++ed by:
Author image Sullivan Beck
and 1 contributors


Shell::Cmd - run shell commands with enhanced support


   use Shell::Cmd;

   $obj = new Shell::Cmd;


A very common use of perl is to act as a wrapper around shell commands where perl is used to prepare the shell commands, execute them, and deal with the resulting output. Even where the bulk of the work is actually done in the perl script, creating small shell scripts within it to do some portion of the task is common.

In the simplest form, running shell commands can be done very simply using the system() call, backticks, or several other ways, but I usually find myself wanting to do a bit (and sometimes a lot) more, especially when I am writing a long-term script that I want to be robust. In these cases, I frequently ended up writing a subroutine to run the shell command(s) for me with added functionality.

This module is designed to take a list of shell commands and automatically turn them into a shell script (using only basic shell commands which are available in any bourne shell variation) which adds some common desirable functionality including:


Commonly, I want to treat STDOUT and STDERR in some way. I may want to keep one or both of them, or discard one or both of them, or merge them.

Command echoing

A common option I want to set is command echoing where the commands I run are echoed as they are run. I want to be able to easily turn this on or off (typically with a command line option in the calling script).


Another common option is to create a dry-run environment where the shell commands may be printed, but not actually run. Again, I want to be able to turn this on and off easily.

Error trapping

Even though I may combine a number of shell commands into a single script (so that it all runs in one shell), I still want to have built in error trapping at a per-command basis. I want to take a series of commands and know exactly which one failed. If I execute the commands one at a time, I can get that information, but typically, I want to combine multiple commands in a single script but still have that ability.

I also want to be able to control what happens to commands that are listed after a failed command. I may want to ignore an error and continue to run the remaining commands. I may want to simply exit. Or I may want to echo, but not run the remaining commands so that I can see what didn't get completed.

Shell environment

I sometimes want to set up some environment for the script such as what directory it will be run in and what environment variables should be set in advance.

Command alternates

Sometimes, especially if you are running the script on multiple platforms, you may not know which command you should use. You can of course generate a platform specific script, but an alternative is to specify alternate commands. If ANY of those commands succeed, then that portion of the script succeeds.

Command retrying

Occasionally you have a command that may fail, but on retrying, it will succeed. This is especially true when some effect from a previous command takes some amount of time to actually go into effect. By allowing a certain number of retries, you can often work around this situation.

Remote execution

Sometimes you want to run the commands locally. Other times, you want to run it remotely using ssh. When running remotely, you may want to run the same script on multiple hosts.

SSH handling

When running on multiple hosts using SSH, sometimes you need to run the script serially (i.e. one host at a time), but other times, it would be nice to run it in parallel to speed up execution. When running in parallel, you should be able specify how many instances to run at at time.

Quoting and special characters

Since shell commands often have quotes, dollar signs, and other special characters, this module can handle that for you by properly escaping them as necessary.

This module is designed to run multiple commands in a single shell, wrapping them in very simple, standard shell commands to automatically add all of this functionality.


   $obj = new Shell::Cmd;

This creates a new object containing commands.

   $vers = $obj->version();

Returns the version of this module.

   $err = $obj->cmd($cmd [,\%options], $cmd [,\%options], ...);

This is used to add one or more commands to the list of commands that will be executed.

Here, each $cmd is either a string containing a command, or a listref where each element in the list is a string containing a command.

In the listref form, the list of commands are alternates to try until one succeeds, and the command only fails if all of the alternates fail. This might be used to specify different paths to an executable, or different executables that perform essentially the same function, but which might not all be available on all platforms.

For example, if you wanted to run a command to get the contents of a web site, and you didn't know which of curl, wget, or lftp were available, you might use something like this:

   $err = $obj->cmd([ "wget $URL", "curl $URL", "lftp $URL"]);

and in this case, it would try wget, and if that failed, it would try curl, and if that failed, it would try lftp. The command will only fail if all three alternates fail.

Each command (or list of alternates) can have options passed in. These options apply only to this command (or list), and are described in the "PER-COMMAND OPTIONS" section below.

All of the commands stored in $obj will be run in a single shell, so it is fine to gather information in one command and store it in a shell variable for use in a later command. Commands must not include a trailing semi-colon as these will interfere with I/O redirection, and will be added automatically as needed.

An error is returned if any of the arguments are invalid.

It should be noted that no attempt is made to see if the syntax of the shell command is correct. That is beyond the scope of this module.

If only simple lists of commands are used, handling them is relatively straightforward, but trying to include commands that affect the flow of the script (such as while...done, if...else, and the like) then handling can be much more complicated. Refer to the "FLOW COMMANDS" section below. Defining functions is NOT supported.

