mesh/check being carried to 8afc3ebdf371 on 93e0a7b50d90 | from 27da949140c6
The group feat/47-G-join11 merged 3 of 17 members on 2026-11-13; G's mesh-sdk alone on main breaks 25 trunk
modules. The operator chose to put the order right: a reader release first, then G whole. This restores main
to 8fbba810 exactly.
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
stream-publish, the capability of StreamPublish (A-44), is named in go/buscalls beside the other bus-state words, as G names every capability word in mesh-sdk.
Builds: design 47 G §622; design 42 zfs-owns-zfs A-41, A-42, A-44, S1, S2
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Conn.CallChunks subscribes to the call's chunks before it asks the verb, hands the verb's answer to an accept hook, then
receives the chunks the answer counts, in order, acknowledging each, into a writer; CallFile answers the whole body as
the member file, the form a bundle's seat call answers. The checker, the indexer, the assigner and the recovery keeper
read a step's answer kept as an object through it.
Builds: design 47 G A-291, A-290, A-254
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
A part handed the binding finding-events publishes on mesh.bus.<module>.finding-events.<event>.<node>, a part with none on
the old stream; the stdio writer reads the event of either form.
Builds: design 47 G A-274, A-13, A-30, §529
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
A process publishes on mesh.seat.<seat>.accept.<kind>.<caller>, its module the caller, as Emit publishes an event, so the
gate's channel over a connection can ask and cancel through the seat operator-channel.
Builds: design 47 G §677, §700; lead answers A-272, A-273
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
seatdef.Resolve, which the seat generator and every reader of a definition go
through, makes each marked verb by-caller and adds the event gated-ended
(gated-ended.<caller>.<call>, kept last per subject one day) and the verb
approvals (class read); the seat files stay as written. approvals is the
gate's verb, so CanHold asks no holder to name it.
Builds: design 47 G §677, §697; lead answers A-264, A-265, A-266
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
A waiting call's ask's 60 minutes are a key with its own time to live in the
holder's state approvals, its end heard as the bus's marker, so no clock waits
in the process; a call whose key ended while the holder was away is judged at
its start. A process holder's act names no node, so two instances raise one ask.
Builds: design 47 G §677, lead answer A-255
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
Against an embedded bus, a module's process calls a holder's marked verb through
process.Conn's call: asked, approved, run once, recorded through
mesh-hand-acts.warranted and its answer {answer, approval} heard on its own
gated-ended; a declined call runs nothing; a marked verb served through no gate,
or a gate whose state approvals is not on the bus, fails closed.
Builds: design 47 G §677, §700, lead answers A-158, A-224, A-244
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
conn.AnswerSeat holds a verb its seat definition marks needs-approval through the
holder's gate (process.Conn.Gate, go/approval held.go): waiting calls call.<id>,
act.<escaped digest> by create or compare-and-set, claimed.<ask id> by create; a
call carrying the x-wait: held header answered {asked, call} in the reply's field
held; on the router's record read under the module's name, the oldest waiting call
run once within 30 minutes of the Approve and recorded through
mesh-hand-acts.warranted, the others asked anew; declined unasked for 10 minutes;
expired and refused ends; every end said on gated-ended.<caller>.<call>; the read
at start; gate-unanswered past the ask's 60 minutes; the verb approvals. A call
without the header takes ADR 0286 §37's path. process.Conn's seat call sends the
header on every seat call and, answered held, waits on its own gated-ended.
Builds: design 47 G §677, §700, lead answers A-158, A-186, A-209, A-219, A-224, A-228, A-244
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
At every start an ordered consumer, made for the one read, hands every message from the first, then the
durable is followed, so a part that keeps what an age-kept seat event said (the assigner's version-released
and version-delivered) knows it again after a restart.
Builds: design 47 G A-233, A-189, A-181
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
A stream entry takes `last-per-subject`, and process.Account.HearStream reads a module's own stream kept so whole through an ordered
consumer before following its durable, as HearSeatEvents does for a seat's event.
