On Tue, 2005-02-15 at 15:38 -0500, Matthias Clasen wrote: >Trying to attack the introspection problem from the api side. >Here is an initial api proposal for an interface repository. >I imagine this would be implemented on top of metadata blobs >which are mmapped readonly, thus the various *Def objects can >be very lightweight, essentially just a pointer to the metadata >plus an offset. > >Does this look reasonable ? (except for the names, which are > completely ad hoc) One thing also that I would also like is a way to sensibly convert a GError domain/code into a string. For example, mapping G_FILE_ERROR/G_FILE_ERROR_EXISTS into a DBus error message named "org.gtk.GLib.FileError.Exists" that callers can dispatch on. GError is a bit odd I guess because it isn't necessarily strongly associated with a particular class. I guess the other approach is to send the domain as a string and the raw code, and expect the caller to somehow have the requisite data to convert it back, but that seems wrong. For example if you have a Qt app that doesn't link to GTK+ invoking a D-BUS method of a GNOME app, it's not going to have access to the quarks etc.
Attachment:
signature.asc
Description: This is a digitally signed message part