docopt.cpp: A C++11 Port

Contents

docopt creates beautiful command-line interfaces

Isn't it awesome how getopt (and boost::program_options for you fancy folk!) generate help messages based on your code?! These timeless functions have been around for decades and have proven we don't need anything better, right?

Hell no! You know what's awesome? It's when the option parser is generated based on the beautiful help message that you write yourself! This way you don't need to write this stupid repeatable parser-code, and instead can write only the help message--the way you want it.

docopt helps you create most beautiful command-line interfaces easily:

#include "docopt.h"

#include <iostream>

static const char USAGE[] =
R"(Naval Fate.

    Usage:
      naval_fate ship new <name>...
      naval_fate ship <name> move <x> <y> [--speed=<kn>]
      naval_fate ship shoot <x> <y>
      naval_fate mine (set|remove) <x> <y> [--moored | --drifting]
      naval_fate (-h | --help)
      naval_fate --version

    Options:
      -h --help     Show this screen.
      --version     Show version.
      --speed=<kn>  Speed in knots [default: 10].
      --moored      Moored (anchored) mine.
      --drifting    Drifting mine.
)";

int main(int argc, const char** argv)
{
    std::map<std::string, docopt::value> args
        = docopt::docopt(USAGE,
                         { argv + 1, argv + argc },
                         true,               // show help if requested
                         "Naval Fate 2.0");  // version string

    for(auto const& arg : args) {
        std::cout << arg.first <<  arg.second << std::endl;
    }

    return 0;
}

Beat that! The option parser is generated based on the docstring above that is passed to docopt::docopt function. docopt parses the usage pattern ("Usage: ...") and option descriptions (lines starting with a dash "-") and ensures that the program invocation matches the usage pattern; it parses options, arguments and commands based on that. The basic idea is that a good help message has all necessary information in it to make a parser.

Getting and using

To get docopt.cpp, the simplest is to use Conda:

conda install -c conda-forge docopt.cpp

Alternatively manual installation is done using (unix):

git clone
cmake .
make install

To link docopt.cpp, the simplest is to use CMake. The general structure of your CMakeLists.txt would be as follows:

cmake_minimum_required(VERSION 3.1)

project(example)

find_package(docopt COMPONENTS CXX REQUIRED)
include_directories(${DOCOPT_INCLUDE_DIRS})

add_executable(${PROJECT_NAME} ...)

target_link_libraries(${PROJECT_NAME} docopt)

C++11 port details

This is a port of the docopt.py module (https://github.com/docopt/docopt), and we have tried to maintain full feature parity (and code structure) as the original.

This port is written in C++11 and also requires a good C++11 standard library (in particular, one with regex support). The following compilers are known to work with docopt:

• Clang 3.3 and later
• GCC 4.9
• Visual C++ 2015 RC

GCC-4.8 can work, but the std::regex module needs to be replaced with Boost.Regex. In that case, you will need to define DOCTOPT_USE_BOOST_REGEX when compiling docopt, and link your code with the appropriated Boost libraries. A relatively recent version of Boost is needed: 1.55 works, but 1.46 does not for example.

This port is licensed under the MIT license, just like the original module. However, we are also dual-licensing this code under the Boost License, version 1.0, as this is a popular C++ license. The licenses are similar and you are free to use this code under the terms of either license.

The differences from the Python port are:

• the addition of a docopt_parse function, which does not terminate the program on error
• a docopt::value type to hold the various value types that can be parsed. We considered using boost::variant, but it seems better to have no external dependencies (beyond a good STL).
• because C++ is statically-typed and Python is not, we had to make some changes to the interfaces of the internal parse tree types.
• because std::regex does not have an equivalent to Python's regex.split, some of the regex's had to be restructured and additional loops used.

API

docopt::docopt(doc, argv, help /* =true */, version /* ="" */, options_first /* =false */)

docopt takes 2 required and 3 optional arguments:

• doc is a string that contains a help message that will be parsed to create the option parser. The simple rules of how to write such a help message are given in next sections. Here is a quick example of such a string (note that this example uses the "raw string literal" feature that was added to C++11):

R"(Usage: my_program [-hso FILE] [--quiet | --verbose] [INPUT ...]

-h --help    show this
-s --sorted  sorted output
-o FILE      specify output file [default: ./test.txt]
--quiet      print less text
--verbose    print more text
)"

• argv is a vector of strings representing the args passed. Although main usually takes a (int argc, const char** argv) pair, you can pass the value {argv+1, argv+argc} to generate the vector automatically. (Note we skip the argv[0] argument!) Alternatively you can supply a list of strings like { "--verbose", "-o", "hai.txt" }.