Builds: design 47 G §703, §556, A-181
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
A message delivered again in the same run keeps its kept time; after a restart, which kept none, it takes the time
the bus stored it, never later than its first delivery, so a message that never settles grows old and the runner's
health check fires for it.
Builds: design 47 G A-187, §76
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
conn.AnswerSeat takes the seat's definition from its caller (the seat list the process follows) and,
before a handler runs, refuses an unproven caller, and every caller while the definition is unread,
names no such verb or opens it to none; a proven caller is among them, since the grants part composes
the call subject only for the effective callers. Every caller of AnswerSeat passes the definition.
Builds: design 47 G §610; lead answers A-165, A-179, A-186
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
process.ConsumerLedger writes stream, name, acknowledged sequence, pending, last event applied with its at and the delivery of
the oldest unacknowledged message into one key per consumer, after each message delivered and applied, so a tool reads them.
Builds: design 47 G §579, A-166, A-176
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
A process answers its seat's verbs through one call: caller-named and
unproven subjects, one queue group per seat, reply {result, error, node}.
The rule that a process answers no call (issue 1171 AC3) is replaced by one
that AnswerSeat is the only way to serve. A compare-and-set write on a
process's own state already exists (State.Create, Update).
Builds: design 47 G A-107 (G-20-lead-answers)
Claude-Session: https://claude.ai/code/session_01NcbgfGKaxw984XyAPfLm4h
A seat event's stream has one bus name, seatdef.KeptStream's, which the
generated contract carries as Stream<E>: HearSeatEvents takes it as it
stands and refuses a dotted or unprefixed name by name, so no part makes
the name a second time (SEAT_SEAT_...).
Builds: design 47 G §260
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
A seat's event kept apart from its family (the node-engine seat's running, kept last per subject; mesh-delivery's owed,
a stream with an age and a cap) had no helper: EventHub binds only the shared events stream, and dev-G-engine and
dev-G-forge each met it. process.SeatStreamName is the bus's name of a contract's Stream<E> (SEAT_<seat>_<event>), one
rule for its maker and its hearers; Account.SeatEventConsumerConfig the module's durable as the maker makes it (last
per subject where the event is kept so); Account.HearSeatEvents binds and pulls it, making nothing.
Builds: design 47 G §260, §24, §39, §146, §153, §199
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
The node-engine hands each process the fingerprint of the want it applies (design 47, The fingerprint handover; mesh-host
W13.1). process.OwnBuild reads it once at the first ask, which Connect and Dial make before they dial, and says it in
the journal; with none, or one that is no single word, it says "own build unknown" and why.
Builds: design 47 G §201, §205, §239
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
Call fails no responders at once on a bus that answers; a callee that re-serves answers no one for
seconds, and the caller kept nothing. CallKept asks again within busretry.Bound.
Claude-Session: https://claude.ai/code/session_01JG73KEq4zoDfM9WtYpcQgZ
mesh/check checking 38659c3937e7 on 55f71b78114f | novox/mesh-sdk#174 in group fix/1966-views-durable
A consumer is made from now (ADR 0443): an event said before it was made is never delivered to it and no later event
shows the loss. The event hub tells a part when the bus made its consumer, and a follower asked before that asks
again. Events passed over, at or below the held revision, or of a state not followed are now said in the log. The
status carries the highest revision heard and why the picture is behind it.
Claude-Session: https://claude.ai/code/session_01NcbgfFSnn24M3GcqttoU7i
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
mesh/check failed: the composition: the composer at mesh-sdk f02f142af128b336c32a7f0059c5e1606bf9b1dc did not run: failed the composer git.novox.be/no…
The check of release sdk1 failed on TestStatePutWritesAKeyWhole: TestACallIsAskedAgainOnlyWhenItReachedNoOne
left busretry's Bound at 50 ms, and on a loaded machine the state test's first pause overran it. Second failure
of the kind "tests leave package-level state set".
Quick fix: restore busretry's waits in the one test and lengthen the state test's bound.
