The takeins durable delivers from the moment it is made, so a take-in said
while the runner was down reached only the main durable, which acknowledged
it with no work: that take-in started no pipeline. Until the takeins durable
is bound, the main durable now works each take-in.taken; from the bind, it
acknowledges with no work those the takeins durable delivers, told apart by
the stream sequence the durable delivers after, so a take-in still waiting on
the main durable is worked whatever the moment it is heard.
Builds: design 47 H2 H-A-500 fourth addendum
The runner heard every event on one serial durable, so a take-in waited behind
hundreds of late step events and its pipeline started ten minutes late. The
runner now makes its own durable <node>_<module>_takeins on EVENTS, filtered
to take-in.taken, and acts on a take-in there alone; the main durable
acknowledges a take-in at once with no work. The test-runs integration test
keeps the runner's findings on their own stream, as the manifest declares:
on EVENTS each reached the main durable unhandled and wrote event-unhandled
again, a flood the take-in now overtakes.
Builds: design 47 H2 H-A-500 (b) and its addenda (2), third addendum (Q2)
A module that hears take-in.taken behind a long run of other events hears it
late; it now hears it on a durable of its own, <node>_<module>_takeins, which
it makes itself. Every module principal, on its own account or carried by a
tool runner, is granted the create, information, pull and acknowledgement of
that one name on EVENTS, none special-cased.
Builds: design 47 H2 H-A-500 (b) and its addenda (2), third addendum (Q1)
The indexer's state heard is declared key-ttl, but its bucket was made before
that declaration, and grants' handling of a held stream only added subjects,
raised caps and moved a source filter: every version-reserved the indexer
kept with a time to live was refused (per-message TTL is disabled, 10166) and
given up after five tries. A held bucket whose state is declared key-ttl now
has per-message time to live turned on, add-only like the cap raise, once per
manifest, and said as asserted; the indexer's TTL stays.
Builds: design 47 H2 H-A-501 (corrected)
Every ask-taken and step-progress listed and read every pipeline to find its
test run, so after a cancel hundreds of late events of ended runs cost one to
two seconds each and held the runner's one event consumer for minutes. The
state test-runs now keeps an ask key ask.<ask> holding the run's key, written
with each ask a run is given and deleted with the run; an event of an ended
run's ask is judged from that key and the run alone and acknowledged at once,
one journal line per hundred.
Builds: design 47 H2 H-A-500 (a) and its addenda (1)
indexer_take_in_pending ran the whole chain of held take-ins inside the one
loop that acts on the takeins asks, minutes long, so its answer was cut and a
second ask waited behind it and answered "holds no source-moved". The ask
now starts the chain beside the loop and answers at once, "chain started at
source <n>, <m> wait"; an ask that meets a running chain answers its state.
The chain's state (branch, source being taken in, since, how many wait) is
the key chain.<branch> of the takeins state, written at each step and
deleted at the chain's end, skipped by the asks' loop and shown by
indexer_pending. A key a stop left is deleted at the indexer's start.
Builds: design 47 H2 H-A-497 (corrected), its addendum and its second addendum
The indexer already takes a branch's held sources in order in one pass, but
nothing said so: on 22 Nov one hand ask took 114, 115 and 116 in turn, about
four minutes each, and it looked as if each needed its own hand call. Each
take-in the chain starts after one landed is now a journal line, "held
source <n> of <branch> taken in next, after <n-1>".
Builds: design 47 H2 H-A-497 (corrected)
The indexer pinned only a pipeline's own revision, so JudgePins removed
revision 292 while main.291's kept mover still had to read it. HeardPins now
pins every revision pipeline-started names in kept under the pipeline's
record, takes the latest started (said again at a re-pin) as the kept set,
freeing a kept revision it no longer names but never the pipeline's own or
its validate base, and pipeline-ended frees them all. JudgePins is
unchanged.
Builds: design 47 H2 H-A-496, H-A-496 addendum
main.291 held the one pipelines-at-once turn 50 minutes waiting for kept
revision 292, which the catalog had removed: nothing pinned it and the wait
had no end. pipeline-started now carries every revision the pipeline's movers
are kept at, and a re-pin says it again with the full list, so the indexer
pins them. A catalog read answered "no such object" is the revision gone;
the placing then ends the pipeline failed "kept revision <r> gone", its
movers are left unfinished and rejoin the next pipeline, and its turn is
freed.
Builds: design 47 H2 H-A-496, H-A-496 addendum
The indexer pinned only a pipeline's own catalog revision, so a revision a
re-pin kept a mover at could be removed while the pipeline still read it
(main.291 waited 50 minutes on removed revision 292). pipeline-started now
carries kept, every revision the pipeline's movers are kept at, said again in
full at each re-pin, so the indexer can pin them. The runner seat's form is
9; the contract is regenerated.
Builds: design 47 H2 H-A-496, H-A-496 addendum
A Go package comes from the Go package registry alone, so a released version
is no longer written as a git tag: the release path and the start's take-in
of a released version call no seat git, and a reserve numbers a series from
the catalog's versions and the registries' only, never reading tags. A
reservation's kept tag field stays for records written before.
