A docopt argument parser for bash 3.2, 4 , and 5 with no external dependencies.
#!/usr/bin/env bash
DOC="Naval Fate.
Usage:
naval_fate.sh <name> move <x> <y> [--speed=<kn>]
naval_fate.sh shoot <x> <y>
Options:
--speed=<kn> Speed in knots [default: 10]."
naval_fate() {
eval "$(docopt "$@")"
if $move; then
printf "The %s is now moving to %d,%d at %d knots.\n" "$_name_" "$_x_" "$_y_" "$__speed"
fi
if $shoot; then
printf "You shoot at %d,%d. It's a hit!\n" "$_x_" "$_y_"
fi
return 0
}
naval_fate "$@"
Run docopt.sh
to insert a parser that matches the helptext into the script:
$ sudo pip3 install docopt-sh
$ docopt.sh naval_fate.sh
naval_fate.sh has been updated
$ ./naval_fate.sh Olympia move 1 5 --speed 8
The Olympia is now moving to 1,5 at 8 knots.
Note that the script is completely independent from the python package.
docopt.sh
is merely a development tool to insert and update the parser, your
scripts will be entirely self contained.
The parser is inlined right beneath the DOC="…"
and takes up ~80 lines of
code (depending on the size of your helptext).
If you ship multiple scripts in the same project
you can reduce that to ~20 lines by
moving the generic parts of the parser to a different file
and reusing them.
Given a script docopt.sh
finds the docopt
(DOC="…"
) help text,
parses it, generates a matching parser in bash, and then inserts it back into
the original script. The patched script will have no dependencies and can be
shipped as a single file.
To reduce the amount of code added to the it, the script will only contain a parser made for that specific help text. For that reason there is no need for the generator itself to be written in bash, instead that part is written in Python 3.11. Though, this also means that you have to regenerate your parser every time you change the help text (see On-the-fly parser generation for automating that part while developing).
Running docopt "$@"
outputs multiple variable declarations (and a
function) whose values match the
command-line arguments that were used.
As an example, invoking naval_fate.sh
from the
quick start section with ./naval_fate.sh shoot 1 5
outputs
the following.
docopt_exit() { [[ -n $1 ]] && printf "%s\n" "$1" >&2
printf "%s\n" "${DOC:12:87}" >&2; exit 1; }
declare -- __speed="10"
declare -- _name_=""
declare -- _x_="1"
declare -- _y_="5"
declare -- move="false"
declare -- shoot="true"
Evaluating (eval
) this in bash will set those variables. If done in a function
the variables will be local and only available inside that function
(like in naval_fate.sh
), otherwise they will be available globally.
docopt.sh
embeds a shasum of the help text into the parser to ensure that the
two always match. In order to update the parser, simply run docopt.sh
again.
The existing parser will be replaced with a new one. If the parser was generated
with any particular options, these options will be re-applied unless instructed
otherwise with --no-auto-params
.
$ docopt.sh --line-length 120 naval_fate.sh
naval_fate.sh has been updated.
$ docopt.sh naval_fate.sh
Adding `--line-length=120` from parser generation parameters that were detected
in the script. Use --no-auto-params to disable this behavior.
The parser in naval_fate.sh is already up-to-date.
Once you have generated the parser, you can move the codeblock to any other place in your script. The script patcher will automatically find the codeblock and replace it with an updated version.
In order to avoid "works on my machine" issues, the parser automatically skips
the help text check on machines without shasum
or sha256sum
(a "command
not found" error will still be printed though). The check can also manually be
disabled with $DOCOPT_DOC_CHECK
(see parser options
for more on that).
Names of arguments, commands, and options are mapped by replacing everything
that is not an alphanumeric character with an underscore. This means --speed
becomes $_speed
, -f
becomes $_f
, and <name>
becomes _name
, while
NAME
stays as $NAME
and set
stays as $set
.
Switches (options without arguments) and commands become true
or false
.
If a switch or command can be specified more than once, the resulting variable
value will be an integer that has been incremented the number of times the
parameter was specified.
Options with values and regular arguments become strings. If an option with a value or an argument can be specified more than once, the value will be an array of strings.
To clarify, given this (somewhat complex, but concise) doc and invocation:
Usage:
program -v... -s --val=VAL multicmd... command ARG ARGS...
$ program -vvv -s --val XY multicmd multicmd command A 1 2 3
The variables and their values will be:
_v=3 # -vvv
_s=true # -s
__val=XY # --val XY
multicmd=2 # multicmd multicmd
command=true # command
ARG=A # A
ARGS=(1 2 3) # 1 2 3
You can use $DOCOPT_PREFIX
to prefix the above variable names with a custom
string (e.g. specifying DOCOPT_PREFIX=prog
would change ARG
to
progARG
). See parser options for additional parser
options.
The commandline options of docopt.sh
only change how the parser is
generated, while global variables specified before eval "$(docopt "$@")"
itself change the behavior of the parser.
The commandline options are:
Option |
Description |
|
Max line length when minifying. Disable with |
|
Generates the dynamic part of the parser and includes the static parts with |
|
Disable auto-detection of parser generation parameters. |
|
Output the parser instead of inserting it in the script. |
|
Show the help screen. |
|
Show docopt.sh version. |
Parser options change the behavior of the parser in various ways. These options
are specified as global variables and must be specified before invoking
eval "$(docopt "$@")"
. You do not need to regenerate the parse when changing
any of these options.
