Kristofer Karlsson's series targets the connectivity check run when objects arrive. Commit traversal stops at the boundary of what is already connected. Trees and blobs reachable from that boundary still have to be walked and marked uninteresting, which allocates a struct object for each. On repositories whose boundary commits have large trees, he says even small incoming changes make the check visit and track more objects than necessary.

The new mode runs in the same rev-list subprocess that check_connected() already spawns, triggered by a new internal flag, --verify-trees-incremental. It verifies each incoming commit incrementally against its parents. The series follows an earlier RFC Karlsson posted, "check_connected: toward incoming-proportional cost".

The second patch adds the transfer.connectivityCheck option, accepting "full" (the default) and "incremental". In that patch the rev-list flag is accepted but has no effect yet. The incremental verification arrives in the following commit. The excerpts also show perf benchmarks (t/perf/p5413) that vary tree size, incoming commit count and files changed per commit. They also show a new technical document, connectivity-check.adoc, which describes the connectivity invariant and the full check.

Junio Hamano's reaction was positive ("This is exciting"). His one concrete suggestion was to reduce the file's reliance on the_repository by adding a struct repository pointer to check_connected_options, and passing it on to check_connected_promisor().

The excerpts include no benchmark results or detailed review of the verifier itself, so how it performs and how reviewers judge its correctness remain to be seen.