interfaces, object paths and activation

Mark McLoughlin
Sun, 21 Sep 2003 19:45:47 +0100


On Sun, 2003-09-21 at 19:36, Havoc Pennington wrote:
> On Sun, 2003-09-21 at 14:22, Mark McLoughlin wrote:
> > > One question I had - why align the field name to 4 byte boundaries?
> > 
> > 	Yeah, good question. The old code did it this way too so I just made it
> > explicit. But sure, there isn't a good reason from a protocol point of
> > view to do it this way. I think, the only problem this will present is
> > making the field deletion logic cope with this ... but it might not be
> > such a bad idea to re-think that code anyway.
> > 
> > 	Will I go ahead and do that, then ?
> The general rule has been to align everything naturally, so e.g. strings
> and bytes are aligned 1, 32-bit ints aligned 4, doubles aligned 8, etc.
> It makes sense to me to keep it that way, so a 1-byte field name would
> be aligned 1. If you want to code it up I'd be in favor, could save us
> some bytes and keep things consistent.

	Okay, I'll get onto that.

> If you want to go ahead and commit either with that change or without
> but with a doc/TODO entry for that change, that would be fine. Just be
> sure "make check" passes (and if you think any new corner cases have
> been introduced, add new test messages for them).
> Also remove the doc/TODO entry for this patch (unless I haven't
> committed it yet, which is possible; I have a big patch outstanding).

	I've committed to avoid conflicts with your patch (let you sort them
out :-) and added the TODO item. make check already passes and the only
new corner cases that I can think of will be when we remove the 4 byte
alignment requirement.