): (run-shell-command "printenv DISPLAY")
192.132.95.213:0
0
cl-user(7): (run-shell-command "printenv DISPLAY"
:environment
'(("DISPLAY" "111.222.33.444:0")))
111.222.33.444:0
0
cl-user(8): (run-shell-command "printenv DISPLAY")
192.132.95.213:0
0
cl-user(9):
On all Windows OS's, run-shell-command executes programs but does not invoke shell commands. The function is therefore misnamed for Windows. It is called run-shell-command to provide cross-platform compatibility between Windows and Unix. It does behave differently on Windows and Unix. The difference is related to differences between UNIX and Windows.
On Unix run-shell-command spawns a Unix process and runs the executable program or Unix shell command provided as argument to the function. On Windows, the argument to run-shell-command command is started directly and so works with programs but the Windows command shell is not invoked and so run-shell-command does not work with shell commands. An example of a program is Notepad. Either of these forms will start Notepad on Windows:
(run-shell-command "notepad") (run-shell-command "notepad.exe")
To run a DOS shell command, the argument to run-shell-command has to start the shell and tell it to run the command.
Here is an example:
On Windows the name of the shell is cmd.exe. Here is an example of using run-shell-command on Windows:
(run-shell-command "cmd /c start \"c:\\Program Files\\\"")
which opens a command prompt in the c:\\Program Files\\ directory.
This displays the Windows explorer and avoids cmd altogether:
(run-shell-command "explorer c:\\acl")
Also:
<<, <, >, and | ). *terminal-io*
will not be accepted
here, nor will a socket stream nor a stream returned by a previous
run-shell-command. The show-window keyword argument only has effect on Windows. The value controls how the window created by the program run by run-shell-command first appears. The value can be an integer. The integer should be the value of one of the SW_ constants defined in the winuser.h include file that is part of the Windows SDK. The symbolic value should be preferred over an integer, however, since this will be portable in the face of changes to the Windows SDK header files and use on other operating systems. The value can also be one of the following symbols:
nil: let the process itself determine how it
will show its window.
:normal: display a window in neither a maximized
nor minimized state. Same as integer value SW_NORMAL.
:showna: show but do not select (i.e. display the
window but do not bring it to the front). Same as integer value
SW_SHOWNA. Warning: this value may work inconsistently on
different versions of Windows. Use :minimized to ensure the window is
not in front but is accessible, or :hide if
accessing the window is never important.
:hide: hide the window completely. It doesn't even
appear in the task bar. Same as integer value SW_HIDE.
:minimized: display the window as an icon in the
task bar. Same as integer value SW_MINIMIZED.
:maximized: display the window as a full screen
maximized window. Same as integer value SW_MAXIMIZED. Console windows
can not be maximized, however.
When share-open-files is true (the default value), behavior is as it has been before this argument was added: all open file handles that could be shared will be shared.
When share-open-files is specified as nil, then the input,
output, and error-output
keyword arguments must also be nil (which is
their default). In this case no open file handles will be shared and
the spawned process will open its own standard io handles if it needs
them.
On Windows, Allegro CL starts an operating system thread for each output stream created, to be a transfer agent for the data. This thread cannot exit until it knows all the data has been transferred from the external shell process to the Lisp world. This will happen if the external process finishes and the agent has read all the data it produced, or if the stream is explicitly closed on the Lisp side. Since the Lisp portion of the stream has a limited buffer space, there may well be data still to be read long after the external process finishes; the agent cannot transfer all the data to the Lisp buffer until Lisp activity makes room for it by reading the earlier data. The Lisp application that uses run-shell-command with stream output should explicitly close any stream it does not read to the end, or it risks finding the available pool of OS threads completely taken up by zombie stream agents.
run-shell-command starts a process with execlp() when command is a string and (as we said above) execvp() when command is a simple vector.
