Record the bases a build copies, keep them by the builds that stood on them, copy each image once (hq ADR 0257, issue 321)

Conversation 5 Commits 3 Files Changed 27

+1968 -36

3 Commits

Author	SHA1	Message	Date
jochen	c6e372896b	Leave an eligible index for a confirmed collect instead of letting it go alone mesh/merge-gate pass: builds build-agent, mesh-controller, route-proxy → ace, g14, novox, shanks; no bus step; every machine composes with the change as it… Details mesh/repo-check pass: its merge-check.sh passed Details mesh/delivery superseded: a newer delivery to the same trunk took over its walk Details Let go of alone after a build, an index's platforms stayed for ever under a record that said collected, so no later collect could reach them (re-review of #144).	2026-10-08 15:47:42 +02:00
jochen	c5663aa18b	Let platform manifests go only on a confirmed collect, and copy again what a sweep took An unrecorded index or a copy in progress can name a platform the records do not see, so only a person's collect, after its dry run, takes an index's platforms, and only once every kept index of each repository it touches was read. A copy missing a platform is copied again, and a copy a build holds again is no longer recorded as collected (review of #144).	2026-10-08 15:47:42 +02:00
jochen	9907df6530	Record the bases a build copies, keep them by the builds that stood on them, and copy each image once A copied base was named only in what a build stood on, and nowhere when the build failed, so the store's sweep could never let one go (hq issue 321). One repository per upstream image stops each module asking the public registry for the same image again, and letting an index go now takes its own platform manifests, which otherwise kept every byte. A person can record the copies no record names through the new mirrors verb (hq ADR 0257). The forge test fix is the same commit as on feat/plain-notifications: main fails without it.	2026-10-08 15:47:42 +02:00

Author

SHA1

Message

Date

jochen

c6e372896b

Leave an eligible index for a confirmed collect instead of letting it go alone

mesh/merge-gate pass: builds build-agent, mesh-controller, route-proxy → ace, g14, novox, shanks; no bus step; every machine composes with the change as it…

Details

mesh/repo-check pass: its merge-check.sh passed

Details

mesh/delivery superseded: a newer delivery to the same trunk took over its walk

Details

Let go of alone after a build, an index's platforms stayed for ever under a record that
said collected, so no later collect could reach them (re-review of #144).

2026-10-08 15:47:42 +02:00

jochen

c5663aa18b

Let platform manifests go only on a confirmed collect, and copy again what a sweep took

An unrecorded index or a copy in progress can name a platform the records do not see, so
only a person's collect, after its dry run, takes an index's platforms, and only once every
kept index of each repository it touches was read. A copy missing a platform is copied
again, and a copy a build holds again is no longer recorded as collected (review of #144).

2026-10-08 15:47:42 +02:00

jochen

9907df6530

Record the bases a build copies, keep them by the builds that stood on them, and copy each image once

A copied base was named only in what a build stood on, and nowhere when the build failed,
so the store's sweep could never let one go (hq issue 321). One repository per upstream
image stops each module asking the public registry for the same image again, and letting
an index go now takes its own platform manifests, which otherwise kept every byte. A
person can record the copies no record names through the new mirrors verb (hq ADR 0257).

The forge test fix is the same commit as on feat/plain-notifications: main fails without it.

2026-10-08 15:47:42 +02:00

Record the bases a build copies, keep them by the builds that stood on them, copy each image once (hq ADR 0257, issue 321) #144