Commit Graph
950 Commits
Author SHA1 Message Date
jschoubben 719956b745 R1-F-parts: the receiving side keeps a transfer's idle key and reads its cap from the binding
distribution and gitea each carried the same idle bound (one expiring key per open transfer, its end
dropping a transfer idle for 24 hours or arming it again for the rest) and the same read of incoming's
cap from its binding: the second instance of the kind, so it moves to the Receiver every holder shares.
Arm is called on each open that leaves a transfer open, a key not written failing it loudly; Expired
handles a key's end; Bound reads the binding's cap over Cap. BusArm, BusCap and FollowIdle bind them to
the holder's own state and objects.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 17:24:55 +01:00
jschoubben eb2978b3ce R1-F-sdk 5: catalogue.Decode refuses a manifest giving one field twice, naming its path
A decoder keeps the last of two equal keys without a word, so the indexer's judge read a manifest a
merge of two branches had given one field twice as if one side never was. mesh-catalog's
checks/manifests refuses it already; Decode now refuses it the same way, at any depth.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 16:58:09 +01:00
jschoubben 5e67b548a0 R1-F-sdk 4: one rule every window keeper steps by, and a read naming no node closes none
distribution's two windows and the builder's old-form window each closed for good on a read that named
no node (an empty node_haves, an unread placement): the second of a kind (security-upgrade review F3).
delivery.StepWindow over delivery.WindowHeld is the one rule both call; a read naming no node judges
nothing, as a first view's empty answer at revision 0 (ADR 0615).

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 16:24:02 +01:00
jschoubben aef9957450 R1-F-sdk 2: a view event not said is said again before the mark moves past it
A failed say was logged and passed over, and the next change's mark moved past it, so a reader
missed the change for good (design 25 §12.2). The emitter now says the same change again on a
backoff until it is said or its context ends.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 16:22:18 +01:00
jschoubben c4de50e91e R1-F-sdk 1, 3: no test imports a seat's contract module, and the guard reads test imports
The import guard skipped _test.go files, so holdings_test.go's import of the mesh-broker seat would have
reached mesh-sdk's go.mod at the pin (ADR 0625 §1, §3). The vault entries' comment says they stay to the
next milestone (issue 1966 note 28).

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 16:21:05 +01:00
jschoubben 64e83c7ee2 J8b: catalogue reads the holder's upstream naming, and a done key may name an upstream copy by it
An upstream copy kept by the artifact store is named by the holder, upstream/<host>/<path>@<digest>, in the seat form
(container-image-registry/upstream/...) and the older one (artifact-store://upstream/...). InRegistrySeat reads it, InUpstream
reads its host and path, and referenceProblem accepts it for an upstream artifact, so readers stop reading it on their own.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 15:58:59 +01:00
jschoubben 7eba326d3d J7 (2): a judge request carries the known packages in a field of its own
The names a manifest's packages may name travelled in the request's Input, whose meaning is the composition's input;
Packages says them plainly, nil being no list and an empty list a list.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 15:40:49 +01:00
jschoubben 0cc28f5b1f J4 (J3 item 7): merge A3's wip/registry-a3-sdk (c974a11): a holder's receiving side of a file in parts joins its sender
Brings A3's parts receiver and bus adapters, the shared caller lists and builder versions question, a transfer read
back from bus state (0424e87) and catalogue.SeatReference (c10ef07). c974a11 adds nothing: its version_numbered.go
and BuildAnswer.OutOfStep are the join's own. publish stays out (ADR 0627, 1a149d4): packages_list_test keeps the
join's form. TestAHolderTakesAFileInParts is a subtest of TestAFileTravelsInTheParts, so unit tests stay at 476.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 15:12:03 +01:00
jschoubben c974a11c97 J3 pick for A3p: delivery.VersionNumbered and BuildAnswer.OutOfStep as join-mesh-sdk 712b21b has them
Only version_numbered.go and stepasks.go of 712b21b, so mesh-build-go builds against them
before the join; identical to the join's, dropped there.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 14:58:14 +01:00
jschoubben 09cf1ea6fc J4 (lead's item): the packages state's series value is a contract its readers take, not the indexer's own type
delivery.SeriesPackage and SeriesVersion, field for field as the indexer writes them (name, the ecosystem name;
package, <module>/<artifact>), with SeriesPackageKey, IsSeriesPackageKey and a checked encode and decode.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 14:54:33 +01:00
jschoubben 28ce325c2e J4 (A2j finding 2): a verdict names each series change the heads make, so none is lost when the status line is cut
delivery.Verdict carries series, each {module, artifact (or the package's ecosystem name), from, to}, refused when
a field is empty, it changes nothing, or one artifact is named twice.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 14:53:01 +01:00
jschoubben fe83d66b62 J4 (A2j finding 1): an owner's read tool answers a key's kept history, so a reader sees what changed before its first view
The view's module history, heard messages and a want's since asked the owners' read tools for a key's history and
were refused: views.ReadTool takes key with history and answers {history: [{revision, at, op, value}]} from the
state's kept values (KeptSource through mesh/state.history; viewstest.Source and Serve from the bucket).

