gsu logo

The global subcommand utility

Introduction

gsu is a small library of bash functions intended to ease the task of writing and documenting large shell scripts with multiple subcommands, each providing different functionality. gsu is known to work on Linux, FreeBSD, NetBSD and MacOS.

This document describes how to install and use the gsu library.

Setting up gsu

gsu is very easy to install:

Requirements

gsu is implemented in bash, and thus gsu depends on bash. Bash version 4.3 is required. Besides bash, gsu depends only on programs which are usually installed on any Unix system (awk, grep, sort, …). Care has been taken to not rely on GNU specific behavior of these programs, so it should work on non GNU systems (MacOS, FreeBSD, NetBSD) as well. The gui module depends on the dialog utility.

Download

All gsu modules are contained in a git repository. Get a copy with

    git clone git://git.tuebingen.mpg.de/gsu.git

There is also a gitweb page.

Installation

gsu consists of several independent modules which are all located at the top level directory of the git repository. gsu requires no installation beyond downloading. In particular it is not necessary to make the downloaded files executable. The library modules can be sourced directly, simply tell your application where to find it. The examples of this document assume that gsu is installed in /usr/local/lib/gsu but this is not mandatory. ~/.gsu is another reasonable choice.

Conventions

Public and private functions and variables

Although there is no way in bash to annotate symbols (functions and variables) as private or public, gsu distinguishes between the two. The gsu_* name space is reserved for public symbols while all private symbols start with _gsu.

Private symbols are meant for internal use only. Applications should never use them directly because name and semantics might change between gsu versions.

The public symbols, on the other hand, define the gsu API. This API must not change in incompatible ways that would break existing applications.

$ret and $result

All public gsu functions set the $ret variable to an integer value to indicate success or failure. As a convention, $ret < 0 means failure while a non-negative value indicates success.

The $result variable contains either the result of a function (if any) or further information in the error case. A negative value of $ret is in fact an error code similar to the errno variable used in C programs. It can be turned into a string that describes the error. The public gsu_err_msg() function can be used to pretty-print a suitable error message provided $ret and $result are set appropriately.

The subcommand module

This gsu module provides helper functions to ease the repetitious task of writing applications which operate in several related modes, where each mode of operation corresponds to a subcommand of the application.

With gsu, for each subcommand one must only write a command handler which is simply a function that implements the subcommand. All processing is done by the gsu library. Functions starting with the string com_ are automatically recognized as subcommand handlers.

The startup part of the script has to source the subcommand file of gsu and must then call

    gsu "$@"

Minimal example:

    #!/bin/bash
    com_world()
    {
        echo 'hello world'
    }
    . /usr/local/lib/gsu/subcommand || exit 1
    gsu "$@"

Save this code in a file called hello (adjusting the installation directory if necessary), make it executable (chmod +x hello) and try

    ./hello
    ./hello world
    ./hello invalid

Here, we have created a bash script (hello) that has a single “mode” of operation, specified by the subcommand world.

gsu automatically generates several reserved subcommands, which should not be specified: help, man, prefs, complete.

Command handler structure

For the automatically generated help and man subcommands to work properly, all subcommand handlers must be documented. In order to be recognized as subcommand help text, comments must be prefixed with two # characters and the subcommand documentation must be located between the function “declaration”, com_world() in the example above, and the opening brace that starts the function body.

Example:

    com_world()
    ##
    ##
    ##
    {
        echo 'hello world'
    }

The subcommand documentation consists of the following parts:

The last three parts are optional. All parts should be separated by lines consisting of two # characters only. Example:

    com_world()
    ##
    ## Print the string "hello world" to stdout.
    ##
    ## Usage: world [-v]
    ##
    ## Any arguments to this function are ignored.
    ##
    ## -v: Enable verbose mode
    ##
    ## Warning: This subcommand may cause the top most line of your terminal to
    ## disappear and may cause DATA LOSS in your scrollback buffer. Use with
    ## caution.
    {
        printf 'hello world'
        [[ "$1" == '-v' ]] && printf '!'
        printf '\n'
    }

Replace hello with the above and try:

    ./hello help
    ./hello help world
    ./hello help invalid
    ./hello man

to check the automatically generated help and man subcommands.

Error codes

As mentioned above, all public functions of gsu return an error code in the $ret variable. A negative value indicates failure, and in this case $result contains more information about the error. The same convention applies for subcommand handlers: gsu will automatically print an error message to stderr if a subcommand handler returns with $ret set to a negative value.

