I agree with Fuseleer it is possible to have temporary partition tolerance, which is in essense what Iota is doing (but I am contemplating that it can be done another way with blocks that I think might have superior qualities). But you still need a global longest chain rule to resolve interpartition activity and thus no unbounded partition tolerance (only temporary). Fuseleer, CfB, and myself all discovered the same conceptual insight.
P.S. I edited my prior post. CfB I also edited the post you replied to.
Sure - partition tolerance I was referring to is the extreme case where an island of connectivity emerges, separate from the main consensus group. Of course there must be a way to objectively merge the two groups together should they be united, and bitcoin's longest chain rule is as good as any model I have heard of.
But there is no way longest chain can merge partitions that have double-spends without reversing transactions that were long ago confirmed. Which includes reversing all the derivative transactions. Our (Fuseleer, myself and apparently CfB) point is partitions can only be tolerated with clearly defined resolution of a well defined temporal condition (and I think this requires blocks but seems Cfb and Fuseleer are attempting to escape from blocks but I think those designs will be found to be unsound[1]).
[1] Which I can say with more confidence than in the past, because I think now I have envisioned every possible design and found issues with all the designs (except perhaps the improvement I mentioned). But readers should note I have not seen a detailed description of eMunie's protocol, so I could possibly be wrong but I will put a very very low chance of that.