I think YAML is leaving the spaces in there and passing ` build-essential` to the shell. There is no package with a space in the name, especially not a leading space.
I think I got the names from the Go source code, but the API emits JSON
that has these names instead. The api/swagger guide even says as much.
This caused the super fun quirk that the upload actually succeedes, but
the program reports an error condition because of the deserialization
failure. Time to bump a minor revision!
The function `std::fs::exists(...)` was stabilized in Rust 1.81, which
means it can't be used in the Debian Bookworm build. This patch swaps to
a compatible implementation leaning on the std::path::Path struct.
I'm both "upstreaming" a Debian-specific patch I had to make for the
package, and fixing the additional usage now in `main.rs`. There doesn't
seem to be any compelling reason to avoid using this function, so I
figure I should merge it into the base release.
I should have done this before publishing the build, but it's too late
now. Instead, I'll update it here and continue forward with gbp-dch to
help generate more correct changelogs.
I'm switching to use nested tags for the packaging branches. I want to
have a deb/bookworm and a deb/experimental branch. The gbp-pq tool
already seems to do this, so I'll follow it's lead.
The Attachment struct exists, but this makes it glaringly obvious that
I've made a bad interface. The create_release_attachment should only
accept one file at a time and the loop over all files should happen in
main.rs
I've changed the signature, removed the loops, and wired in the newer
error handling routines. Needs fixing at call sites.
I can't meaningfully unit test these things like this. I'll explore creating a tarball of a known Gitea configuration and using Docker to test against that. For now, just... kinda keep the test building.
I only dealt with the scenario where I get back an Err(_) value. This is
not what happens when the file is missing, it's what happens when there
was some other problem interacting with the file. Instead, an Ok(bool)
is returned to indicate positive identification of the presence or
absence of the file.
Something should be done about the std::io::error::Error that gets
handed back. Since I'm not expecting to see it very often, nor to really
do anything about it, I'm just emitting an error message and panicking.
There's a check in the create_release_attachment() function to ensure
the files exist before attempting to read them. Clearly, it isn't
working correctly.
I've dropped the `#[should_panic]` annotation and added a better Result
unpacking routine. It will report specific messages based on the result.
I'm not sure I want this code path to panic, actually... It should
probably return some error result so the app can print an error
diagonstic.
Oh well. That's a future problem. The end result is functionally
identical.
I was about to copy-paste the entire body of the attach_file_exists test
function into the attach_file_missing function. The only difference is
the file that they upload -- or don't, in the second case. I could make
a try-file-upload function and pass it many different files, but I don't
think I need that. Instead, I'll separate the test setup from the test
sequence itself.
I need some kind of automated testing, and Cargo has a test framework
built-in. This isn't the smartest test and depends upon another
test-less function.
Building a mocking environment is possible but means I'm trying to
emulate some behavior I don't fully understand -- that's why I need
tests in the first place. That path leads to tests for my tests and
that's stupid.
Instead, I'm having the test read config from the environment and hit a
real Gitea server. It's up to the user to ensure they have one.
I never got around to implementing the create-and-upload behavior for
the file attachment command. I'll do it later, but for now I'm disabling
the code to quiet down the compiler warnings.
Releases in a "pre-release" state are... weird. I'm not going to deal
with their oddities right now, so instead I'm dropping the command
option and hard-coding a `false` at it's usage site.
These arguments don't accept a value, they *are* the value. Presence or
absence tells the program everything it needs, so there's no point in
setting a default.
I've been putting this off because I don't want to have many small
formatting commits, and I don't have any kind of pre-commit hook to do
it for me. So here it is :v
This thing only needs to exist so Serde can decode the response JSON
properly. I'll either get a success, a missing token, *no body at all
which doesnt' make sense*, or some other ApiError text. In all cases,
the crate::Result is returned. This thing doesn't need to be public.
Now that all API functions are the new versions using my gt_tools::Error
type natively, I don't need to map from the reqwest::Error into the
wrapped variant.
