Answer A-475a: the stale count, the catch-up's wait and validate-tries' bounds

This commit is contained in:
2026-11-17 16:17:38 +01:00
parent fd3d7371bc
commit 2eb4059897
@@ -1859,3 +1859,8 @@ the principle that settles it, cited as "design 47 G A-n".
`mesh-catalog`'s verb `module` is said with its reason and asked again on the next pass, so a store holder left
name-only by a failed ask is loud and recovers without a restart. (From the review of the store fetchers' fix: the
live case is a name-only holder, which A-461b's "every other fault still refuses the node" would have refused.)
- **A-475a** (A-475; principles 7, 2 and 11) "In a row" is counted on the pending revision's validation itself
(`stale`, the stale ends since its last ask that was not stale), the only place that needs it. The catch-up waits
while a validation ask of any repository is out, since every take-in moves the one revision a request is built from;
it still compares the heads and raises A-463's merge-untaken condition meanwhile, and only its take-in waits.
`validate-tries` defaults to 3 and is at most 20. (From the develop and review of A-475.)