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 14:48:21 +01:00
jschoubben 712b21bd74 J3 (A3n finding 3, A2k finding 1): the builder's version.numbered is a contract, and a build's answer names its files out of step
delivery.VersionNumbered with VersionNumberedEvent and VersionNumberedSubject is the body the builder says when it
numbers a version, so a build module consumes it from mesh-sdk instead of the builder's internal package.
BuildAnswer gains out-of-step, so the builder decodes a trunk build's compare by its field, not by name.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 14:42:35 +01:00
jschoubben dae61959b9 J3 (lead's a): go/views names no module; its caller gives the owner's read tool and views stream
ProcessReader.Follow(owner, state, tool, stream) takes the owner's read tool and its views stream as the owner's
declaration names them, and refuses another owner's tool or a second stream for one owner; OwnerReadTool, StreamOf,
IndexerStream and AssignerStream are gone, and viewstest takes the same two values. The second instance of a module
name in shared code (the earlier review finding), so the package test now refuses a module's name in go/views.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 14:35:56 +01:00
jschoubben 4827df4d48 J3 (lead's b): two leaves of one module are both kept, the empty accepted event named by its revision
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 14:29:56 +01:00
jschoubben c10ef07ea3 ADR 0605 §9: catalogue.SeatReference names what a build published in the seat's form, the inverse of InRegistrySeat
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 14:29:35 +01:00
jschoubben 3218da4ee4 J3: mesh-sdk within its test budget, each file of declaration and the provider harness one test of subtests
The unit test functions fall from 587 to 476 (the budget is 500): the cases of each of twenty test files in
declaration and provider/harness become subtests of one test named for the file, under their own names, their bodies
unchanged. Integration tests stay at 30 (budget 50).

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 14:28:23 +01:00
jschoubben 635aed57fd J3 (A2i finding 2): a build's end names the generated files its test step found out of step
delivery.BuildLogEntry gains out-of-step, an end's alone, as the build module's test step answers it, so the indexer
reads the field from the entry, not the raw message, and its trunk finding can be raised once the builder writes it.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 14:26:26 +01:00
jschoubben 1a3ef8101f J3 (A2f findings 2, 4): a leaving module's accepted event is named by its module; a user naming no series accepts the first numbered one
ManifestAccepted.EventID names the module when the digest is empty, so the indexer's own id for a module leaving the
trunk can go and the id lives in one place. Manifest.AcceptedSeries takes the series the package has a numbered
version in and answers the lowest, the series of its first numbered version (design 32), DefaultSeries while none is.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 14:25:25 +01:00
jschoubben 5bb0662d19 J3 (A2g ruling): a reader's again tool records its hand act through mesh-hand-acts before it asks
StatusTools takes the reader's handacts.Recorder: again records the act through seat:mesh-hand-acts.hand-act before
it writes the hand ask, and refuses, asking nothing, when the seat does not take it, as the stdio Reader's again does.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 14:23:40 +01:00
jschoubben 315d7e9b8c J3 (A2g, A2h): a process's pictures of its owners' states are go/views', and a first view says when each key was written
ProcessReader and Picture (A2g's mesh-delivery sdk/firstview at c6c6687) and the test owner (viewstest) move into
go/views, so the builder, checker, teller, view and watchman read the indexer's and the assigner's states through the
one shared follower. The first view's answer carries each key's written time as the bus keeps it, and
Follower.Changed hands it on, so a reader's want time is the owner's write, not its own start (A2h's request).

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 14:22:20 +01:00
jschoubben 0424e87399 ADR 0627 §2: a transfer read back from the bus takes its first chunk
A transfer is kept in a key-value state as JSON, and its empty Chunks map is omitted, so the one
read back for the first chunk had none and Take panicked: every publish in parts through
BusTransfers. The unit tests' transfers now round-trip JSON as the bus does. Found by the forge
module's whole test (CAT-G3).

