L3: a source revision ordered by the module graph, the check waits on its candidate's builds, and an emptied repository never drops a module another trunk holds (ADR 0603, ADR 0604, issues 1367, 774) #147

2026-11-03T16:59:35Z

mesh-tools commented

2026-11-03 16:59:35 +00:00

Milestone L3 of design 47, one delivery group on branch feat/l3b-bus with mesh-sdk, mesh-catalog and mesh-tools; pins mesh-sdk b7abe203a929. (mesh-host goes alone after this group: its own check needs this group's checker, which reads not-judged.)

ADR 0603: the indexer keeps the module graph as its state (§2; the graph is this repository's own Go module sdk/, package graph); view_tiers (§6); a source revision delivered over the nodes present (§7); the watchman's waits past their bound (§8); only the newest manifest's provisions hold a want (§9); the check waits on the builder's candidate builds before it composes, new setting build-wait-bound, and asks itself again when a failed build is done again (§10; bases are composed from manifests and named, ADR 0487 §1); the check refuses a graph cycle.
ADR 0603 §5 in the assigner: a consumer's want also waits on a provision the provisioner has keyed while its provider is not yet answering, and on a requirement nothing offers (raised past its bound by the watchman); a requirement offered by a module's own name or by a module the mesh provides itself is not waited on (the second check's failure: a consumer moved before its provider).
ADR 0604 §1: the seat mesh-catalog's verbs answered from the indexer (served once the indexer claims the seat, in the next group).
Issue 1367: a merge that leaves a repository no module never deletes a module another trunk holds; indexer_take_in. (Commits also cite issue 1126, its origin.)
ADR 0601 §1, §3, §4 rows carried from L3a; the indexer's integration test expects a field its judge does not know to wait while its running build is unknown.

Tests: 876 test functions (unchanged; over the 500 budget, no new functions); the assigner and checker integration tests extended; every delivery part's integration tests run locally on this head.

Issues: 774, 1367. Replaces #145 (feat/l3a-waits).

Reviewed-by: mesh-code-reviewer@shanks at 945906ac97
Pushed-by: lead-47@shanks

https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn

Milestone L3 of design 47, one delivery group on branch feat/l3b-bus with mesh-sdk, mesh-catalog and mesh-tools; pins mesh-sdk b7abe203a929. (mesh-host goes alone after this group: its own check needs this group's checker, which reads not-judged.) - ADR 0603: the indexer keeps the module graph as its state (§2; the graph is this repository's own Go module sdk/, package graph); view_tiers (§6); a source revision delivered over the nodes present (§7); the watchman's waits past their bound (§8); only the newest manifest's provisions hold a want (§9); the check waits on the builder's candidate builds before it composes, new setting build-wait-bound, and asks itself again when a failed build is done again (§10; bases are composed from manifests and named, ADR 0487 §1); the check refuses a graph cycle. - ADR 0603 §5 in the assigner: a consumer's want also waits on a provision the provisioner has keyed while its provider is not yet answering, and on a requirement nothing offers (raised past its bound by the watchman); a requirement offered by a module's own name or by a module the mesh provides itself is not waited on (the second check's failure: a consumer moved before its provider). - ADR 0604 §1: the seat mesh-catalog's verbs answered from the indexer (served once the indexer claims the seat, in the next group). - Issue 1367: a merge that leaves a repository no module never deletes a module another trunk holds; indexer_take_in. (Commits also cite issue 1126, its origin.) - ADR 0601 §1, §3, §4 rows carried from L3a; the indexer's integration test expects a field its judge does not know to wait while its running build is unknown. Tests: 876 test functions (unchanged; over the 500 budget, no new functions); the assigner and checker integration tests extended; every delivery part's integration tests run locally on this head. Issues: 774, 1367. Replaces #145 (feat/l3a-waits). Reviewed-by: mesh-code-reviewer@shanks at 945906ac976c4ec61a0947f128bdc07ac2e3f358 Pushed-by: lead-47@shanks https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn

mesh-tools added 49 commits 2026-11-03 16:59:36 +00:00

A manifest with a field the judge does not know waits instead of being refused; a held module keeps its bus users; tools list and redo the verdicts (ADR 0578 amended, issue 774) 6645ac58c2

