In various cases, such a the DecayTreeFitter, we have particles and their descendants living in the same Particle::Container, and similarly their vertices. Packing these events is fine, but when unpacking it seems to get stuck in an infinite loop.
Example is available at /eos/lhcb/user/l/ldufour/segfault_unpacking.
Related: Moore!3710 (merged)
added Persistency bug labels
mentioned in merge request Moore!3710 (merged)
Without looking at the actual problematic case, and just inspecting the code (again), the only way I (currently, naively) think it can get into an infinite recursion is in resolveLinks - specifically when it looks at the dependencies of the requested buffer prior to unpacking the elements in the buffer. In the case you describe, there will (most likely) be a dependency of the buffer on itself, but I would think that case is explicitly handled by this branch which is basically 'do nothing' as opposed to the other branch which requested loading of the dependent buffers.
So given that it does go recursive in the reported case, the condition which decides which branch to enter must be incomplete...
We should test if the problem occurs when we only have the same particle, or the same vertex container, or whether both come into play for some reason.
In any case, we can debug this properly later. For the upcoming release we could make sure we're not using SubstitutePID or the DTF in the trigger, to be sure.