right now the default address of the system bus is hardcoded, so if you change it in the global config file suddenly you have to set DBUS_SYSTEM_BUS_ADDRESS env variable. Might be nice if the client lib somehow read the config file, or if the bus on startup somehow wrote out its address to a well-known spot, but might also not be worth it.
we don't run filters on messages while blocking without entering the main loop, since filters are run as part of dbus_connection_dispatch(). This is probably a feature, as filters could create arbitrary reentrancy. But kind of sucks if you're trying to filter METHOD_RETURN for some reason.
FIXME what if we call out to application code to handle a message, holding the dispatch lock, and the application code runs the main loop and dispatches again? Probably deadlocks at the moment. Maybe we want a dispatch status of DBUS_DISPATCH_IN_PROGRESS, and then the GSource etc. could handle the situation? Right now our GSource is NO_RECURSE
this should be based on min header size plus some average body size, or something. Or rather, the min header size only, if we want to try to read only the header, store that in a DBusMessage, then read only the body and store that, etc., depends on how we optimize _dbus_message_loader_get_buffer() and what the exact message format is.
this function (or some lower-level non-convenience function) needs better error handling; should allow the application to distinguish between out of memory, and bad data from the remote app. It also needs to not leak a bunch of args when it gets to the arg that's bad, as that would be a security hole (allow one app to force another to leak memory)
DBusString needs a lot of cleaning up; some of the API is no longer used, and the API is pretty inconsistent. In particular all the "append" APIs, especially those involving alignment but probably lots of them, are no longer used by the marshaling code which always does "inserts" now.
avoid inserting the source into dest, then deleting the replaced chunk of dest (which creates a potentially large intermediate string). Instead, extend the replaced chunk of dest with padding to the same size as the source chunk, then copy in the source bytes.
We need to have a way to wake up the select sleep if a new iteration request comes in with a flag (read/write) that we're not currently serving. Otherwise a call that just reads could block a write call forever (if there are no incoming messages).