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 14:19:21 +01:00
jschoubben f9e035fbee J1: a failed have carries the pull it waits on (ENG-9, the lead's ruling on pull-waits; local only) 2026-11-04 14:17:29 +01:00
jschoubben 85b31412b9 J1: node-engine's compose-candidate checks leave mesh-sdk for its client; pull-waits leaves the seat-set row (local only) 2026-11-04 14:17:29 +01:00
jschoubben 3273f5b219 Join: L3's mesh-sdk (f30a53b2) onto A1's wip/registry-a1 (c5fc2ee)
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 14:17:29 +01:00
jschoubben 54e1ecce45 ADR 0627 §2: a holder reads its caller list as mesh-config resolved it, not from an environment nothing sets
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 13:40:29 +01:00
jschoubben c5fc2ee859 DLV-T (A2d): a fetch request's package named at a Go semantic version is taken, a branch name refused
The row of buildstep that refused v0.1.7 now refuses main: c1d7218 lets PackageVersions take a group's tagged version.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 13:35:11 +01:00
jschoubben a0465bf0b7 ADR 0625 §10: the README says a seat's code is mesh-contracts', and what mesh-sdk keeps
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 13:34:50 +01:00
jschoubben e802e67f9a ADR 0627 §2: a holder's caller list per verb and its question to the builder are shared, not copied per holder
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 13:32:20 +01:00
jschoubben c1d7218b32 DLV-T (A2d): a package's version is a full commit or a Go semantic version, a group's tagged version
delivery.PackageVersions.Check takes v0.1.15-feat-x.2 beside a full commit (ADR 0607 §3, §5): a delivery group's
check names the group's tagged version. A2d's patch, whole.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 13:31:38 +01:00
jschoubben a410447ff5 SD-M1: a holder's event kept in its own stream is refused under its module's subjects
A family on a holder's event names the holder's own stream, so its subject is mesh.stream.<holder>.event.<event>;
operator-channel's offered and closed said the router's stream but mesh.mod subjects.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 13:31:16 +01:00
jschoubben 7faa4a1b20 ADR 0627 §1: the vault's answers, verbs and signed event are its contract's, not mesh-sdk's
SignAnswer, Signed, EnrolmentTokenAnswer, NodeAccountAnswer and the names SignVerb, SignedEvent, SignedSubject,
EnrolmentTokenVerb, NodeAccountVerb go; mesh-contracts meshvault holds them. The signers state, the kinds, the
signing contexts and the redeem stay: they are the vault module's and the enrolment's, not the seat's.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 13:29:47 +01:00
jschoubben 44a97d9ad6 ADR 0627 §2: gofmt the bus adapters
Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 13:27:21 +01:00
jschoubben ce592a94d0 ADR 0627 §2: a holder binds the receiving side to its own objects and transfers state through the runtime
Every holder of a registry seat needs the same two adapters over its bus state; kept here so none writes its own.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 13:27:16 +01:00
jschoubben 5bc1bf9f0a ADR 0627 §4: asks and asker leave mesh-sdk for operator-channel's client; the channel's benches stay as package channel
The gate's and the assigner's asks carry the router's ask as its JSON, so mesh-sdk names no type of a seat
defined whole; the ask's id, options, digest and an option's level are read by the gate, the assigner and the
teller through operatorchannelclient. ViewAnswerFits keeps the asking's check, from the key. The channel and
intake benches are not defined whole: their shapes move to go/channel (ADR 0625 §9).

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 13:27:00 +01:00
jschoubben 772b441be3 ADR 0627 §2: a refusal for want of room is told from the others, so a holder's finding for it can clear
A holder raises a finding when it refuses a file for want of room; that finding is the holder's, not the file's,
and clears once it takes a file again. AtCap and RefusedAtCap let it tell that refusal apart without its own copy
of the reason's words.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 13:24:40 +01:00
jschoubben d15c1bc814 ADR 0625 §3, ADR 0627 §4: mesh-sdk imports the seat set in catalogue alone, and names no client seat outside it
buscompose, delivery's builder and step asks, handacts, health and provisions take their seat names from catalogue.
The package-list test refuses any other contracts import, the client seats' names outside catalogue, and the moved
packages (asks, asker, publish) by name; what the group's join still moves (asks, asker, the vault module's name in
delivery and provisions) is listed and only shrinks.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 13:20:35 +01:00
jschoubben de3af0cc65 buscalls.Unwrap: one reading of a tool's answer whatever wraps it
A JSON string, the content list, {output, ok, answer} or the tool runner's {result, error}: mesh-config's
unwrapAnswer and the distribution's copy take it, and views.AskVia reads through it.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 13:18:32 +01:00
jschoubben 3249b30e8d SD-M1: a holder's event is under its module's subjects, or its own stream's when a family keeps it
The definition reader refuses any other subject, and a mesh.stream.<holder> subject no family names (A2c's
finding: the checker's events in its own stream were written as mesh.mod.).