To allow for error codes defined by the application, the $gsu_errors variable must be set before calling gsu(). Each non-empty line in this variable should contain an identifier and error string. Identifiers are written in upper case and start with E_. For convenience the $GSU_SUCCESS variable is defined to non-negative value. Subcommand handlers should set $ret to $GSU_SUCCESS on successful return.

To illustrate the $gsu_errors variable, assume the task is to print all mount points which correspond to an ext3 file system in /etc/fstab. We’d like to catch two possible errors: (a) the file does not exist or is not readable, and (b) it contains no ext3 entry. A possible implementation of the ext3 subcommand could look like this (documentation omitted):

    #!/bin/bash

    gsu_errors='
        E_NOENT     No such file or directory
        E_NOEXT3    No ext3 file system detected
    '

    com_ext3()
    {
        local f='/etc/fstab'
        local ext3_lines

        if [[ ! -r "$f" ]]; then
            ret=-$E_NOENT
            result="$f"
            return
        fi
        ext3_lines=$(awk '{if ($3 == "ext3") print $2}' "$f")
        if [[ -z "$ext3_lines" ]]; then
            ret=-$E_NOEXT3
            result="$f"
            return
        fi
        printf 'ext3 mount points:\n%s\n' "$ext3_lines"
        ret=$GSU_SUCCESS
    }

Printing diagnostic output

gsu provides a couple of convenience functions for output. All functions write to stderr.

Subcommands with options

Bash’s getopts builtin provides a way to define and parse command line options, but it is cumbersome to use because one must loop over all given arguments and check the OPTIND and OPTARG variables during each iteration. The gsu_getopts() function makes this repetitive task easier.

gsu_getopts() takes a single argument: the optstring which contains the option characters to be recognized. As usual, if a character is followed by a colon, the option is expected to have an argument. On return $result contains bash code that should be eval'ed to parse the position parameters $1, $2, … of the subcommand according to the optstring.

The shell code returned by gsu_getopts() creates a local variable $o_x for each defined option x. It contains true/false for options without argument and either the empty string or the given argument for options that take an argument.

To illustrate gsu_getopts(), assume the above com_ext3() subcommand handler is to be extended to allow for arbitrary file systems, and that it should print either only the mount point as before or the full line of /etc/fstab, depending on whether the verbose switch -v was given at the command line.

Hence our new subcommand handler must recognize two options: -t for the file system type and -v. Note that -t takes an argument but -v does not. Hence we shall use the optstring t:v as the argument for gsu_getopts() as follows:

    com_fs()
    {
        local f='/etc/fstab'
        local fstype fstab_lines
        local -i awk_field=2

        gsu_getopts 't:v'
        eval "$result"
        ((ret < 0)) && return

        [[ -z "$o_t" ]] && o_t='ext3' # default to ext3 if -t is not given
        [[ "$o_v" == 'true' ]] && awk_field=0 # $0 is the whole line
        fstab_lines=$(awk -v fstype="$o_t" -v n="$awk_field" \
            '{if ($3 == fstype) print $n}' "$f")
        printf '%s entries:\n%s\n' "$o_t" "$fstab_lines"
        ret=$GSU_SUCCESS
    }

Another repetitive task is to check the number of non-option arguments and to report an error if this number turns out to be invalid for the subcommand in question. The gsu_check_arg_count() function performs this check and sets $ret and $result as appropriate. This function takes three arguments: the actual argument count and the minimal and maximal number of non-option arguments allowed. The last argument may be omitted in which case any number of arguments is considered valid.

Our com_world() subcommand handler above ignored any given arguments. Let’s assume we’d like to handle this case and print an error message if one or more arguments are given. With gsu_check_arg_count() this can be achieved as follows:

    com_world()
    {
        gsu_check_arg_count $# 0 0 # no arguments allowed
        ((ret < 0)) && return
        echo 'hello world'
    }

Global documentation

Besides the documentation for subcommands, one might also want to include an overall description of the application which provides general information that is not related to any particular subcommand.

If such a description is included at the top of the script, the automatically generated man subcommand will print it. gsu recognizes the description only if it is enclosed by two lines consisting of at least 70 # characters.

Example:

    #/bin/bash

    #######################################################################
    # gsu-based hello - a cumbersome way to write a hello world program
    # -----------------------------------------------------------------
    # It not only requires one to download and install some totally weird
    # git repo, it also takes about 50 lines of specially written code
    # to perform what a simple echo 'hello world' would do equally well.
    #######################################################################

HTML output

The auto-generated man subcommand produces plain text, html, or roff output.

    ./hello man -m html > index.html

is all it takes to produce an html page for your application. Similarly,

    ./hello man -m roff > hello.1

creates a manual page.

