kjcolley7 / kjc_argparse Goto Github PK

The main feature this library is currently missing is proper support for defining subcommands, such as docker login or git pull. These subcommands should be defined similarly to arguments, and they need to have a recursive argparse context with support for their own arguments and help.

It would be nice to automatically support the --help argument without the programmer needing to manually declare an argument handler like:

ARG('h', "help", "Display this help message") {
    ARGPARSE_HELP();
    return 0;
}

The automatically defined one should register the long argument --help, but probably not the small argument -h. Perhaps it would also be good to keep the description set to NULL so it doesn't appear in help output. There should be a configuration parameter to disable auto help.

The current "test suite" consists of running an example program a bunch of times from a shell script with different arguments and then diffing the output against a file. I'd prefer to have a better testing framework to do unit testing in a way that is hopefully much better and cleaner than diffing files.

Mainly considering using https://github.com/sheredom/utest.h.

Using the docker login example, if I didn't set a custom usage string, docker login --help would show the usage like this (assuming argv[0] was docker):

Usage: docker [-hpu] [OPTIONS] COMMAND ...

Right now, ARGPARSE_HELP() has no way of knowing if it's being called from a subcommand.

There are a couple of approaches to solve this problem:

1. Add another configuration parameter to set the program name. Something like ARGPARSE_CONFIG_PROGNAME("docker login"); could be placed under ARG_COMMAND("login", "...") {.
2. When a subcommand is encountered and the child argparse context is created, it could have a link to the parent argparse context, along with the subcommand name that was used to enter this subcommand.

Option 1 is nice because it requires substantially fewer core argparse changes.

Option 2 is better for automatically handling many subcommands, even deeply nested ones (like docker builder imagetools inspect). It also has the benefit of producing the specific command prefix that was used to reach the argparse context. By that I mean in cases where there are multiple ways to invoke the same command (docker container ls, docker container list, docker container ps, docker ps).

When the -- argument is encountered, every subsequent argument should be passed to the ARG_POSITIONAL handler. This should be an automatic, built-in behavior. There should also be a configuration parameter to disable this functionality if the program author doesn't want it. This should be an easy change.