   $ret = $obj->run();

This prepares a shell script based on the commands and options entered and runs it as appropriate. The script is stored in a temporary file that can be set using the tmp_script option (refer to the "GLOBAL OPTIONS" section below).

There are several different ways in which the commands can be run, and these are described in the options method below. The most important option is the mode option which determines the form of the script, and how it is run.

If the mode is run, the method is called as:

   $err = $obj->run();

In this mode, the script is run, and output is sent directly to STDOUT and STDERR as appropriate for the options specified. In essence, this generates a script and runs it with the system() call.

The error code returned is described below in the "ERROR CODES" section.

In dry-run mode, the method is called as:

   $script = $obj->run();

In this mode, the commands are not actually executed. Instead, the script is built and returned. The form of the script is determined by the script option described below.

In script mode, the method is called as:

   $err = $obj->run();

In this case, the output from the commands are kept for further analysis. The $obj-output(...)> method may then be used to examine the resulting output.

The error codes are described in the "ERROR CODES" section below.


This behaves similar to the run method except it will run the commands on each host in @hosts using ssh. The return values for each mode are identical to the return methods from the run method except that for both the run mode and script mode, the output is returned as a hash where the keys are the hosts and the values are the value for that host.

For example, in run mode, the call would be:

   %err = $obj->ssh(@hosts)

In dry-run mode, the call is identical to the run method, and it will return the script that would be run on each host.

   $script = $obj->ssh(@hosts);

Note that when running in parallel in run mode, the output that is printed to the terminal will be a mix of the output from each of the hosts the commands are being run on.

   $ret = $obj->output(%options);
   @ret = $obj->output(%options);
   %ret = $obj->output(%options);

This will return the output produced by running the commands in script mode depending on the options passed in.

The %options argument is described below in the OUTPUT OPTIONS section.

   $obj->flush( [@opts] );

If @opts is not given, it removes all the data stored in the object, resetting it to a clean object. If @opts is given, you can clear specific parts of the object. Any of the following options can be given:

   commands   : clears all commands and their options
   env        : clears the environment
   opts       : clears all options
   out        : clears the output from running the command
                in B<script> mode
   $err = $obj->dire($dire);

This method is used to set the dire option. For a description, please see the entry in "GLOBAL OPTIONS" below. This is a shortcut for:

   $err = $obj->options('dire',$dire);

You can also check the value that is set using:

   $dire = $obj->dire();
   $err = $obj->mode($mode);

This method is used to set the mode option. For a description, please see the entry in "GLOBAL OPTIONS" below. This is a shortcut for:

   $err = $obj->options('mode',$dire);

You can also check the value that is set using:

   $mode = $obj->mode();
   $obj->env($var1, $val1, $var2, $val2, ...);

This can be called any number of times to set some environment variables. If $val is undef, the environment variable will be explicitly unset.

You can also query the environment variables with:

   %env = $obj->env();
   $err = $obj->options(%options);

This can be used to set some options about what will be done when the commands are run.

The hash is of the form:

   %options = ( OPTION => VALUE,
                OPTION => VALUE, ...)

The options are defined in the "GLOBAL OPTIONS" section below.


The error code returned by the run or ssh methods are described in the following table:

   0       No error
   1-200   The number of the command that failed.
           Commands entered with the B<cmd> method
           are numbered starting at 1.  If 200 or
           less commands are entered, the error code
           will correspond to the command that
   201     If more than 200 commands are entered,
           and any of them beyond the 200th fail,
           the error code will be 201.
   252     An error in the script.  Usually this
           indicates that flow commands not correctly
   253     If the temporary script cannot
           be copied to a remote host (for use in
           the B<ssh> method), this is returned.
   254     If a temporary script could not be
           created, this will be returned.
   255     If a global directory was specified that
           does not exist, this will be returned.


The following global options exist can can be set using the options method:


The mode option determines how the commands will be handled by the run/ssh methods. The following values are available.

   run      (default)

The run mode is the standard way to run commands in an interactive setting. It will run the commands in real-time and allow you to watch STDOUT and/or STDERR (depending on the options you choose) as they run.

The dry-run mode will not execute any commands. Instead, it will generate a script that WOULD have been run and returns it. The script can take several different forms, and is described in the script option below.

The script mode is more appropriate for running in an unattended script. It gathers the output and post-processes it allowing for more useful handling of the output. For example, you could discard the output from commands that succeed and keep only the output for the one that failed, or a number of other options.

The mode option can also be set using the mode method.


The dire option is use to specify the directory where all of the the commands should be run. This can be overridden on a per-command basis using per-command options in the cmd method, but all commands not specifically set will run in this directory.

