Gryadka is not Paxos, so it's probably wrong (RETRACTED)

Leslie Lamport purportedly once said (though I can’t prove it):

If it’s not Paxos, it’s probably wrong.

In the original version of this post, I claimed that gryadka, which claims to be built on top of a Paxos-backed CAS register, was incorrect by providing a “counter-example”.

It turns out that you really drop 20 IQ points on a vacation, and that I had not at all provided a counter-example, despite having looked closely at the algorithm.

My claim that the Paxos-style CAS register exhibited anomalies is thus wrong (or, at least, unproven).

My apologies to @rystsov, who kindly pointed out my mistake. I’ll look into the algorithm more (post-vacation), but now considering that it might be correct, which would be quite exiting - perhaps there are Paxos-like things which are not Paxos but are not wrong.

The original version of this post is available in the commit history, preserved for posterity. My mistake not carrying out the read phase in the middle fully: If you do, then you’ll commit the old value “again”, making the anomaly-exhibiting later CAS fail.