theoreticalbts
commented
3 years ago The algorithm implemented in this PR only deletes blocks of height lower than LIB.
The architectural intent is that TopologyCache holds unverified claims of peers' topology, while MyTopologyCache holds the topology that has been accepted by the local node.
By definition the local node cannot accept orphaned forks, that is, blocks whose history branched at any point older than LIB. When advancing the topology cache, MyTopologyCache should prune all blocks on an orphaned fork.
For an example of this situation, suppose our head is 100, our LIB is 90, and there were two forked blocks 91, let us call them 91a and 91b. We have five blocks on the b fork, 91b...95b, as well as 91a...100a on the main fork. Then we receive block 101a, which advances LIB to 92a. The current code discards only 90, 91a and 91b. That is, the code implements only the first of these three rules:
- (1) Discard all blocks with height less than LIB.height (i.e. 90, 91a and 91b should be discarded)
- (2) Discard all blocks of LIB.height except LIB itself (i.e. 92b should be discarded)
- (3) Discard all descendants of blocks discarded according to rule (2) (i.e. 93b-95b should be discarded)
Closes #115