`clap_generate` originally intended to be "generate anything". With
`fig`, we already broke one part out. With #3174's man support, we are
also looking at keeping it separate:
- More freedom to iterate on the API
- Uniqueness (and potential weight) of its dependencies
- man generation is normally more for distribution while completions are
a mix of being distributed with the app or the app generating the
completions (which will be exacerbated if we move most completion
parsing logic to be in Rust)
So `clap_generate` is having a lot more limited of a role than the
original name conveys. I worry the generic name will be a hindrance to
people discovering and using it (yes, documentation can help but there
are limits).
I hesitated because we are on the verge of releasing 3.0. However, doing
it even later will be even more disruptive because more people will be
using it (crates.io lists ~70 people using `clap_generate`).
To ease things, we are still releasing `clap_generate` as a wrapper
around `clap_complete`.
`App::get_matches` lazily post-processes `App`s and `Arg`s so we don't
do it to subcommands that are never run (downside being people have to
exercise their full app to get debug_asserts).
`clap_generate` was only post-processing the top-level `App` and `Arg`s,
ignoring the sub-commands. In #2858, we noticed that `--version` was
being left in the completions instead of being removed during the
`_build` step. We would also have an incorrect `num_vals` and a host of
other problems.
This change adds a `App::_build_all` function for `clap_generate` to use
to eagerly build everything. By having it there, we make sure
everywhere that needs eager building, gets it (like some tests).
In `clap_generate::utils`, we add a unit test to ensure the subcommand's
`--version` was removed.
For some other tests specifying `.version()`, I added
`AppSettings::PropagateVersion` to make it behave more consistently.
The places I didn't were generally where the version was conditionally
set.
For `clap_generate/tests/generate_completions.rs`, I had to adjust the
`conflicts_with` because the subcommand was inheriting the argument with
it defined *but* the subcommand did not have the argument, tripping up a
debug assert.
Fixes#2860
Right now, adjusting test cases to cover some special edge-cases for a
specific shell is a little problematic.
For instance, if you add a special character that needs escaping
to one of the help strings, not only the tests for the shell you're
currently working on need adjustments, but rather all shell tests!
Contributors can often only work on one specific shell. The
need to adjust shells they don't know anything about discourages
testing or makes it simply impossible for a single contributor.
Arguably, this commit introduces a lot of code duplication, but on
the other hand it also makes testing a lot easier.