Review round 1: the indexer learns its build from the node-engine's seat; a held module's seats and hearings come from its running manifest; indexer_judge; one judge pass at a time (issue 774) 78fe1ce960

Review round 2: a held module's object reads come from its running manifest; one with none keeps the list in force, said; indexer_judge asks the engine for the running build again (issue 774) 6301e49c73

The bus has the upgrade policy its manifest says, as any module: the judge and the assigner rows read it from the manifest (ADR 0601 §1) 054f089614

The judge knows the tool runner by what its manifest says and keeps the kept streams itself, and this repository's check that no manifest names the ended controller reads them as text: mesh-sdk names no module (ADR 0601 §4) ab3cb1ca9e

The judge's row for a field it does not know names a field mesh-sdk does not know now: core-credentials is one of its manifest words (ADR 0601 §4) ecf96be8ce

The assigner moves the bus module by its rules as any module, with no bar and no snapshot of its own, and a person may choose its policy as any module's: its node backs its streams up before it applies (ADR 0601 §1, §2) a9aa6d29ea

A release's Approve answers the record ask of a move of the build alone to the build it carries: the assigner reads the release by the build's source revision from the gate's source and merges, and names it in the want; a move that changes a setting or a secret too is asked as before (ADR 0601 §1) daf178c4ca

The grants part knows the tool runner by what its manifest says, never by its name, and keeps a node's tool runner user as the list in force holds it while no manifest there says it yet, said in the held condition (ADR 0601 §4; the prepare's seams finding on mesh-grants's own name) b342c67b19

The checker decides a node the sandbox did not judge by the contract's field (composer.Machine.NotJudged, verdict not-judged) and names it with the composer's words, never by a phrase in its reason (ADR 0593 §9; the prepare's seams finding on "not judged") 6b9d211f72

Pin mesh-sdk 09ec82cb3551, the group's head of feat/l3b-bus (ADR 0601) ea5be10532

A release answers only the build whose fingerprint it moved: the indexer stamps the source revision at which a fingerprint last changed, a recompute with no merge the revision of its moment, and the assigner reads that stamp, never the Src a recompute keeps (ADR 0601 §1; review round 1, finding 1) a41d2851c7

The watchman raises what a node keeps of a module as it runs, for a manifest that does not yet say it, from the module's have, and clears it once the have says none (ADR 0601 §4; review round 1, finding 3) 0fa00d5e6c

The node-engine's witness ping is granted by the tool runner's name its manifest gives, any service's instance on its own node while none says so, never a name in mesh-grants (ADR 0601 §4; review round 1, finding 4) 6d24922172

Pin mesh-sdk 825483043549, the group's head of feat/l3b-bus (review round 1) 478f0110e9

A recompute with no merge writes the fingerprint's source revision as not known, so no release's Approve answers a build it never carried, though the source revision is still that release's (ADR 0601 §1; review round 2) 22aedbe97f

Pin mesh-sdk e7ce86ff62b5, the group's head of feat/l3b-bus (review round 2) 2f387a3f4c

Pin mesh-sdk 25b371bc8397, the group's head of feat/l3b-bus with the module graph and the seat mesh-catalog's contract (ADR 0603 §2, ADR 0604 §1) 5ee3b4c15f

The indexer keeps the module graph as its state graph: each module's build and provision edges written at every recompute when they change, at most 2000 modules, a module's key deleted once the modules state no longer holds it (ADR 0603 §2) f70faed5b6

Pin mesh-sdk acad60547c84, and the indexer's graph nodes take their bases and provision edges from the contract's GraphEdgesOf (ADR 0603 §2) a901b840a0

The check refuses a change whose module graph has a cycle and names it, worked out from the catalogue as the change leaves it; a cycle the trunk already has among the same modules is said, not the change's failure (ADR 0603 §2) 7ba2668aa8

