On Nov 6, 2007, at 15:29 , dormando wrote:
How much work do you think it'd be to make the protocol more pluggable? I don't feel comfortable making a lot of these ideas part of the primary distribution. Given large install folks might try it then realize it is in fact incredibly slow.
There was talk about doing such things, but the code doesn't have very clean separation in that respect. The progress I made was very slight. I was hoping to avoid having much impact on the text protocol to add the binary.
But people want to try, and we could make it easier for them to maintain out of tree plugins. Brian's also been hinting at pluggable backend support (which looks fairly easy, lacking time).
It does sound like a decent goal. Of course, from where I stand, I'd rather get the two in and separate them first (having already done the first part).
Besides, having two distinct protocols to work with would help guide one through the separation while the tests validate nothing broke along the way.
Sorry for derailing the original topic, but this appears to be the problem worth discussing :) I agree with what Dustin/etc saying in that extending the binary protocol isn't a big deal. Making the commands all "meta" would overload the fact that the binary protocol is already very meta, and small commands just seem to be a better idea for maintenance.
And the idea of combining commands comes with some high costs. The combinations add up fast.
-- Dustin Sallings
