Comment on attachment 102125 [details] [review]
documentation patch for dbus-glib

Review of attachment 102125 [details] [review]:
-----------------------------------------------------------------

> A GByteArray is implemented as GArray with element size 1

It might have the same in-memory layout, but it isn't the same GType; and I don't see anything in the GLib documentation that guarantees that you can pass a GArray* containing guchar where a GByteArray* is expected, or vice versa. So I'm reluctant to apply this.

(Also, I would actually prefer people to think "dbus-glib is awful, I'm going to stop using it"; because it is, in many other ways. Use GDBus if at all possible.)

(In reply to comment #1)
> Comment on attachment 102125 [details] [review] [review]
> documentation patch for dbus-glib
> 
> Review of attachment 102125 [details] [review] [review]:
> -----------------------------------------------------------------
> 
> > A GByteArray is implemented as GArray with element size 1
> 
> It might have the same in-memory layout, but it isn't the same GType; and I
> don't see anything in the GLib documentation that guarantees that you can
> pass a GArray* containing guchar where a GByteArray* is expected, or vice
> versa. So I'm reluctant to apply this.

Correct. But GByteArray and GArray are not GObjects. Having a gpointer to such instances you must already know what you have at hand. E.g. you cannot dereference the pointer to check whether you have a G_TYPE_ARRAY or G_TYPE_BYTE_ARRAY. Hence that point doesn't matter much.


Note that glib states:

 * #GByteArray is a mutable array of bytes based on #GArray, to provide arrays
 * of bytes which grow automatically as elements are added.

also note:

GByteArray*
g_byte_array_new (void)
{
  return (GByteArray *)g_array_sized_new (FALSE, FALSE, 1, 0);
}


I would argue that glib will never change this to (re)implement GByteArray not to be effectively a GArray. Doing so might break many users. Which would be bad even when arguing that they should not rely on non-guaranteed implementation details.


> (Also, I would actually prefer people to think "dbus-glib is awful, I'm
> going to stop using it"; because it is, in many other ways. Use GDBus if at
> all possible.)

Ok.


Anyway. Feel free to reject the patch and close the bug.
Thanks.

(In reply to Thomas Haller from comment #2)
> I would argue that glib will never change this to (re)implement GByteArray
> not to be effectively a GArray. Doing so might break many users. Which would
> be bad even when arguing that they should not rely on non-guaranteed
> implementation details.

Sorry, I am not convinced by this reasoning - I think it would be entirely valid for the GLib maintainers to apply that argument, so I'm not going to apply this patch unless/until the GLib documentation explicitly says that GByteArray* and (GArray-of-byte)* can be used interchangeably.

(In reply to Thomas Haller from comment #2)
> Anyway. Feel free to reject the patch and close the bug.

Done.