Thorough fix (chosen, the kind met twice): busretry.Waits sets the waits and returns their restore, used as
t.Cleanup(busretry.Waits(...)) by every test that sets them, each with a bound of seconds where it waits to
succeed; and servesAlone gives every test that serves its own runtime channel and Served, put back after it,
instead of tests clearing the channel an earlier one left. A scan of every test assignment to a package-level
variable found no other unrestored one.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
The lease write asked seat:node-service-manager.stop@novox-be through Conn.Call, which published
the '@' inside the subject: nobody served it and the call timed out after 30 s with nothing said
(issue 774 notes 84-87). Every key now parses in one place, calladdress: Call routes '@node' as
CallOn does, and stdio Ask refuses a key the runtime could not route, before anything is sent.
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
ADR 0571 clause 10: the grants part publishes memberships on mesh.assignment until the lease write and stops
then. FollowMembership now reads and watches its own key of the state memberships and nothing else, and refuses
to follow, naming the state, when it cannot read it (there is nothing else to follow, and a silent follow of
nothing is the fault). A module's principal and the runtime lose their subscribe on mesh.assignment; the grants
part's publish grant stays until the write (S21 stops the publishing).
Tests: TestFollowMembershipHearsWhatIsIssuedToTheAccount (integration) edited: no state is a refusal, a
membership put in the state is applied, one published on the subject is not; run locally against its
embedded bus. Rows added to TestTheCopiedComposerComposesAsTheControllers: nobody subscribes mesh.assignment,
and the controller's golden grant of it is the one row the copy now lacks. No new test function.
(issue 774, 774/S5)
Claude-Session: https://claude.ai/code/session_01NcbgfJ9Gum47zkFKWRaovL
mesh/check checking a37181261c24 on 4c1384faefa0 | novox/mesh-sdk#138 in group fix/sweep-2026-10-30
The review found that busretry could ask a delivered call again: Conn.request turned a callee's error answer into
errors.New(words), and Undelivered matched the words "no responders" and "jetstream not enabled", so a callee that
answered so during a blackout ran its work twice. A callee's answer is now *busretry.Answered, which Undelivered
never counts: only the transport's own no responders makes a call undelivered. The other busretry sites read or
write through JetStream's API (State, Ask) or the runtime's state verbs, where the words are the transport's.
At each mesh-grants restart and user-list reload (16:37:24, 16:55:15) nats-server 2.11's reloadAuthorization left
the mesh's account without its imports, $JS.API.> among them, for a few seconds: requests had no responders,
JetStream calls said it was not enabled, and a build-test result on ace was lost.
Quick fix, chosen as asked; the thorough fix is a reload that keeps the account's imports, in the bus's holder.
The package busretry asks a request again, backing off within about 30 s, only while its error says it reached no
one (no responders, JetStream not enabled, no response from the stream, or either in words): nothing was done, so
even a create or an update is safe to ask again. A timeout or any answer is never asked again. After the bound it
is said on standard error and returned. Used by process.State's requests, process.Conn's calls (asked again only
while $JS.API.INFO says the bus itself answers nothing, so a callee that is not running still fails at once) and
asks, and the bundle's state requests through the runtime.
mesh-grants' sealed grants and the controller's merge record each met "message size exceeds maximum" (JetStream
10054) only after the put, and a writer that retries looped on a value that can never land; an update over the
limit was even answered as a revision mismatch, which every compare-and-set writer retries.
Quick fix: cap the two values in their modules. Thorough fix, chosen because the kind was met twice and every
module writes through the SDK: the package kvlimit checks a value against its bucket's largest message (read from
its stream and kept) before the put, naming the bucket, the key, the size and the limit, and turns the bus's own
refusal into the same *TooLarge (errors.Is kvlimit.ErrTooLarge), so no writer retries it. The SDK's own writers
use it: process.State (Put, Create, CreateTTL, Update), process.AwaitedBucket, the runtime state of a bundle
(stdio KeptState, against the module state's 256 KiB) and mesh-hand-compose. A value that can outgrow its bucket
is split over keys or kept in the object store with its name in the key.
jschoubben