Lets go back to basics.
In terms of chains, nice or not, no difficulty arises if two nodes share cells. The same applies if the same set of cells is visited several times by the same chain. As long as each link is sound and the links are properly joined, the logic of the chain is unaffected. Programmers will evidently need to avoid the opportunity to circle endlessly in some loop.
In terms of links such as A -x- B, Myth Jellies pointed the way. All that matters is what the link means. In this case:
- if the set of cells, A, has x as an entry, B does not and
- if the set of cells, B, has x as an entry, A does not.
Suppose A and B are ALSs. Then the link represents the statement x is a restricted common candidate of A and B. Havard, I think, was effectively saying that a candidate for A∩B cannot be a restricted common candidate. That is true though it is probably more efficient to target the restricted common candidates then pursue a subset of those that are not.
If a nice chain starts at A and comes back to A again, you have a nice loop. Allowing for groups requires minor changes to the definition of continuity. For example, continuity at x= A =y demands |A| = 1 as well as x ≠ y. In other respects I believe the traditional pattern carries through.
I have an uneasy feeling that some of the difficulty may arise from allowing improper links to slip through the net. At any rate, if someone can explain where they see problems with overlap, perhaps we shall both become little wiser.
Steve