`gio::SubprocessLauncher::set_environ` and `gio::ApplicationCommandLine::environ` don't compose well

[pjungkamp]

I noticed that gio::ApplicationCommandLine::environ returns a Vec<OsString> and that gio::SubprocessLauncher::set_environ only takes a &[&Path] as argument.

This seems rather unfortunate. My application should share a primary instance for multiple windows, but still needs to spawn processes with the environment that spawned a new window on the primary instance. This currently needs an intermediate collection to collect the Path references that can be passed to gio::SubprocessLauncher::set_environ.

let cmd: gio:ApplicationCommandLine; // from the parameter of the `gio::Application::command_line` signal
let launcher = gio::SubprocessLauncher::new();

launcher.set_environ(cmd.environ().iter().map(<OsString as AsRef<Path>>::as_ref).collect::<Vec<&Path>>().as_slice());

---

This could be generic over all ToGlibContainerFromSlice types that coerce properly like &OsStr, &str, &Path and their owned variants OsString, String, PathBuf.

I think an intermediary trait for all kinds of OsStr-like values could help to make the gio::SubprocessLauncher::set_environ method more usable.

Maybe something like fn set_environ<'a>(&self, envs: &'a [impl AsOsStr<'a>])? Where AsOsStr<'a> is a subtrait of glib::translate::ToGlibContainerFromSlice<'a, *const *const i8> so the .to_glib_none() in the generated bindings still works?

trait AsOsStr<'a> : glib::translate::ToGlibContainerFromSlice<'a, *const *const i8> {}
impl<'a> AsOsStr<'a> for T where T: AsRef<OsStr> + glib::translate::ToGlibContainerFromSlice<'a, *const *const i8> {}

On a side-note: Why doesn't CString and &CStr implement ToGlibContainerFromSlice?

---

It's just a minor inconvenience but I wanted to start some discussion on these OsStr-like slices in function arguments.

[sdroege]

I noticed that gio::ApplicationCommandLine::environ returns a Vec and that gio::SubprocessLauncher::set_environ only takes a &[&Path] as argument.

It should really take an &OsStr instead of &Path, that would be the first problem here :)

And then this could simply become a &[&impl AsRef<Path>] or not? Do you want to give that a try?

Why doesn't CString and &CStr implement ToGlibContainerFromSlice?

Probably just forgotten / wasn't needed yet. Do you want to add that?

[pjungkamp]

Why doesn't CString and &CStr implement ToGlibContainerFromSlice?

Probably just forgotten / wasn't needed yet. Do you want to add that?

Just opened PR #1641.

[pjungkamp]

And then this could simply become a &[&impl AsRef] or not? Do you want to give that a try?

The problem is that &[&impl AsRef<OsStr>] or &[impl AsRef<Path>] do not implement ToGlibPtr. This is only implemented for slices where the element type implements ToGlibContainerFromSlice. But AsRef<OsStr> and AsRef<Path> are not strict supertraits of ToGlibContainerFromSlice and thus the &[&impl AsRef<OsStr>] or &[&impl AsRef<Path>] slices do not implement ToGlibPtr for the ffi translation.

[pjungkamp]

Here's the Rust error message:

error[E0599]: the method `to_glib_none` exists for reference `&[impl AsRef<OsStr>]`, but its trait bounds were not satisfied
  --> gio/src/subprocess_launcher.rs:25:79
   |
25 |             ffi::g_subprocess_launcher_set_environ(self.to_glib_none().0, env.to_glib_none().0);
   |                                                                               ^^^^^^^^^^^^ method cannot be called on `&[impl AsRef<OsStr>]` due to unsatisfied trait bounds
   |
   = note: the following trait bounds were not satisfied:
           `impl AsRef<OsStr>: glib::translate::ToGlibContainerFromSlice<'_, _>`
           which is required by `[impl AsRef<OsStr>]: glib::translate::ToGlibPtr<'_, _>`
           `[impl AsRef<OsStr>]: glib::translate::ToGlibPtr<'_, _>`
           which is required by `&[impl AsRef<OsStr>]: glib::translate::ToGlibPtr<'_, _>`

I've found another place that takes a slice of environment variables at environ_getenv.

gtk-rs-core/glib/Gir.toml

Lines 356 to 359 in 35e15b7

	[[object.function]]
	name = "environ_getenv"
	# manual input &[OsString]
	ignore = true

So maybe just use &[OsString] for set_environ to be consistent?

[sdroege]

So maybe just use &[OsString] for set_environ to be consistent?

Yes that probably makes most sense for now