• help, by default true, specifies whether the parser should automatically print the help message (supplied as doc) and terminate, in case -h or --help option is encountered (options should exist in usage pattern, more on that below). If you want to handle -h or --help options manually (as other options), set help=false.

• version, by default empty, is an optional argument that specifies the version of your program. If supplied, then, (assuming --version option is mentioned in usage pattern) when parser encounters the --version option, it will print the supplied version and terminate. version could be any printable object, but most likely a string, e.g. "2.1.0rc1".

  Note, when docopt is set to automatically handle -h, --help and --version options, you still need to mention them in usage pattern for this to work (also so your users to know about them!)

• options_first, by default false. If set to true will disallow mixing options and positional argument. I.e. after first positional argument, all arguments will be interpreted as positional even if the look like options. This can be used for strict compatibility with POSIX, or if you want to dispatch your arguments to other programs.

The return value is a map<string, docopt::value> with options, arguments and commands as keys, spelled exactly like in your help message. Long versions of options are given priority. For example, if you invoke the top example as:

naval_fate ship Guardian move 100 150 --speed=15

the return dictionary will be:

{"--drifting": false,    "mine": false,
 "--help": false,        "move": true,
 "--moored": false,      "new": false,
 "--speed": "15",        "remove": false,
 "--version": false,     "set": false,
 "<name>": ["Guardian"], "ship": true,
 "<x>": "100",           "shoot": false,
 "<y>": "150"}

If any parsing error (in either the usage, or due to incorrect user inputs) is encountered, the program will exit with exit code -1.

Note that there is another function that does not exit on error, and instead will propagate an exception that you can catch and process as you like. See the docopt.h file for information on the exceptions and usage:

docopt::docopt_parse(doc, argv, help /* =true */, version /* =true */, options_first /* =false)

Help message format

Help message consists of 2 parts:

• Usage pattern, e.g.:

  Usage: my_program [-hso FILE] [--quiet | --verbose] [INPUT ...]
  

• Option descriptions, e.g.:

  -h --help    show this
  -s --sorted  sorted output
  -o FILE      specify output file [default: ./test.txt]
  --quiet      print less text
  --verbose    print more text
  

Their format is described below; other text is ignored.

Usage pattern format

Usage pattern is a substring of doc that starts with usage: (case insensitive) and ends with a visibly empty line. Minimum example:

"""Usage: my_program

"""

The first word after usage: is interpreted as your program's name. You can specify your program's name several times to signify several exclusive patterns:

"""Usage: my_program FILE
          my_program COUNT FILE

"""

Each pattern can consist of the following elements:

• <arguments>, ARGUMENTS. Arguments are specified as either upper-case words, e.g. my_program CONTENT-PATH or words surrounded by angular brackets: my_program <content-path>.
• --options. Options are words started with dash (-), e.g. --output, -o. You can "stack" several of one-letter options, e.g. -oiv which will be the same as -o -i -v. The options can have arguments, e.g. --input=FILE or -i FILE or even -iFILE. However it is important that you specify option descriptions if you want your option to have an argument, a default value, or specify synonymous short/long versions of the option (see next section on option descriptions).
• commands are words that do not follow the described above conventions of --options or <arguments> or ARGUMENTS, plus two special commands: dash "-" and double dash "--" (see below).

Use the following constructs to specify patterns:

• [ ] (brackets) optional elements. e.g.: my_program [-hvqo FILE]
• ( ) (parens) required elements. All elements that are not put in [ ] are also required, e.g.: my_program --path=<path> <file>... is the same as my_program (--path=<path> <file>...). (Note, "required options" might be not a good idea for your users).
• | (pipe) mutually exclusive elements. Group them using ( ) if one of the mutually exclusive elements is required: my_program (--clockwise | --counter-clockwise) TIME. Group them using [ ] if none of the mutually exclusive elements are required: my_program [--left | --right].
• ... (ellipsis) one or more elements. To specify that arbitrary number of repeating elements could be accepted, use ellipsis (...), e.g. my_program FILE ... means one or more FILE-s are accepted. If you want to accept zero or more elements, use brackets, e.g.: my_program [FILE ...]. Ellipsis works as a unary operator on the expression to the left.
• [options] (case sensitive) shortcut for any options. You can use it if you want to specify that the usage pattern could be provided with any options defined below in the option-descriptions and do not want to enumerate them all in usage-pattern.
• "[--]". Double dash "--" is used by convention to separate positional arguments that can be mistaken for options. In order to support this convention add "[--]" to your usage patterns.
• "[-]". Single dash "-" is used by convention to signify that stdin is used instead of a file. To support this add "[-]" to your usage patterns. "-" acts as a normal command.

