Hi! So I've finally got to try the thing we have briefly discussed at RustConf: removing duplication of docopt options in different commands. This PR is a rough proof of concept to gather feedback and decide if we need this at all.

The approach here is to use hand-rolled code generation: all commands are specified in options.txt file, from which options.rs with docopt usage strings and options strutcs is generated. In options.txt, we can defined option groups and then use them to assemble full docopt description of the command.

I've decided to go with code generation and not with runtime string concatenation for the following reasons:

• In some sense, it is simpler: to understand the end result, you need to look only at options.rs, which contains usual self-describing docopt usages and structs. You don't need to understand the tricky string concatenation stuff.

• We can generate not only the usage strings, but also the structs with options. I think this can be extended to also generate bash completions.

• This can be extended to generate structs for different option groups, so instead of config.configure(options.flag_verbose, options.flag_quiet, &options.flag_color, options.flag_frozen, options.flag_locked)?; one can write config.configure(options.color_options(), options.lockfile_options()).

So that's the plan :)

Current implementation applies it only to build and publish commands, to show what's possible without creating a huge diff. Generation is done with a separate binary generate-options.rs and not via build.rs just because it was easier to debug. However I think it maybe better to avoid build.rs and just to commit the generated options.rs together with options.txt: the generated code is supposed to be very readable, and you might want to inspect it to see if the help message is wrapped properly, for example.

generate-options.rs is very rough: lots of unwraps and does not handle multiline flag descriptions :)