Pin mesh-sdk 182d8757dba9, the group's head of feat/l3b-bus with the graph's ordering edges (ADR 0603 §1, §6) 027c445bae

view_tiers: a source revision's tiers worked out from the module graph at each call, and per module what it waits on now, its build, each node's want and have and each provision it awaits, the nodes offline named away (ADR 0603 §6) 4f6e1021c6

The seat mesh-catalog's four verbs answered from the indexer's modules, manifests and graph, with the old catalog's field names (captured answers as fixtures); served once the indexer claims the seat, in the release after L3's (ADR 0604 §1, §2) 9f49bc92ad

Write and take in a merge that leaves no module.json, so a module-less repository's reach is not held for ever and a removed last module is dropped (issue 1126) 6bc16d74db

The forge bridge wrote no merged message for such a merge, so taken.<repo> never moved
and the view held 'the indexer has not yet taken in <head>' (hq #490-#492). The bridge
now writes it with no trees; the indexer drops every module of the repository on it,
after the same order check as any merged message. The view needed no change: with the
taken key moving it reaches the merge, which its new row proves.

Needs mesh-sdk fix/1126-no-module-reach (a merged message may carry no trees).

A merge that leaves a repository no module never deletes a module another trunk holds, and indexer_take_in takes a repository in (issue 1367) e1d0c9fe03

dropRemoved hands the module over from the other trunk, read from the forge at the head the source names, so no wait in memory is needed. indexer_take_in with a repository takes in a head with no module.json as a merge that leaves none, and answers by name for a head with modules.

builder_builds_since names the operator's command nox builds as its reader, not mesh-catalog (ADR 0604) 39b3ce3c73

The judge's copy of the mesh-catalog seat is the SDK's entry: four optional verbs, so the old module's claim is not outgrown (ADR 0604 §1; F3) aacdba44e3

On B1: the ledger test names its tool runner by its manifest's tool-runner, and ShelfWaiting takes B1's old-section counts (rebase onto main c4470f42) 875d42fe3e

Pin mesh-sdk 18b3eac47c8e, the group's head of feat/l3b-bus rebased onto main 2341e8e9 0ce16ad773

The module graph is mesh-delivery's own: a Go module sdk/ with package graph, used by path by the indexer, the view and the checker, and the seat mesh-catalog's answers built from mesh-sdk's contract types (the operator, 2026-11-03: no graph-related code in mesh-sdk, only contracts; research 081 D11, D13) 3411b91571

Pin mesh-sdk cf5812101b0b, the group's head of feat/l3b-bus with the module graph taken out 4a736858b6

A source revision is delivered over the nodes present: the view leaves out a node the watchman says is away and names it, and the node counts once it runs its wants (ADR 0603 §7) 2c37f48f9d

The watchman mirrors a want held by a rule, a provision asked and a build queued behind a package as waits with a bound, and raises a finding past it (ADR 0603 §8) 92b4307922

Assigner row: the newest source revision's provider edge replaces the older one, and only it is waited on (ADR 0603 §9) a1814c6571

Take a missing graph state as no graph yet and a Publisher's directory as the nearest module above it 1387b10e78

The indexer's recompute and merge failed on listing a graph bucket not yet on the bus, and a package published from inside a module's directory had no publisher, so the dependents of a build module missed what imports its packages (review findings 3 and 4 of L3).

Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn

Test rows: a missing graph state fails no recompute (review finding 3); a recipe change builds a Go module again (D9) b4855f21c3

Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn

The check names the modules built with a changed build module or base; rows for D9 and the assigner's integration test (ADR 0603 §3, §4, §10) 5c4d770aa1

D13b: the composition's warnings, and so the check's line, name the changed modules built with the Go build module the change alters and those standing on a changed base. D9: builder and verdict rows. D15: the assigner's integration test gets a source revision changing a provider, its new consumer and a build module.

Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn

Merge D11, D12, D14: delivered over the nodes present, the watchman's waits with a bound, the newest provider edge (ADR 0603 §7, §8, §9; L3 review round 1, finding 2) 0d1edef846

Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn

Merge D9, D13b, D15 and review round 1 findings 3 and 4: the indexer runs without its graph state, a package inside a module's directory has its publisher (ADR 0603 §3, §4, §10; ADR 0604 §1) fd638b99d5

Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn

The assigner waits only on the keys of provisions the newest manifest still requires: an older edge's asked or retiring key holds nothing (ADR 0603 §9) 9da14440e8

Fold the three new tests into existing ones (test budget); the base note says what the check does effa12c4b4

The check builds no artifacts and no changed base before its dependents; build_change.go now says only that they are composed from their manifests.

Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn

A wait the watchman mirrors keeps its earliest start while the same mark stands, so a key written again with new words does not start the bound again (ADR 0603 §8) 45bd63427f

Merge round 2 findings A and B: only the newest manifest's provisions hold a want, and a wait keeps its first start while it stands (ADR 0603 §8, §9) 2773885600

Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn

Merge round 2 finding C and the check's true words: the new rows fold into existing tests, the check names what it composes from manifests (ADR 0603 §10; test budget) 8f7945a109

Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn

The check waits on the builder's builds of its candidate before it composes, so it composes with them (ADR 0603 §10) 453cdfa607

The builder already builds every candidate fingerprint into its own key. The check now watches exactly those keys on the
bus and composes once each is done or failed, within the new setting build-wait-bound (1800 s). A failed build fails the
check, naming it; a build not ended within the bound is named and the check composes without it. A module built by a
build module the change alters (mesh-build-go or mesh-build-node), one standing on a base the change alters, and one the
trunk holds no record of are not waited on, and the composition's warnings name each with why.

Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn

checker_status names the builds a running check waits on (ADR 0603 §10) 877b09e92b

Worked out from the builder's builds state at each read, stored nowhere: of each running check, the candidate's builds
that are neither done nor failed.

Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn

A check failed on a build asks itself again once the build is done again, and a failed check never waits on builds (ADR 0603 §10) 3842a464ed

A failed candidate build no longer needs mesh-delivery.recheck, an exception verb (ADR 0421 §12): the checker hears the
builder's keys the failed verdict names and, once each is written done (builder_build_again), retakes the check of the
same heads as it does after a failure of the nodes, at its start too. When the tests have already failed the check
composes at once and names the builds it did not wait on, so its red status is never held back by a build.

Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn

Merge ADR 0603 §10: the check waits on the builder's builds of its candidate before it composes, asks itself again when a failed build is done again, and names what it composes from manifests (the operator, 2026-11-03: build §10 in L3)

mesh/check checking 51deed51e3b4 on c4470f42dba9 | novox/mesh-delivery#147 in group feat/l3b-bus

mesh/delivery-group group feat/l3b-bus checking: 4 of 4 member(s) ready

Details

mesh/delivery closed: a newer head of the same pull request

Details

51deed51e3

Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn

mesh-tools commented

2026-11-03 16:59:40 +00:00

Delivery novox/mesh-delivery@945906ac976c — merged since 2026-11-03T20:13:00Z

Source — source.main moved to novox/mesh-delivery@4154c3040c. Whether it reached the machines is mesh-delivery-view's, derived from source.main, the wants and the haves.

Group feat/l3b-bus: novox/mesh-catalog@5fab8f118e, novox/mesh-delivery@945906ac97, novox/mesh-sdk@b7abe203a9, novox/mesh-tools@8b8a760047

composed together: pass — the checker's verdict at the members' heads (eb182f6015c0)

Transitions

2026-11-03 18:50 (new) → proposed (announced): novox/mesh-delivery#147's head announced
2026-11-03 18:50 proposed → ready (accepted): a verdict names this commit and the trunk's head as its trunk commit, the gate passed or warned, and the repository's own check did not fail; or the check gates no merge (check-gates-merge false)
2026-11-03 20:13 ready → merged (merged): on the trunk its modules follow: merged there

The commit's note under refs/notes/mesh-plan keeps every transition of this change up to its merge: git log --notes=mesh-plan.