The SHELL environment variable is used to determine the type of shell spawned. If SHELL has no value. /bin/csh is used. If that fails, /bin/sh is used.
run-shell-command starts up subprocesses of the Lisp process. In order to modify an environment variable for these subprocesses, you must modify the environment for the Lisp process -- it does no good to modify the environment for the parent process of the Lisp process (i.e. the shell where Lisp was started, or Emacs, if Lisp is started as a subprocess of Emacs).
You can poll environment variables with getenv, and you can set environment variables for the Lisp process, and thus for subprocesses created by Lisp (by run-shell-command, e.g.) with setf and getenv.
These arguments only have meaning on UNIX platforms. They are not supported on Windows.
uid and gid are numbers representing user and group ids. effective is a boolean which indicates that uid and gid are effective user and group ids. initgroups-user is a string naming a user.
Here is an example using command-output (which also has the new arguments):
cl-user(1): (require :osi)
t
cl-user(2): (excl.osi:command-output "whoami" :uid 483)
("layer")
nil
0
cl-user(3): (excl.osi:command-output "whoami")
("root")
nil
0
cl-user(4):
The gid, initgroups-user, and uid arguments are independent and are processed in the following order, using the indicated system calls (on most UNIX platforms):
| gid | setgid(), setegid() |
| initgroups-user | initgroups() |
| uid | setuid(), seteuid() |
The group is always set first, since after changing users that user may not have permission to change groups.
;; The current values of the environment variables on your system may, ;; of course, be different from what appears in this example. user(2): (sys:getenv "SHELL") "/bin/csh" user(3): (setf (sys:getenv "SHELL") "/bin/sh") "/bin/sh" user(4): (sys:getenv "SHELL") "/bin/sh"
Here are a couple of examples of run-shell-command. In the first, we simply have run-shell-command execute a simple command (who).
USER(1): (run-shell-command "who") rlogin ttyb Aug 19 08:26 sdj ttyp0 Aug 18 16:08 (rubix) adam ttyp2 Aug 18 16:17 (rubix) dm ttyp4 Aug 19 10:24 (rubix) 0 USER(2):
The second example is more complicated. We cause run-shell-command to spawn a shell and then we communicate with the shell via the stream set up by run-shell-command.
;; First we define a function to read the output from the shell. This
;; function is pretty simple -- it reads characters and prints them
;; out but it does show how a more useful function could be implemented.
USER(24): (defun get-from-shell (stream)
(do ((ch (read-char-no-hang stream)
(read-char-no-hang stream)))
((null ch))
(write-char ch)))
GET-FROM-SHELL
;; Now we initiate the shell:
USER(25): (setq shell-stream
(excl:run-shell-command "csh"
:input :stream
:output :stream
:wait nil))
#<EXCL::BIDIRECTIONAL-TERMINAL-STREAM @ #x10a4aa6>
USER(26): (format shell-stream "who~%")
NIL
USER(27): (force-output shell-stream)
NIL
USER(28): (get-from-shell shell-stream)
rlogin ttya Aug 19 07:06
rlogin ttyb Aug 19 08:26
sdj ttyp0 Aug 18 16:08 (rubix)
cheetham ttyp1 Aug 18 17:17 (frozen)
adam ttyp2 Aug 18 16:17 (rubix)
NIL
;; We exit the shell:
USER(29): (format shell-stream "exit~%")
NIL
;; and close the stream.
USER(30): (close shell-stream)
T
;; We call sys:reap-os-subprocess because we called
;; run-shell-command with :wait nil:
USER(31): (sys:reap-os-subprocess)
0
3995
nil
USER(32):
A call to run-shell-command
with :wait t (the default for
the wait keyword argument
is t) in an SMP Lisp blocks the Lisp process
than calls run-shell-command
but does not block other running processed. This is in contrast to the
behavior of non-SMP multiprocessing, where a call to run-shell-command with :wait
t blocks everything until the shell command completes, as we
describe under the next heading.
run-shell-command does not take multiprocessing into consideration. Therefore, if it is called with the :wait argument true (the default is t), all of Lisp waits for the call to complete, not just the process or thread that called run-shell-command. It is that behavior which is multiprocessing unfriendly. The following is a multiprocessing friendly call to run-shell-command. It does cause the calling process or thread to wait but does not cause the entire Lisp process to wait for the shell command to finish.