This does NOT check the existence of the directory until the commands are actually run since the commands may be run via. ssh.

The dire option can also be set using the dire method.


The output option can be one of the following:

   both     (default)

In the run mode, these determine what output will be displayed. In script mode, it determines which output is stored in the object. Obviously, if output is not kept, it will not be available to examine using the output method.

It can display only STDOUT, only STDERR, or both, or both can be discarded with the 'quiet' option. The default is to include 'both'. The 'merged' option is used to display both but merge STDERR into STDOUT (using a 2>&1 redirection).


The script option is used only in dry-run mode.

When commands are run in dry-run mode, a script is produced. The form of that script is controlled by this option. The value may be any of:

   run     (default)

If the value is run or script, the script produced will be exactly the script produced in those modes, including all of the wrapping shell structure to add the requested functionality.

If the value is simple, the script will simply be the list of commands with the minimum necessary additions to handle directory and environment variables. No additional scripting will be added to do error checking or add other functionality.


The echo option is used only in run mode. With it, you can choose whether or not the commands should be displayed when they are run.

The values are:


With echo and noecho values, commands will be displayed or NOT displayed respectively. With echo the commands will be displayed before they are run.

If the value is failed, a command that failed will be displayed. Since it has already run, the command will be echoed after execution rather than before.

Note that flow commands are not echoed.


The failure option is used in run and script modes. When a command fails, there are several alternatives that can be done. Values for this option are:


The default is exit. With this option, the script will stop executing commands once one has failed.

The display option is only used in run mode. With it, if any command fails, a simple script will be displayed showing what commands failed to run.

With the continue option, remaining command are executed, but the overall exit values is still set to point at the first failed command.

tmp_script, tmp_script_keep

The tmp_script option is used to specify a temporary script name.

The script that is generated by this module may exceed the length of a string that can be passed directly to a shell. In order to avoid this problem, the script will be stored in a temporary script file (set with the tmp_script option) which will be executed. If not set, the default value for tmp_script will be:


Once execution is complete, the temporary script file will be removed unless the tmp_script_keep option is set.

ssh_script, ssh_script_keep

These are related to the tmp_script and tmp_script_keep options. If tmp_script is created, then when the ssh method is used to run the script remotely, it is copied to the remote host (via. scp) to a temporary location (given by ssh_script). The remote script is then removed (unless ssh_script_keep is passed in).

If tmp_script is set but ssh_script is NOT, ssh_script defaults to the same value as tmp_script.

ssh_script_keep defaults to 0, even if tmp_script_keep is set.


When running a command on multiple hosts via SSH, it is possible to run them serially (one at a time) or in parallel.

This option can be set to a number 0 or more. If the number is 1, then only a single ssh connection will be made at a time so the hosts will all be contacted serially.

If the option is set to 0, all of the hosts will be run simultaneously.

If the option is set to N, N simultaneous connections will be allowed and additional hosts will be run only after others have completed.


When running a command on multiple hosts via SSH, it is sometimes desirable to stagger them slightly so multiple copies are running at the same time, but not at EXACTLY the same time.

If this option is set to 0 (the default), all of the commands will be run with no delay. If it is set to the value N, commands will sleep a random amount of time (from 0 to N seconds) before running. If it is set to a negative value -N, it will sleep for exactly N seconds.


When running a command on a remote host via. ssh, the Net::OpenSSH module is used.

Every option that can be passed to the Net::OpenSSH::new method can be set here. For example, if you want to call Net::OpenSSH as:

   $ssh = Net::OpenSSH->new($host, user => $user_name);

just set the option:

   ssh:user = $user_name


The following options exists that can be applied to individual commands. They can be set in the cmd method.


The dire option refers to the directory which this single command should be executed in. The value of the option is the directory.

This will basically wrap a command in:

   cd $dire
   cd $CURR_DIR

If the noredir option is included, no command line redirection is done for this command. Most commands automatically redirect STDOUT and STDERR based on the output global option.

If the command explicitly sends these to somewhere (such as a log file or temporary file), use the noredir option so automatic redirection is not done.

Since the command is not parsed to see whether or not redirection is handled by the command, this option must be used with every shell command which includes any type of I/O redirection.

retry, sleep

The retry and sleep options can be used to retry a command.

Sometimes, a command may fail but running it a second time can succeed. Often, a command completes, but for various reasons, it takes a certain amount of time after the command completes for the full results to take effect. A later command might be run before those results have taken effect, but rerunning it a few seconds later would succeed.

With the retry option, you can retry a command. The value of the retry option should be an integer (N). If N is greater than 1, the command will be run up to N times total. Any other value of N will be ignored, and the command will run only a single time.