Interactive completion

The auto-generated complete subcommand provides interactive bash completion. To activate completion for the hello program, it is enough to put the following into your ~/.bashrc:

    _hello()
    {
        eval $(hello complete 2>/dev/null)
    }
    complete -F _hello hello

This will give you completion for the first argument of the hello program: the subcommand.

In order to get subcommand-sensitive completion you must provide a completer in your application for each subcommand that is to support completion. Like subcommand handlers, completers are recognized by name: If a function xxx_complete() is defined, gsu will call it on the attempt to complete the xxx subcommand at the subcommand line. gsu has a few functions to aid you in writing a completer.

Let’s have a look at the completer for the above fs subcommand.

    complete_fs()
    {
        local f='/etc/fstab'
        local optstring='t:v'

        gsu_complete_options $optstring "$@"
        ((ret > 0)) && return

        gsu_cword_is_option_parameter $optstring "$@"
        [[ "$result" == 't' ]] && awk '{print $3}' "$f"
    }

Completers are always called with $1 set to the index into the array of words in the current command line when tab completion was attempted (see COMP_CWORD in the bash manual). These words are passed to the completer as $2, $3,…

gsu_complete_options() receives the option string as $1, the word index as $2 and the individual words as $3, $4,… Hence we may simply pass the $optstring and "$@". gsu_complete_options() checks if the current word begins with -, i.e., whether an attempt to complete an option was performed. If yes gsu_complete_options() prints all possible command line options and sets $ret to a positive value.

The last two lines of complete_fs() check whether the word preceding the current word is an option that takes an argument. If it is, that option is returned in $result, otherwise $result is the empty string. Hence, if we are completing the argument to -t, the awk command is executed to print all file system types of /etc/fstab as the possible completions.

See the comments to gsu_complete_options(), gsu_cword_is_option_parameter() and gsu_get_unnamed_arg_num() (which was not covered here) in the subcommand file for a more detailed description.

The gui module

This module can be employed to create interactive dialog boxes from a bash script. It depends on the dialog(1) utility which is available on all Unix systems. On Debian and Ubuntu Linux it can be installed with

    apt-get install dialog

The core of the gui module is the gsu_gui() function which receives a menu tree as its single argument. The menu tree defines a tree of menus for the user to navigate with the cursor keys. As for a file system tree, internal tree nodes represent folders. Leaf nodes, on the other hand, correspond to actions. Pressing enter activates a node. On activation, for internal nodes a new menu with the contents of the subfolder is shown. For leaf nodes the associated action handler is executed.

Hence the application has to provide a menu tree and an action handler for each leaf node defined in the tree. The action handler is simply a function which is named according to the node. In most cases the action handler will run dialog(1) to show some dialog box on its own. Wrappers for some widgets of dialog are provided by the gui module, see below.

Menu trees

The concept of a menu tree is best illustrated by an example. Assume we’d like to write a system utility for the not-so-commandline-affine Linux sysadmin next door. For the implementation we confine ourselves with giving some insight in the system by running lean system commands like df to show the list of file system, or dmesg to print the contents of the kernel log buffer. Bash code which defines the menu tree could look like this:

    menu_tree='
        load_average
        processes
        hardware/
            cpu
            scsi
        storage/
            df
            mdstat
        log/
            syslog
            dmesg
    '

In this tree, hardware/, block_devices/ and log/ are the only internal nodes. Note that these are written with a trailing slash character while the leaf nodes have no slash at the end. All entries of the menu tree must be indented by tab characters.

Action handlers

Action handlers are best explained via example:

Our application, let’s call it lsi for lean system information, must provide action handlers for all leaf nodes. Here is the action handler for the df node:

    lsi_df()
    {
        gsu_msgbox "$(df -h)"
    }

The function name lsi_df is derived from the name of the script (lsi) and the name of the leaf node (df). The function simply passes the output of the df(1) command as the first argument to the public gsu function gsu_msgbox() which runs dialog(1) to display a message box that shows the given text.

gsu_msgbox() is suitable for small amounts of output. For essentially unbounded output like log files that can be arbitrary large, it is better to use gsu_textbox() instead which takes a path to the file that contains the text to show.

To illustrate gsu_input_box() function, assume the action handler for the processes leaf node should ask for a username, and display all processes owned by the given user. This could be implemented as follows.

    lsi_processes()
    {
        local username

        gsu_inputbox 'Enter username' "$LOGNAME"
        ((ret != 0)) && return
        username="$result"
        gsu_msgbox "$(pgrep -lu "$username")"
    }

Once all other action handlers have been defined, the only thing left to do is to source the gsu gui module and to call gsu_gui():

    . /usr/local/lib/gsu/gui || exit 1
    gsu_gui "$menu_tree"

Example

The complete lsi script below can be used as a starting point for your own gsu gui application. If you cut and paste it, be sure to not turn tab characters into space characters.

