I wrote:
> First cut [of having predicate locks linked to later row versions
> for conflict detection purposes]
> It passes all the usual regression tests, the new isolation tests,
> and the example posted earlier in the thread of a test case which
> was allowing an anomaly. (That is to say, the anomaly is now
> prevented.)
>
> I didn't get timings, but it *seems* noticeably slower; hopefully
> that's either subjective or fixable. Any feedback on whether this
> seems a sane approach to the issue is welcome.
After sleeping on it and reviewing the patch this morning, I'm
convinced that this is fundamentally right, but the recursion in
RemoveTargetIfNoLongerUsed() can't work. I'll have to just break the
linkage there and walk the HTAB in the general cleanup to eliminate
orphaned targets.
I'm working on it now, and hope to have it all settled down today. I
fear I've messed up Heikki's goal of a commit this weekend, but I
think this is a "must fix".
-Kevin