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 13:17:26 +01:00
jschoubben 7efb17386d Build log: a build narrates between its start and its end (build.said)
Its module, fingerprint, line and why, as design 51 has a build narrate its publish; a start or end carries no
line (A3g's finding).

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 13:14:43 +01:00
jschoubben 8d883c373e ADR 0627 §2: the receiving side of a file in parts is seat-neutral runtime beside its sender
parts.Receiver takes a transfer's chunks into the holder's own objects under the caller's prefix, checks each whole
part's hash, drops a transfer left idle and tells the caller the objects' fill; the opening's shape, the check before
storing, the storing and the saying of a refusal are the caller's, so it names no seat, verb or seat type and
distribution and the forge's two package seats bind it to the seats they hold.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 13:13:51 +01:00
jschoubben 1a149d44be ADR 0627 §2: the sending of a file in parts is seat-neutral runtime; publish leaves mesh-sdk
parts takes the open and the chunk calls from its caller and names no seat, verb or seat type; each registry
seat's client in mesh-contracts binds it to its own seat (ADR 0627 §1).

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 13:09:19 +01:00
jschoubben e39faa0f38 SDK-1: the parts helper reads the registry part's shapes from a registry seat's module
Interim, for the lead's ruling: ADR 0625 §1 cuts the registry part into each registry seat's own Go module, and §3
lets no mesh-sdk package but catalogue import a contracts module; this helper is logic shared over a seat, a client
in §4's sense, which no clause lets the registry milestone build. Kept building meanwhile.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 12:54:26 +01:00
jschoubben 293df49377 SD-M1 sdk: a seat's family is the shared events stream's, 7 days, no cap of its own
The lead's ruling of 13:0x replaces 12:5x's said-subject: mesh-broker's events follow every seat, on
mesh.seat.mesh-broker.event.<event> in the shared events stream, 7 days; the bus module keeps a longer history on a
stream of its own module. So every seat family is that stream filtered to the seat's subjects: a definition says 7
days and no cap of its own, and the reader refuses anything else.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 12:53:19 +01:00
jschoubben fc50486190 views: an ask through a tool reads the tool runner's {result, error} envelope
A5's request: the bus module's keeper asks the indexer's first view through the tool runner, which answers in that
envelope; its error is the ask's.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 12:51:56 +01:00
jschoubben 0b25bce239 ADR 0626 §1: a process's bus may say keeps, for the bus module's keeper
A2b's judge test waits on it: the strict decoder refused the key, so the bus module's new manifest did not decode.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 12:50:41 +01:00
jschoubben 6fb778a889 SDK-A6: a bundle hears its consumer's heartbeat between events, OnHeartbeat
A6's request: the knowledge base's page marks a source unanswered from its subscription; with events alone a quiet
source and a lost subscription look the same. The runtime sends mesh/heartbeat at the bus's idle heartbeats, and once
with missed when they stop; the tool runner's side is a request to its owner.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 12:45:52 +01:00
jschoubben 3622c129f8 SDK-A6: a tool reads its call's id from the runtime, CallIDOf
A6's request: the ledger's apply names the call it serves (ADR 0597 §7) so the runtime stamps the caller it proved as
the operation's author; the id is read from the call's _meta under mesh/call, never from the arguments.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 12:44:16 +01:00
jschoubben f30a53b242 A compose-candidate answer's open input keeps its numbers as written when decoded
The answer's input is now the object the node-engine seat's contract names, so a decoder without UseNumber turns its
numbers into float64 and a checker re-encoding it for the composer would hand on different digits.

Claude-Session: https://claude.ai/code/session_01JG73KG7AfSdh6FP8VrBqCn
2026-11-04 12:43:11 +01:00
jschoubben c515074714 SD-M2d sdk: a seat family on the shared events stream keeps no longer than the stream
Holds operator-channel's decided to the shared stream's 7 days (the lead's ruling of 12:5x), and each registry seat's
family with it.

Claude-Session: https://claude.ai/code/session_01JG73KJNrH5EgFXMKKYSrig
2026-11-04 12:42:57 +01:00