Design 47 H: H-A-244 and H-A-245, a moved mover leaves a waiting pipeline; a tagged number is never reused

This commit is contained in:
2026-11-21 11:29:40 +01:00
parent 830850d40c
commit 85345a7c05
@@ -1389,3 +1389,17 @@ Where two preparers disagreed, the chosen one is named.
accepted form: the first id's time part plus six hex digits of sha256 of the first id and ".2"; the runner matches
an end under either by deriving it from the step's ask. P-NATS-native (an event, not a timer, is the first
recovery), P4, P7.
- **H-A-244 (a moved mover leaves a waiting pipeline too; a done build with a withdrawn number says so, 11:29)** Met
live: a pipeline placed from the queue nine hours after its take-in still carried a mover whose fingerprint a newer
revision had moved (the series taken over by the newer build at 03:33), so its build step found the old build done
but holding no number and failed. (1) H-A-223 holds for every pipeline not ended: a mover a newer revision moves
leaves a waiting pipeline as it leaves a running one, so a pipeline is placed with only the movers still its own,
and ends superseded when none is left. (2) The builder's answer for a done build whose number was taken over says
"built at <fp> but its series moved to <fp2> <version>: stale", never "no version is reserved". P4, P7 (one rule
for every pipeline not ended).
- **H-A-245 (a version's tag is written once; a number met twice is a fault said loud, 11:29)** Met live: the tag
go/v0.1.59 of a package stood at another commit, so the version-released event could not write it and was given
up after five deliveries while the publish went on. The builder refuses to reserve a number whose tag the forge
already holds at another commit (it reads the tag before reserving, through the seat git) and takes the next free
number, with a journal line and the finding of kind version-tagged-elsewhere (scope module); a tag is never moved.
P4, P2 (the forge's tag is the truth the number must follow).