(multiple-value-bind (s errs my-pid)
(run-shell-command "sleep 5; ls /usr/bin" :wait nil)
(declare (ignore errs s))
(let ((my-status nil))
(mp::process-wait "for run-shell-command to finish"
#'(lambda ()
(setq my-status
(or my-status
(sys:reap-os-subprocess
:pid my-pid :wait nil)))))
my-status))
Notes:
nil allows Lisp to continue in any
case, and that might be what you want.Code written similar to the following skeleton of code may hang:
(multiple-value-bind (shell-stream error-stream process)
(excl:run-shell-command cmd
:input :stream :output :stream :error :stream)
(when process
(loop (when (sys:reap-os-subprocess :pid process :wait nil)
(return))))
;; now read from shell-stream and then close the streams
)
In the code sample, the process is reaped prior to reading the process output. While this often works, because many programs don't bother to wait for all of their writes to complete before exiting, it may cause hanging if the pipe to which the data is sent fills up and thus not all data can be written until some reading is done, or the child program waits until each input has been read before writing more data. Some operating systems will cause select() to not return ready status on the output descriptor if any data at all is in that pipe, regardless of whether a call to write() would have succeeded.
The correct outline for the code is:
(multiple-value-bind (shell-stream error-stream process)
(excl:run-shell-command cmd
:input :stream :output :stream :error :stream)
;; do all the reading from shell-stream
(when process
(loop (when (sys:reap-os-subprocess :pid process :wait nil)
(return))))
;; close the streams
)
When you are using multiprocessing, you can use multiprocessing tools such as process-wait to ensure that no hanging occurs, as is done in the example under the heading excl:run-shell-command and multiprocessing above.
Such processes often exit on there own or are killed by the user of the application but you may wish to kill them yourself from within Lisp. To do this, you need the pid value (which is the third returned value from run-shell-command when :wait is nil). We assume in what follows that the pid value is pid. Then:
The function in the excl.osi package kill takes pid and a kill signal as arguments. For example,
(multiple-value-setq (v1 v2 xterm-pid) (run-shell-command "exec xterm" :wait nil)) nil nil 3187 ;; An xterm appears. The pid is 3187, which is the value of XTERM-PID ;; Later: (excl.osi:kill 3187 excl::*sigkill*) ;; excl::*sigkill* is the constant associated with that signal ;; The xterm is killed. ;; ;; ALWAYS reap the killed process (sys:reap-os-subprocess :pid 3187) 0 3187 9
Note the call to sys:reap-os-subprocess. This cleans the process from the system. If you neglkect to do this and you start many subprocesses, the system will clog up.
This method does not work on Windows (kill signals an error if called in a Windows Lisp.
This is a bit more complicated because:
In light of these differrences, be sure to test potential uses of killing spawned processes to ensure that what you want dome is actually done.
The following form kills the process with pid pid:
(run-shell-command (format nil "taskkill /f /pid ~a" pid))
So, for example, we start notepad and then kill it:
(multiple-value-setq (v1 v2 pid) (run-shell-command "notepad" :wait nil)) ;; Notepad runs (run-shell-command (format nil "taskkill /f /pid ~a" pid)) ;; Notepad is killed (sys:reap-os-subprocess :pip pid) ;; we reap the process
See also shell. See os-interface.htm for general information on the interface between Allegro CL and the operating system and for information on shell commands.
Copyright (c) 1998-2022, Franz Inc. Lafayette, CA., USA. All rights reserved.
This page was not revised from the 10.0 page.
Created 2019.8.20.
| Allegro CL version 10.1 Unrevised from 10.0 to 10.1. 10.0 version | ||||||||||