**Delivery** `novox/mesh-delivery@945906ac976c` — **merged** since 2026-11-03T20:13:00Z **Source** — source.main moved to novox/mesh-delivery@4154c3040cee. Whether it reached the machines is `mesh-delivery-view`'s, derived from `source.main`, the wants and the haves. **Group** `feat/l3b-bus`: novox/mesh-catalog@5fab8f118e18, novox/mesh-delivery@945906ac976c, novox/mesh-sdk@b7abe203a929, novox/mesh-tools@8b8a7600479e - composed together: pass — the checker's verdict at the members' heads (eb182f6015c0) **Transitions** - 2026-11-03 18:50 (new) → proposed (announced): novox/mesh-delivery#147's head announced - 2026-11-03 18:50 proposed → ready (accepted): a verdict names this commit and the trunk's head as its trunk commit, the gate passed or warned, and the repository's own check did not fail; or the check gates no merge (check-gates-merge false) - 2026-11-03 20:13 ready → merged (merged): on the trunk its modules follow: merged there The commit's note under `refs/notes/mesh-plan` keeps every transition of this change up to its merge: `git log --notes=mesh-plan`.

mesh-tools commented

2026-11-03 17:51:33 +00:00

mesh-delivery: the operator's Approve of this pull request's merge into main is ended: its head a24a1d72ad45 is no longer the pull request's head, or the pull request closed. Nothing is merged on it. The operator is asked again once the checks pass at the new head.

jschoubben added 3 commits 2026-11-03 17:55:03 +00:00

The indexer's integration test expects a field from later to wait, not be refused, while no engine says which build runs (ADR 0601 §3) 70ac4ff79a

Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn

Pin mesh-sdk b7abe203a929, the group's head of feat/l3b-bus: it keeps, deprecated, the exported names the running parts still use b96a4f1d99

Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn

Merge the indexer's integration test: a field its judge does not know waits while the running build is unknown (ADR 0601 §3; the group check's failure)

mesh/check failed: integration: mesh-delivery-assigner/cmd/mesh-delivery-assigner.TestWantsAreWrittenAndNoDifferenceIs:… | f12823681c06 on c4470f42dba9

mesh/delivery-group group feat/l3b-bus rejected: composed together: the checker's verdict at the members' heads (62ca67fc6b58): integration: mesh-delivery-assi…

Details

mesh/delivery closed: a newer head of the same pull request

Details

f12823681c

Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn

jschoubben added 3 commits 2026-11-03 18:50:32 +00:00

The assigner's integration test puts the provider's module before the consumer's, so the consumer's requirement always has a provider to be judged against (the group check's flaky failure, 5 of 40 runs) 34ef185ba5

The assigner waits for a provision whose provider is not known yet: an unknown provider has provided nothing, so a consumer never moves ahead of a provider the indexer has not taken in (ADR 0603 §5); the integration test's write order is unordered again 1c66d58bce

The assigner waits for a requirement nothing offers, and for a provision the provisioner keys whoever is placed yet; a requirement something offers (a module's own name, a module the mesh provides itself) is left to Awaited as before (ADR 0603 §5)

mesh/check checking 945906ac976c on 4154c3040cee | novox/mesh-delivery#147 in group feat/l3b-bus

mesh/delivery merged: source.main moved to novox/mesh-delivery@4154c3040cee

Details

mesh/delivery-group group feat/l3b-bus merged: every member is merged

Details

945906ac97

mesh-tools commented

2026-11-03 18:50:40 +00:00

mesh-delivery: the operator's Approve of this pull request's merge into main is ended: a head it covers moved on: the heads it covers are no longer the group's heads. Nothing is merged on it. The operator is asked again once the checks pass at the new head.

mesh-tools merged commit 4154c3040c into main

2026-11-03 20:12:56 +00:00

mesh-tools deleted branch feat/l3b-bus

2026-11-03 20:12:57 +00:00

Sign in to join this conversation.

2 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: novox/mesh-delivery#147