Design 47 H: H-A-151 to H-A-153 carry the clock's times

This commit is contained in:
2026-11-20 16:45:41 +01:00
parent 94f45529cd
commit abe42c4240
@@ -740,14 +740,14 @@ Where two preparers disagreed, the chosen one is named.
when the store answered the fetch and it failed (a 404, bytes not of the digest), never when the bus did not answer
(that fetch is tried again as today, the module not held for it). The take-in's fetch and the apply's fetch of a
bundle both forget, since either can meet a wrong digest. P7, P3.
- **H-A-151 (validate's base where the previous source number was never kept, 16:46)** Met in the review of H-A-147:
- **H-A-151 (validate's base where the previous source number was never kept, 16:42)** Met in the review of H-A-147:
a take-in that moves nothing (the seed, a merge with no mover) keeps no revision for its source number, so a base
read as number minus one names nothing, and the pin of H-A-147 cannot hold what was never kept. Validate's "before"
case reads the newest kept revision below its own that its take-in pinned (H-A-147: the indexer pins the newest
kept revision at the time of the take-in, not the previous number); where the take-in pinned none (the branch's
first take-in) the before case is empty and says so in the step's words, never a failure and never a could-not-run.
P3, P7.
- **H-A-152 (one set of unlisted assignments, read from the watched branches, 16:50)** Met in the review of H-A-148:
- **H-A-152 (one set of unlisted assignments, read from the watched branches, 16:43)** Met in the review of H-A-148:
the build left out every assigned module the case's manifests lacked, but said in the judgement only those the
watched branches' newest revisions list nowhere. There is one set: the modules assigned to the machine that no
watched branch's newest revision lists, read once at the run's start and carried on the run to the leaving-out, the
@@ -755,7 +755,7 @@ Where two preparers disagreed, the chosen one is named.
while a machine still holds it assigned stays in the composition and fails it as "nothing provides", the change's own
fault (ADR 0578 §2), never a stale assignment. Where the newest revisions are not followed yet, nothing is unlisted
and nothing is left out. P3, P7.
- **H-A-153 (the run value carries its unlisted set, 16:52)** Met in the build of H-A-152: the judgement and the
- **H-A-153 (the run value carries its unlisted set, 16:45)** Met in the build of H-A-152: the judgement and the
resume after a restart read only the stored run (state `composition-runs`), and no field of the run value (G-24 §719)
holds the set; the kept request is gone before the judgement. The run value gains the field `unlisted`, the module
names of the assignments left out, written when the run starts and read by every step's judgement and by the