If your pattern allows to match argument-less option (a flag) several times:

Usage: my_program [-v | -vv | -vvv]

then number of occurrences of the option will be counted. I.e. args['-v'] will be 2 if program was invoked as my_program -vv. Same works for commands.

If your usage pattern allows to match same-named option with argument or positional argument several times, the matched arguments will be collected into a list:

Usage: my_program <file> <file> --path=<path>...

I.e. invoked with my_program file1 file2 --path=./here --path=./there the returned dict will contain args['<file>'] == ['file1', 'file2'] and args['--path'] == ['./here', './there'].

Option descriptions format

Option descriptions consist of a list of options that you put below your usage patterns.

It is necessary to list option descriptions in order to specify:

• synonymous short and long options,
• if an option has an argument,
• if option's argument has a default value.

The rules are as follows:

• Every line in doc that starts with - or -- (not counting spaces) is treated as an option description, e.g.:

  Options:
    --verbose   # GOOD
    -o FILE     # GOOD
  Other: --bad  # BAD, line does not start with dash "-"
  

• To specify that option has an argument, put a word describing that argument after space (or equals "=" sign) as shown below. Follow either <angular-brackets> or UPPER-CASE convention for options' arguments. You can use comma if you want to separate options. In the example below, both lines are valid. However you are recommended to stick to a single style.:

  -o FILE --output=FILE       # without comma, with "=" sign
  -i <file>, --input <file>   # with comma, without "=" sing
  

• Use two spaces to separate options with their informal description:

  --verbose More text.   # BAD, will be treated as if verbose option had
                         # an argument "More", so use 2 spaces instead
  -q        Quit.        # GOOD
  -o FILE   Output file. # GOOD
  --stdout  Use stdout.  # GOOD, 2 spaces
  

• If you want to set a default value for an option with an argument, put it into the option-description, in form [default: <my-default-value>]:

  --coefficient=K  The K coefficient [default: 2.95]
  --output=FILE    Output file [default: test.txt]
  --directory=DIR  Some directory [default: ./]
  

• If the option is not repeatable, the value inside [default: ...] will be interpreted as string. If it is repeatable, it will be split into a list on whitespace:

  Usage: my_program [--repeatable=<arg> --repeatable=<arg>]
                       [--another-repeatable=<arg>]...
                       [--not-repeatable=<arg>]
  
  # will be ['./here', './there']
  --repeatable=<arg>          [default: ./here ./there]
  
  # will be ['./here']
  --another-repeatable=<arg>  [default: ./here]
  
  # will be './here ./there', because it is not repeatable
  --not-repeatable=<arg>      [default: ./here ./there]
  

Examples

We have an extensive list of examples which cover every aspect of functionality of docopt. Try them out, read the source if in doubt.

There are also very interesting applications and ideas at that page. Check out the sister project for more information!

Subparsers, multi-level help and huge applications (like git)

If you want to split your usage-pattern into several, implement multi-level help (with separate help-screen for each subcommand), want to interface with existing scripts that don't use docopt, or you're building the next "git", you will need the new options_first parameter (described in API section above). To get you started quickly we implemented a subset of git command-line interface as an example: examples/git

Compiling the example / Running the tests

The original Python module includes some language-agnostic unit tests, and these can be run with this port as well.

The tests are a Python driver that uses the testcases.docopt file to then invoke a C++ test case runner (run_testcase.cpp):

$ clang++ --std=c++11 --stdlib=libc++ docopt.cpp run_testcase.cpp -o run_testcase
$ python run_tests.py
PASS (175)

You can also compile the example shown at the start (included as example.cpp):

$ clang++ --std=c++11 --stdlib=libc++ -I . docopt.cpp examples/naval_fate.cpp -o naval_fate
$ ./naval_fate --help
 [ ... ]
$ ./naval_fate ship Guardian move 100 150 --speed=15
--drifting: false
--help: false
--moored: false
--speed: "15"
--version: false
<name>: ["Guardian"]
<x>: "100"
<y>: "150"
mine: false
move: true
new: false
remove: false
set: false
ship: true
shoot: false

Development

Comments and suggestions are very welcome! If you find issues, please file them and help improve our code!

Please note, however, that we have tried to stay true to the original Python code. If you have any major patches, structural changes, or new features, we might want to first negotiate these changes into the Python code first. However, bring it up! Let's hear it!

Changelog

docopt follows semantic versioning. The first release with stable API will be 1.0.0 (soon).

• 0.6.2 Bugfix release (still based on docopt 0.6.1)
• 0.6.1 The initial C++ port of docopt.py (based on docopt 0.6.1)