Option |
Default |
Description |
|
|
The string to print when |
|
|
Set to |
|
|
Set to |
|
|
Prefixes all variable names with the specified value |
|
|
Set to |
|
|
Set to |
Oftentimes additional verification of parameters is necessary (e.g. when an
option value is an enum). In those cases you can use docopt_exit "message"
in order to output a message for the user, the function automatically appends a
short usage message (i.e. the Usage:
part of the doc) and then exits with
code 1
.
Note that this function is only defined after you have run
eval "$(docopt "$@")"
, it is part of the docopt output.
Instead of inlining the entirety of the parser in your script, you can move the
static parts to an external file and only insert the dynamic part into your
script. This is particularly useful when you have multiple bash scripts in the
same project that use docopt.sh
. To generate the library run
docopt.sh generate-library > DEST
. The output is written to stdout
, so
make sure to add that redirect.
Once a library has been generated you can insert the dynamic part of your parser
into your script with docopt.sh --library DEST SCRIPT
. The generator will
then automatically add a source DEST
to the parser. Make sure to quote your
library path if it contains spaces like so
docopt.sh --library '"/path with spaces/docopt-lib.sh"'
. You do not need to
specify --library
on subsequent refreshes of the parser, docopt.sh
will
automatically glean the previously used parameters from your script and re-apply
them.
--library
can be any valid bash expression, meaning you can use things like
"$(dirname "$0")/docopt-lib-$v.sh"
(the $v
is the version of docopt
the parser AST was generated with).
On every invocation docopt checks that the library version and the version of the dynamic part in the script match. The parser exits with an error if that is not the case.
ATTENTION: The method outlined below relies on docopt.sh
being installed
and is only intended for development use, do not release any scripts that use
this method.
When developing a new script you might add, modify, and remove parameters quite
often. Having to refresh the parser with every change can quickly become
cumbersome and interrupt your workflow. To avoid this you can use the
--parser
flag to generate and then immediately eval
the output in your
script before invoking eval "$(docopt "$@")"
.
The script from the introduction would look like this (only
eval "$(docopt.sh --parser "$0")"
has been added):
#!/usr/bin/env bash
DOC="Naval Fate.
Usage:
naval_fate.sh <name> move <x> <y> [--speed=<kn>]
naval_fate.sh shoot <x> <y>
Options:
--speed=<kn> Speed in knots [default: 10]."
naval_fate() {
eval "$(docopt.sh --parser "$0")"
eval "$(docopt "$@")"
if $move; then
printf "The %s is now moving to %d,%d at %d knots.\n" "$_name_" "$_x_" "$_y_" "$__speed"
fi
if $shoot; then
printf "You shoot at %d,%d. It's a hit!\n" "$_x_" "$_y_"
fi
return 0
}
naval_fate "$@"
Since docopt.sh
is not patching the script, you also avoid any line number
jumps in your IDE. However, remember to replace this with the proper parser
before you ship the script.
You can turn of minifaction with -n 0
. This outputs the parser in its full
form. The parser and the generated AST code is heavily documented and includes
references to the analyzed DOC, showing what each part does.
e.g. docopt.sh -n 0 naval_fate.sh
#!/usr/bin/env bash
DOC="Naval Fate.
...
--speed=<kn> Speed in knots [default: 10]."
# docopt parser below, refresh this parser with `docopt.sh naval_fate.library.sh`
# shellcheck disable=2016,2086,2317
docopt() {
...
# This is the AST representing the parsed doc. The last node is the root.
# Options are first, as mentioned above. The comments above each node is
# shows what part of the DOC it is parsing (with line numbers).
# 03 naval_fate.sh <name> move <x> <y> [--speed=<kn>]
# ~~~~~~~
node_0(){
value __speed 0
}
# 03 naval_fate.sh <name> move <x> <y> [--speed=<kn>]
# ~~~~~~
node_1(){
value _name_ a
}
...
# Unset exported variables from parent shell
# that may clash with names derived from the doc
for varname in "${varnames[@]}"; do
unset "$p$varname"
done
# Assign internal varnames to output varnames and set defaults
eval $p'__speed=${var___speed:-10};'\
...
}
# docopt parser above, complete command for generating this parser is `docopt.sh --line-length=0 naval_fate.library.sh`
naval_fate() {
eval "$(docopt "$@")"
...
}
naval_fate "$@"
docopt.sh
uses pytest for testing. You can run the
testsuite by executing pytest
in the root of the project.
All
use
cases from the original docopt are used to validate correctness. Per default
pytest uses the bash version that is installed on the system to run the tests.
However, you can specify multiple alternate versions using
--bash-version <versions>
, where <versions>
is a comma-separated list of
bash versions (e.g. 3.2,4.0,4.1
). These versions need to be downloaded and
compiled first, which you can do with get_bash.py
. The script downloads,
extracts, configures, and compiles the specified bash versions in the
tests/bash-versions
folder. Use --bash-version all
to test with all the
bash versions that are installed.