Add gio::Vfs subclass #1726

fbrouille · 2025-05-21T10:20:51Z

Allow custom implementation of gio::Vfs

gio/Gir.toml

gio/src/subclass/vfs.rs

gio/tests/vfs.rs

sdroege · 2025-05-25T07:11:29Z

Looks good to me except for the two open discussions about the string array.

glib/src/collections/strv.rs

sdroege · 2025-06-01T07:55:00Z

glib/src/collections/strv.rs

+/// It can be constructed safely from a `&StrV` and unsafely from a pointer to a C array.
+/// This type is very similar to `[GStringPtr]`, but with one added constraint: the underlying C array must be `NULL`-terminated.
+#[repr(transparent)]
+pub struct CStrV {


The name seems inconsistent with everything else. This is basically the String/PathBuf (StrV) vs. str/Path (CStrV) duality. Maybe this one should've actually been StrV and the other one StringV. Using Buf here seems wrong.

In other places (e.g. in gstreamer) there's a pattern of Foo and FooRef, so maybe we can just call this one StrVRef? @bilelmoussaoui what do you think?

Also GStringPtr is an ugly name, should probably reconsider that before the next release. Its difference with GStr is that it doesn't store the length and is equivalent with the raw C pointer. GStrLite or so... doesn't have to be solved as part of this PR but in case someone has ideas while we talk about naming anyway.

CStrV because of std::ffi::CStr. However I agree that name should be consistent with others gtk-rs types. I think also that CStrV should be StrV and StrV should be StringV or if you don't want breaking changes, maybe renameCStrV to GStrV and rename StrV to GStringV and create a type alias (deprecated) StrV = GStringV.

Anyway let me know for CStrV new name.

sdroege · 2025-06-01T08:47:32Z

gio/src/subclass/vfs.rs

+        self.parent_get_file_for_uri(uri)
+    }
+
+    unsafe fn get_supported_uri_schemes(&self) -> &'static CStrV {


Suggested change

unsafe fn get_supported_uri_schemes(&self) -> &'static CStrV {

fn get_supported_uri_schemes(&self) -> &'static CStrV {

no need to for this to be unsafe anymore (but also if it was necessary then the trait impl would have to be unsafe: unsafe trait is unsafe to implement, unsafe trait function is unsafe to call)

sdroege · 2025-06-01T08:49:29Z

gio/src/subclass/vfs.rs

+                    let _ = File::for_path("path");
+                });
+                glib::gobject_ffi::g_type_from_name("GLocalVfs".to_glib_none().0)


Not that it really matters because this is a test, but why not call g_type_from_name() also inside the Once (and make it a LazyLock)?

No it does not work. Calling File::for_pathis the only way to initialize GLocalVfs.

Whatever, I have removed binding to existing (private) gio types and refactor the tests.

I meant doing both the File::for_path() and g_type_from_name() inside the LazyLock (or OnceCell)

sdroege · 2025-06-01T08:49:56Z

gio/src/subclass/vfs.rs

+                static SUPPORTED_URI_SCHEMES: std::sync::OnceLock<glib::StrV> =
+                    std::sync::OnceLock::new();
+                SUPPORTED_URI_SCHEMES
+                    .get_or_init(|| glib::StrV::from(["myvfs"]))


Why not LazyLock?

Just to be consistent with initialization way of other static variables in the gtk-rs. LazyLock is not (yet) used in the project.

Whatever, I changed it.

Also seems fine to me. I think I should go over the other code and update it one of these days :) Thanks for the reminder!

sdroege · 2025-06-01T08:51:04Z

glib/src/collections/strv.rs

@@ -123,6 +123,20 @@ impl std::ops::Deref for StrV {
    }
 }

+impl AsRef<CStrV> for StrV {


Should probably also Deref into &CStrV (which in turn Deref into &[GStringPtr]).

sdroege · 2025-06-01T08:51:50Z

glib/src/collections/strv.rs

+    inner: [GStringPtr],
+}
+
+impl CStrV {


Should probably get all the same trait impls that StrV has (PartialEq, etc)

I tried to implement the same traits except Clone and Drop because CStrV does not own the null-terminated array.

This type represents a borrowed reference to a `NULL`-terminated array of `NULL`-terminated UTF-8 strings. Signed-off-by: fbrouille <[email protected]>

Signed-off-by: fbrouille <[email protected]>

fbrouille mentioned this pull request May 21, 2025

Add gio::File, gio::FileEnumerator, gio::FileMonitor, gio::Vfs subclasses #1676

Open

bilelmoussaoui reviewed May 21, 2025

View reviewed changes

gio/Gir.toml Outdated Show resolved Hide resolved

bilelmoussaoui reviewed May 21, 2025

View reviewed changes

gio/src/subclass/vfs.rs Show resolved Hide resolved

fbrouille force-pushed the gio_vfs_subclass branch from 63b4266 to 6427fa2 Compare May 22, 2025 16:09