Stuff Michael Meeks is doing

Up late, checked mail. Corresponded with my friend Torsten who seems to be finding most of the holes in bonobo with unerring accuracy; excellent chap.
Committed updated docs with the convention for reference counting in Bonobo projects. Did some time trials on my totaly un- optimized new stub code, having had the dire performance warnings, I expected it to be ~2 times slower, steeled myself for the results: only 20% slower, wow. The miracle of cache locality, branch prediction etc. messing with the traditional size / speed tradeoff.
Part way through writing the ORBit orbit-idl backend I tested for size savings, a little more needs to go in but I got 150K stubs for Bonobo ( and most of this will move to common ), or 3.6K stripped compared with 440K stubs normally which are 74Kb when stripped. Also a far faster build time, with no warnings.
Realized looking at my code that the type information I write out could, with a little effort, be turned into a type database that could be used instead of parsing idl and about 1 / 5th of the size. I forget what a type database is actualy useful for though, but it would certainly accelerate and simplify the process of writing language bindings I think. After adding all the exception and method information it is now 9Kb stripped; hmm.
Spent a long time writing chunks of the idl compiler backend, joyfully discarding scads of the code that was there originaly.