We probably should look into removing `xtask codegen` altogether. The
test workflow works perfectly for package.json config.
There are two things preventing that:
* Lint completions are generated on demand.
* Docs are not committed to the repository.
The idea here is that, on CI, we only want to cache crates.io
dependencies, and not local crates. This keeps the size of the cache
low, and also improves performance, as network and moving files on
disk (on Windows) can be slow.
This builds on #2231 but was actually done before that. You see, the
cause for #2231 was that I got this error message:
Error: Error { kind: Io(Os { code: 2, kind: NotFound, message: "No such file or directory" }) }
Just switching to `anyhow::Result` got me stack traces (when setting
`RUST_LIB_BACKTRACE=1`) that at least showed
stack backtrace:
0: std::backtrace::Backtrace::create
1: std::backtrace::Backtrace::capture
2: anyhow::error::<impl core::convert::From<E> for anyhow::Error>::from
3: xtask::install_server
4: xtask::install
5: xtask::main
6: std::rt::lang_start::{{closure}}
7: std::panicking::try::do_call
8: __rust_maybe_catch_panic
9: std::rt::lang_start_internal
10: std::rt::lang_start
11: main
With the added contexts (not at all exhaustive), the error became
Error: install server
Caused by:
0: build AutoCfg with target directory
1: No such file or directory (os error 2)
Since anyhow is such a small thing (no new transitive dependencies!),
and in general gives you `Result<T, Box<dyn Error>>` on steroids, I
think this a nice small change. The only slightly annoying thing was to
replace all the `Err(format!(…))?` calls (haven't even looked at whether
we can make it support wrapping strings though), but the `bail!` macro
is shorter anyway :)