    #!/bin/bash

    menu_tree='
        load_average
        processes
        hardware/
            cpu
            scsi
        storage/
            df
            mdstat
        log/
            syslog
            dmesg
    '

    lsi_load_average()
    {
        gsu_msgbox "$(cat /proc/loadavg)"
    }

    lsi_processes()
    {
        local username

        gsu_inputbox 'Enter username' "$LOGNAME"
        ((ret < 0)) && return
        username="$result"
        gsu_msgbox "$(pgrep -lu "$username")"
    }

    lsi_cpu()
    {
        gsu_msgbox "$(lscpu)"
    }

    lsi_scsi()
    {
        gsu_msgbox "$(lsscsi)"
    }

    lsi_df()
    {
        gsu_msgbox "$(df -h)"
    }

    lsi_mdstat()
    {
        gsu_msgbox "$(cat /proc/mdstat)"
    }

    lsi_dmesg()
    {
        local tmp="$(mktemp)" || exit 1

        trap "rm -f $tmp" EXIT
        dmesg > $tmp
        gsu_textbox "$tmp"
    }

    lsi_syslog()
    {
        gsu_textbox '/var/log/syslog'
    }

    . /usr/local/lib/gsu/gui || exit 1
    gsu_gui "$menu_tree"

The config module

Some applications need config options which are not related to any particular subcommand, like the URL of a web service, the path to some data directory, or a default value which is to be used by several subcommands. Such options do not change frequently and are hence better stored in a configuration file rather than passed to every subcommand that needs the information.

The config module of gsu makes it easy to maintain such options and performs routine tasks like reading and checking the values given in the config file, or printing out the current configuration. It can be used stand-alone, or in combination with either the subcommand or the gui module.

Defining config options

To use the config module, you must define the $gsu_options bash array. Each config option is represented by one slot in this array. Here is an example which defines two options:

    gsu_options=(
    "
    name=fs_type
    option_type=string
    default_value=ext3
    required=false
    description='file system type to consider'
    help_text='
        This option is used in various contexts. All
        subcommands which need a file system type
        use the value specified here as the default.
    '
    "
    "
    name=limit
    option_type=num
    default_value=3
    required=no
    description='print at most this many lines of output'
    "
    )

Each config option consists of the following fields:

To enable the config module you must source the config module of gsu after $gsu_options has been defined:

    . /usr/local/lib/gsu/config || exit 1

Passing config options to the application

There are two ways to pass the value of an option to a gsu application: environment variable and config file. The default config file is ~/.$gsu_name.rc where $gsu_name is the basename of the application, but this can be changed by setting $gsu_config_file. Thus, the following two statements are equivalent

    fs_type=xfs hello fs
    echo 'fs_type=xfs' > ~/.hello.rc && hello fs

If an option is set both in the environment and in the config file, the environment takes precedence.

Checking config options

The gsu config module defines two public functions for this purpose: gsu_check_options() and gsu_check_options_or_die(). The latter function exits on errors while the former function only sets $ret and $result as appropriate and lets the application deal with the error. The best place to call one of these functions is after sourcing the config module but before calling gsu() or gsu_gui().

Using config values

The name of an option as specified in $gsu_options (fs_type in the example above) is what users of your application may specify at the command line or in the config file. This leads to a mistake that is easy to make and difficult to debug: The application might use a variable name which is also a config option.

To reduce the chance for this to happen, gsu_check_options() creates a different set of variables for the application where each variable is prefixed with ${gsu_name}. For example, if $gsu_options as above is part of the hello script, $hello_fs_type and $hello_limit are defined after gsu_check_options() returned successfully. Only the prefixed variants are guaranteed to contain the proper value, so this variable should be used exclusively in the application. The prefix may be changed by setting $gsu_config_var_prefix before calling gsu_check_options().

com_prefs()

For scripts which source both the subcommand and the config module, the auto-generated prefs subcommand prints out the current configuration and exits. The description and help text of the option as specified in the description and help_text fields of $gsu_options are shown as comments in the output. Hence this output can be used as a template for the config file.

List of public variables

License

gsu is licensed under the GNU LESSER GENERAL PUBLIC LICENSE (LGPL), version 3. See COPYING and COPYING.LESSER.

Contact

Send beer, pizza, patches, improvements, bug reports, flames, (in this order), to Andre Noll maan@tuebingen.mpg.de.

References