Snapshots: the in-memory state costs most of a clean status issue #2 closed: completed andymccall opened this 1 month ago · 0 comments

A clean snapshot of 10 000 files spends most of its time loading the base state, and a checkpoint does not help: decoding one costs as much as replaying every change, because both build the same in-memory file graphs. Make those graphs cheaper to build (hashing, allocations, copies), measure against the previous build, and add opt-in phase timings.