PREBITFLIP=1 clients will have all data and only need to refresh permission changes. Currently when the bitflip happens, they will see a load of permissions go from "d/f" to "d/f" and the diff will include those in the file list.

This hurts nothing, but does make for a lot of stat calls on bitflip day and does slow down the processing of the run where the bitflip happens (on a day when the servers are probably loaded anyway). What we really want is to find things where "d-/f-" goes to 'd/f'.

The problem here is that our diff call doesn't tell us what the line used to be, so after the diff happens we have no way to eliminate the problematic lines. We could preprocess the file list (as part of one of the other awk calls) to turn all "d/f" lines into "d/f" lines before doing any comparison, but... that seems like a lot of effort to avoid 300K extra stat calls once every six months.