Design 47 H: H-A-210, a take-in again recomputes the modules state as a re-pin does

This commit is contained in:
2026-11-20 23:11:42 +01:00
parent 2a5a741d72
commit d7d18ef084
@@ -1134,3 +1134,10 @@ Where two preparers disagreed, the chosen one is named.
the result as `<branch>.<source>.again.<n>` on every watched branch holding the repository; movers whose fingerprint
moved get pipelines as a re-pin's do, under the re-pin's call form; the gate is not asked (P1: source revisions stay
the gate's; derived revisions stay the indexer's). P1, P4.
- **H-A-210 (a take-in again recomputes the modules state, as a re-pin does; H-A-209 amended, 23:11)** Met in the build
of H-A-209: a re-pin on a publish keeps no catalog revision (only a take-in does, H-A-49) and the names of the form
`<branch>.<source>.published.<module>.<version>` are the runner's pipeline keys, not revisions. So a take-in again
keeps no revision either: the tool re-reads the repository's module graph at its current merge commit (the kept
requirements of every module of the repository, on every watched branch holding it) and recomputes the modules
state, exactly as a publish re-pin does; a moved fingerprint reaches the runner by the same path and event a re-pin's
does, and gets its pipeline the same way; the gate is not asked and the revision numbering is untouched. P1, P4.