There can be an optional sleep time between running the command. The optional sleep option (which should also be an integer) sets the number of seconds between retries. If the value is 0, or not an integer, there will be no delay between retries.

This command will be marked as failed only if all of the retries fail.

You cannot retry a flow command.


Sometimes, a command is written such that the exit code does not accurately reflect whether the command failed or not. It may produce a zero exit code but still have failed, or it may have succeeded but still produce an error code.

In these cases, you can supply a command with this option which will check the result of the command and set the error flag appropriately.

If the command succeeded, the error flag should be set to zero. If it failed, it should be set to something non-zero.

If this is given for a command which has alternatives, it will be run after every alternative.


Each command can have a label attached to it which will allow you to refer to that command by the label. This is useful in analyzing the output.

The label should not consist only of digits (i.e. be an integer).


When simple shell commands are given, there is no ambiguity about how to treat each, so handling them is relatively simple. Simple commands are fully supported, and all of the functionality described above can be added.

In order to add the desired functionality, the commands are wrapped with some enclosing shell structure using very basic shell command which add the requested features. Simple command are very easy to wrap in a basic enclosing shell structure. For example, it is easy to turn:

   mycommand arg1 arg2


   if [ SOME_CONDITION ]; then
      mycommand arg1 arg2

However, when commands are added which affect the flow of the script, they must be handled differently than simple commands in order to deal with them properly. Wrapping them in other shell structure would produce invalid shell scripts. As a result, each type of flow must be considered carefully.

Currently, the supported flow commands are:


In order to recognize them, the commands will be partially parsed, and they must be of the forms:

   if ... ; then
   elif ... ; then

where '...' may be any string. In other words, the first line must start with if, followed by whitespace, and end with a ';' followed by optional whitespace, followed by 'then'.

The alternate formatting of:

   if ...

is not supported.


The commands must be of the form:

   while ... ; do
   until ... ; do
   for ... ; do

If flow commands are entered, but not correctly closed and/or nested, an error will be returned.

Also, flow commands must generally be simple tests. If complex shell commands are entered which produce output, this output will NOT be handled correctly, and may actually break things when running in script mode.

At some point, the select and case structures may be supported, but this in not yet available.

Also, shell functions are not currently supported.


When commands are run in the 'script' mode, the output is stored in the object.

To access the output, use one of the following calls to the output method:


Each call to the method will return one part of the output. To following options may be used to determine what is returned.


If the output was generated using the ssh method, this option is required (since it is possible to run the commands both locally and remotely, and the output is stored separately).

HOST can be any of:

   all      The output for all hosts will be returned.  The return value
            will be a hash of the form:

               %ret = ( HOST1 => OUTPUT1, HOST2 => OUTPUT2, ...)

            The output for all of the hosts listed will be returned in a

   HOST     If only a single host is specified, the output for only
            that host will be returned.  It will not be returned as a hash.

This tells what output will be returned. TYPE can be any of:

   stdout   STDOUT will be returned for the command(s) selected.
            This is the default.

   stderr   STDERR will be returned for the command(s) selected.

   command  The command itself will be returned.

   num      The command number will be returned.

   label    The command label (if any) will be returned.

   exit     The exit code will be returned for the command(s) selected.
            The exit code is the one returned by the final alternative
            on the final try.

This tells which commands will be included in the output. COMMAND can be any of:

   curr     An internal flag is kept which starts at the 1st command
            which produced any type of output. The value is returned
            for this command.  This is the default.

   next     The internal flag is incremented, and that becomes the new
            current command.

   all      The value for all commands will be returned in the order they

   CMD_NUM  The commands are numbered starting at 1.  This will return the
            output for only the command given.  Note however that a command
            may occur multiple times (due to retries, being in a loop, etc.)
            so the output will be a list of values, one per occurrence.

   LABEL    This will return the output for all commands assigned the
            given label (using the per-command B<label> option).  Multiple
            commands may be assigned the same label, so the output from
            all occurrences of all commands with this label will be returned
            as a list.

   fail     This will return the output for the command that failed (if any).


Minimal support for complex scripts

These methods work best for simple lists of commands. Using simple command flow (<if...then...else>, etc.) is allowed, but must be used carefully. The use of functions is NOT supported and will not work.

Maximum of 200 commands fully supported

In order to determine which command fails, a unique error code is assigned to each command. Since exit codes must be between 0-255, and some are reserved, there is a limit of 200 commands that can be entered if accurate error tracking is needed.


This script is free software; you can redistribute it and/or modify it under the same terms as Perl itself.


Sullivan Beck (sbeck@cpan.org)