double marshalling ...

[Michael Meeks]

Hi Havoc,

On Wed, 2004-04-14 at 23:39 -0400, Havoc Pennington wrote:
> It isn't immediately obvious to me what you're suggesting here, maybe
> give me more concrete pointers to what old and new code would be like?

	Well - it's a minor issue but our marshalling code has:

	if (msg_endianness == FOO_ENDIANNESS)
		marshal_one_way;
	else
		marshal_another_way;

	[ in simple terms ]. The ORB misses most of this out (with it's
branching penalty) - since it is assumed that we are always marshalling
in the host byte order; I was suggesting switching to this sort of
scheme - however, it's clearly useful to be able to marshal in both byte
orders for the dbus-message-builder - so I was wondering about moving
the endianness switching code to there instead - not really a big win; I
think we can get a nicer win from genericising the (de)marshalling code,
virtualising it, then doing a per-message endianness split early-on; but
of course inching towards that will take a little while. [ I have a
sizeable code-reduction / genericisation patch in progress here ].

	Hmm,

		Michael.

-- 
 michael at ximian.com  <><, Pseudo Engineer, itinerant idiot