Design 47 H: H-A-213, revisions are the indexer's numbers; a release revision starts one pipeline

This commit is contained in:
2026-11-20 23:20:39 +01:00
parent d07dc5aad6
commit 0e6b7f97be
@@ -1161,3 +1161,16 @@ Where two preparers disagreed, the chosen one is named.
carries the re-pinned fingerprints; pipeline keys stay as they are. The take-in again (H-A-210) keeps its revision
the same way with the call "take-in again of <repository>" and the runner treats it as a publish's revision for
every mover whose fingerprint moved. P1, P4, P7.
- **H-A-213 (revisions are the indexer's numbers; a release revision starts one pipeline, 23:20)** Met in the build of
H-A-212, four questions. (1) A branch's revision number is the indexer's own counter: every kept revision, a take-in's
or a re-pin's, is the newest + 1; the gate's source is the revision's `source` field, and the take-in's dedupe and
"source to take" read that field, never the number (P1: the source is the gate's truth, the revision the indexer's).
(2) A re-pin runs once per module release, so its revision's call is "release of <module> <version>", not per
artifact or package path; the runner starts ONE pipeline per release revision (movers: the re-pinned importers),
replacing its per-artifact Published pipelines, which go away. (3) The call is a new field `call` of the catalog
revision in the seat mesh-catalog's definition, of kind take-in (the gate's call), release (module, version) or
take-in-again (repository), and the catalog-revision event carries revision, source and call. (4) The runner keys a
pipeline by the revision number (`<branch>.<revision>`; a take-in's number is unchanged where it equalled the
source) and, on a catalog-revision event, reads that revision by number through the revision verb, never the newest
half; "waits for revision" compares numbers. Every revision a pipeline runs on is pinned (H-A-147, S33), a release
revision too. P1, P4, P7.