Mark void pointers as gpointer

[thiblahute]

Otherwise the ABI is not properly respected as for some reason, mem::size_of::<c_void>() returns 1 on a 64bits system. In the case of GstVideoDecoder(Class), the ABI mistmatch because of the last void * padding[14] field which is translated to pub padding: [c_void; 14] which size was 14(bytes) instead of 112 (on 64bits systems).

[sdroege]

Yes they have to be *mut c_void aka gpointer

[sdroege]

I'm not sure the change is correct though. It is detected as a pointer already in the code you change, but that knowledge disappears somewhere along the way.

[thiblahute]

I'm not sure the change is correct though. It is detected as a pointer already in the code you change, but that knowledge disappears somewhere along the way.

I am not sure what you mean here :-)

[EPashkin]

Seems this really not right: gio-sys:

pub struct GInputStreamClass {
     pub parent_class: gobject::GObjectClass,
-    pub read_fn: Option<unsafe extern "C" fn(*mut GInputStream, *mut c_void, size_t, *mut GCancellable, *mut *mut glib::GError) -> ssize_t>,
+    pub read_fn: Option<unsafe extern "C" fn(*mut GInputStream, *mut gpointer, size_t, *mut GCancellable, *mut *mut glib::GError) -> ssize_t>,
     pub skip: Option<unsafe extern "C" fn(*mut GInputStream, size_t, *mut GCancellable, *mut *mut glib::GError) -> ssize_t>,

glib-sys

-    pub fn g_once_init_enter(location: *mut c_void) -> gboolean;
-    pub fn g_once_init_leave(location: *mut c_void, result: size_t);
+    pub fn g_once_init_enter(location: *mut gpointer) -> gboolean;
+    pub fn g_once_init_leave(location: *mut gpointer, result: size_t);

gtk-sys

pub struct GtkFileChooserButtonClass {
     pub parent_class: GtkBoxClass,
     pub file_set: Option<unsafe extern "C" fn(*mut GtkFileChooserButton)>,
-    pub __gtk_reserved1: *mut c_void,
-    pub __gtk_reserved2: *mut c_void,
-    pub __gtk_reserved3: *mut c_void,
-    pub __gtk_reserved4: *mut c_void,
+    pub __gtk_reserved1: *mut gpointer,
+    pub __gtk_reserved2: *mut gpointer,
+    pub __gtk_reserved3: *mut gpointer,
+    pub __gtk_reserved4: *mut gpointer,
 }

etc.

And this code don't changed

pub struct GtkAppChooserButtonClass {
    pub parent_class: GtkComboBoxClass,
    pub custom_item_activated: Option<unsafe extern "C" fn(*mut GtkAppChooserButton, *const c_char)>,
    pub padding: [gpointer; 16],
}

as it have right type in gir, and other paddings in gtk too.

      <field name="padding" readable="0" private="1">
        <array zero-terminated="0" c:type="gpointer" fixed-size="16">
          <type name="gpointer" c:type="gpointer"/>
        </array>
      </field>

[thiblahute]

Interesting, in Gst it is doing the right thing, but I clearly missed something, I will fix.

[sdroege]

@thiblahute any news on this?

[ghost]

gir for GstVideoDecoderClass looks as follows:

    <record name="VideoDecoderClass"
            c:type="GstVideoDecoderClass"
            glib:is-gtype-struct-for="VideoDecoder">
      ...
      <field name="padding" readable="0" private="1">
        <array zero-terminated="0" c:type="void" fixed-size="14">
          <type name="gpointer" c:type="void*"/>
        </array>
      </field>
    </record>

The value of array@c:type doesn't seem to make sense, and this is the one that is used to determine how many pointers stick there. You can't also simply fix this in codegen by swapping those two around as the second c:type is not retained from parsing stage (neither would it be correct in general).

Furthermore, after investigating this further and cross checking with C, I am even more confused. The values of array@c:type and element type@c:type values are so inconsistent. A few related bugzilla items:

• https://bugzilla.gnome.org/show_bug.cgi?id=756122
• https://bugzilla.gnome.org/show_bug.cgi?id=792275
• https://bugzilla.gnome.org/show_bug.cgi?id=755234
• https://bugzilla.gnome.org/show_bug.cgi?id=646080

[sdroege]

See https://github.com/gtk-rs/gir/issues/534#issuecomment-363826593

Not a problem for this specific case anymore