Argument Aggregator {#mainpage}
===================

This is the Doxygen documentation for Argument Aggregator, a simple C++11 argument parser.

To use just create an argagg::parser object. However, the struct doesn't provide any explicit methods for defining flags. Instead we define the flags using initialization lists.

    argagg::parser argparser {{
        { "help", {"-h", "--help"},
          "shows this help message", 0},
        { "delim", {"-d", "--delim"},
          "delimiter (default: ,)", 1},
        { "num", {"-n", "--num"},
          "number", 1},
      }};

An option is specified by four things: the name of the option, the strings that activate the option (flags), the option's help message, and the number of arguments the option expects.

With the parser defined you actually parse the arguments by calling the argagg::parser::parse() method. If there are any problems then an exception is thrown.

    argagg::parser_results args;
    try {
      args = argparser.parse(argc, argv);
    } catch (const std::exception& e) {
      std::cerr << e.what() << std::endl;
      return EXIT_FAILURE;
    }

You can check if an option shows up in the command line arguments by accessing the option by name from the parser results and using the implicit boolean conversion. You can write out a simplistic option help message by streaming the argagg::parser instance itself.

    if (args["help"]) {
      std::cerr << argparser;
      //     -h, --help
      //         shows this help message
      //     -d, --delim
      //         delimiter (default: ,)
      //     -n, --num
      //         number
      return EXIT_SUCCESS;
    }

That help message is only for the flags. If you want a usage message it's up to you to provide it.

    if (args["help"]) {
      std::cerr << "Usage: program [options] ARG1 ARG2" << std::endl
                << argparser;
      // Usage: program [options] ARG1 ARG2
      //     -h, --help
      //         shows this help message
      //     -d, --delim
      //         delimiter (default: ,)
      //     -n, --num
      //         number
      return EXIT_SUCCESS;
    }

A special output stream, argagg::fmt_ostream, is provided that will run the usage and help through `fmt` for nice word wrapping (see `./examples/joinargs.cpp` for a better example).

    if (args["help"]) {
      argagg::fmt_ostream fmt(std::cerr);
      fmt << "Usage: program [options] ARG1 ARG2" << std::endl
          << argparser;
      return EXIT_SUCCESS;
    }

Generally argagg tries to do a minimal amount of work to leave most of the control with the user.

If you want to get an option argument but fallback on a default value if it doesn't exist then you can use the argagg::option_results::as() API and provide a default value.

    auto delim = args["delim"].as<std::string>(",");

If you don't mind being implicit an implicit conversion operator is provided allowing you to write simple assignments.

    int x = 0;
    if (args["num"]) {
      x = args["num"];
    }

Finally, you can get all of the positional arguments as an std::vector using the argagg::parser_results::pos member. You can alternatively convert individual positional arguments using the same conversion functions as the option argument conversion methods.

    auto y = 0.0;
    if (args.pos.size() > 0) {
      y = args.as<double>(0);
    }

One can also specify `--` on the command line in order to treat all following arguments as not options.

For a more detailed treatment take a look at the examples or test cases.

Mental Model
------------

The parser just returns a structure of pointers to the C-strings in the original `argv` array. The @ref argagg::parser::parse() method returns a @ref argagg::parser_results object which has two things: position arguments and option results. The position arguments are just a @ref std::vector of `const char*`. The option results are a mapping from option name (@ref std::string) to @ref argagg::option_results objects. The @ref argagg::option_results objects are just an @ref std::vector of @ref argagg::option_result objects. Each instance of an @ref argagg::option_result represents the option showing up on the command line. If there was an argument associated with it then the @ref argagg::option_result::arg member will *not* be `nullptr`.

Consider the following command:

    gcc -g -I/usr/local/include -I. -o test main.o foo.o -L/usr/local/lib -lz bar.o -lpng

This would produce a structure like follows, written in psuedo-YAML, where each string is actually a `const char*` pointing to some part of a string in the original `argv` array:

    parser_results:
      program: "gcc"
      pos: ["main.o", "foo.o", "bar.o"]
      options:
        version:
        debug:
          all:
          - arg: null
        include_path:
          all:
          - arg: "/usr/local/include"
          - arg: "."
        library_path:
          all:
          - arg: "/usr/local/lib"
        library:
          all:
          - arg: "z"
          - arg: "png"
        output:
          all:
          - arg: "test"

Conversion to types occurs at the very end when the `as<T>()` API is used. Up to that point `argagg` is just dealing with C-strings.

Installation
------------

There is just a single header file (`argagg.hpp`) so you can copy that whereever you want. If you want to properly install it you can use the CMake script. The CMake script exists primarily to build the tests and documentation, but an install target for the header is provided.

The standard installation dance using CMake and `make` is as follows:

    mkdir build
    cd build
    cmake -DCMAKE_INSTALL_PREFIX=/usr/local ..
    make install
    ctest -V # optionally run tests

Override [`CMAKE_INSTALL_PREFIX`](https://cmake.org/cmake/help/v2.8.12/cmake.html#variable:CMAKE_INSTALL_PREFIX) to change the installation location. By default (on UNIX variants) it will install to `/usr/local` resulting in the header being copied to `/usr/local/include/argagg/argagg.hpp`.

If you have [Doxygen](http://www.stack.nl/~dimitri/doxygen/) it should build and install documentation as well.

There are no dependencies other than the standard library.

Edge Cases
----------

There are some interesting edge cases that show up in option parsing. I used the behavior of `gcc` as my target reference in these cases.

### Greedy Arguments

Remember that options that require arguments will greedily process arguments.

Say we have the following options: `-a`, `-b`, `-c`, and `-o`. They all don't accept arguments except `-o`. Below is a list of permutations for short flag grouping and the results:

- `-abco foo`: `-o`'s argument is `foo`
- `-aboc foo`: `-o`'s argument is `c`, `foo` is a positional argument
- `-aobc foo`: `-o`'s argument is `bc`, `foo` is a positional argument
- `-oabc foo`: `-o`'s argument is `abc`, `foo` is a positional argument

For whitespace delimited arguments the greedy processing means the next argument element (in `argv`) will be treated as an argument for the previous option, regardless of whether or not it looks like a flag or some other special entry. That means you get behavior like below:

- `--output=foo -- --bar`: `--output`'s argument is `foo`, `--bar` is a positional argument
- `--output -- --bar`: `--output`'s argument is `--`, `--bar` is treated as a flag
- `--output --bar`: `--output`'s argument is `--bar`