Greg Stein <gst...@gmail.com> writes: > Seems like this will make things even more complicated. I'd be in > favor of *not* switching to NODES unless/until the op_depth is done > properly. If you switch early, then you're going to require another > format bump to reset all the op_depth fields to their proper values. I > don't think an early-switch to NODES before proper op_depth > computation will buy us anything.
There is an overhead to maintaining both sets of code. It's possible to build the wrong version, run regression tests with the wrong version, or break the other version. -- Philip
