Design 47 H: H-A-259, the assigner's pass never waits on the keeping of revisions

This commit is contained in:
2026-11-21 12:02:50 +01:00
parent b727fb1bce
commit 43628f09a4
@@ -1484,3 +1484,12 @@ Where two preparers disagreed, the chosen one is named.
read from its history. A done key whose end could not be read stays marked unread, and no pass judges while any
done key's end is unread (the pass ends with a journal line), never from a view missing a build. P-NATS-native,
P4.
- **H-A-259 (the assigner's pass never waits on the keeping of revisions, 12:02)** Corrects H-A-256 on the cause: the
pipeline-ended and running events do wake the pass, but each pass that keeps a fresh assignment graph revision
sits six to eight minutes in the keeping and removing of revisions (every unlisted entry object read one at a
time), so a wake waits for the blocked pass. (1) The keeping and removing of revisions runs off the pass loop, one
at a time, folded (a second need marks it to run once more), and the pass never waits on it: a pass judges and
writes wants from the graph it holds. (2) The removal's judgement reads no object per entry: it takes the entries
it needs in one read (the revision's entry list the catalog seat answers, or the store's listing), the kind of
row 125. (3) Each pass and each keeping says its duration in the journal. Test: a pass runs and writes wants while
a keeping is blocked. P-NATS-native, P4, P7.