Builds: design 47 H2 H-A-494 (withdraws H-A-484's tag steps and H-A-493; design 51's fallback tag retired)
Importers were pinned to mesh-delivery-contracts 0.1.82 to 0.1.84, reserved by
builds a competing build withdrew before release and never published, so their
builds failed "unknown revision". The series' numbered version a reservation
writes no longer pins: a pin is the series' newest released version, so a
version-reserved moves no pin and the release's re-pin moves the importers. A
pin held on a version never released is moved at the next take-in or re-pin and
said per module ("<module>: pinned to <pkg> <v>, never released; moved to <v2>").
Builds: design 47 H2 H-A-495 and its addendum (hq 9e0923ce)
Review round 1 (principle 11, configurable, never hard-coded): the bound of
the record of a version's digests is the builder's setting
record-bound-seconds (default 60), read from its file as revision-wait and
package-wait are, set on the Builder and used for the record. A build key
the builder cannot read, or that holds no build, is said in its journal
instead of being taken silently as not done (principle 3, failure is loud).
Builds: design 47 H2 H-A-492
A takeover or an end cancelling a build's context cancelled the record of its
reservation's digests eight times in a day, each leaving a passed build
unreleased. The record now runs under the builder's own bound (one minute,
not cancelled with the build), and a done build whose digests are not
recorded is recorded from its done key at the builder's next look and at its
start (before H-A-484's start check holds it against the released versions),
said each time.
Builds: design 47 H2 H-A-492
A put wrote its want by fingerprint alone, which the node-engine cannot
compose. The put now finds the version released at its build's parts, as the
pass does, names it in the want, and is refused when none is released.
Builds: design 47 H2 H-A-490 (addendum)
An again pipeline on a built, published key ends naming no version, and the
runner never says its end again, so its movers kept the old want. At the pass,
a mover with no version is matched to the release of its build's parts.
Builds: design 47 H2 H-A-490
Review round 2: a durable nobody ever read (no pull waiting, no delivery ever) was
named behind when its twin was not current, and the plain durable of a module whose
program left was never judged (live: claude-code's plain durables 1,000 to 3,000
behind beside a current `_tools` twin). A never-read durable is now an orphan
whatever its twin; one read once and lagging stays behind. The pass judges both
twins of every heard entry wherever no pass composes them, also on a stream where
only such durables are held.
Builds: design 47 H2 H-A-472 (eleventh addendum)
Review round 1 found two paths untested: a step ask of a withdrawn key
still newest answers failed at once, and a newer entry heard between the
older build's claim and its reservation is not taken over. Each row
fails without its line. The comments of H-A-205 say that H-A-489
overrides it once the newer entry is heard.
Builds: design 47 H2 H-A-489 (review round 1)
Review round 1 found that with no `_tools` twin composed, the behind and orphan
checks never looked at one: live, five twins on EVENTS held 1,000 to 58,000
messages and none was named. The pass now reads, on the streams each consuming
module's heard entries name, every held `<consumer>_tools` and judges it as any
consumer: behind by the bound, or an orphan when no pull waits and its twin is
current. The stale comments on HasProgram and the orphan finding's text follow.
Builds: design 47 H2 H-A-472 (tenth addendum)
Two fingerprints of one module took the series from each other (nats
flipped between 0.1.140 and 0.1.152): a later look built an older
fingerprint again, and its reservation took the newer one's over.
The builder now refuses, before it reserves or builds, a fingerprint of
the first watched branch that is no longer the newest entry heard there,
saying once per key that it is superseded; a step ask of it fails so,
never waits. It looks again just before it reserves. A key withdrawn is
taken again by no look, only by a person's build again.
Builds: design 47 H2 H-A-489
Review round 1 found the fingerprint guard of the build read unproven: every
entry of the 100-key listing carried a time. The listing is now also read with
entries that say none, and no history may be read.
Builds: design 47 H2 H-A-487 (review round 1 finding)
builder_build read every key's whole history to learn when its entry was
written, one more round trip per key of the module; on nats's hundreds of keys
the read crawled. Each entry's own write time now comes with the key's read,
the history is read only for the one key a fingerprint names when its entry
says no time, and a listing cut at the 20 newest says so in its note.
Builds: design 47 H2 H-A-487
The grants part stays the one composer of filters while the tool runner makes its
own `_tools` twin: each heard entry, EVENTS included, carries the twin's filters
(the again filter named for it), ack wait, max deliver, max ack pending and
deliver policy. The pass composes no `_tools` consumer any longer, so it makes no
twin nothing reads; the reader grants add the create and delete of the module's
own twin by name. A membership of a module no longer assigned on its node is
withdrawn by a whole pass, so the runner there removes that module's twins. The
part's own hearing loop leaves EVENTS to the hub, now that the membership names it.
Builds: design 47 H2 H-A-472 (fifth, sixth, seventh, eighth and ninth addenda)
Design 51 copies an upstream image by its pinned digest, and distribution's
mirror says published with no version, so nats, hello-web, mailu and showcase
could never be released. The reservation keeps its upstream refs (Upstream,
filled at the reserve from the plan's upstream tasks) and Missing never awaits
them; the indexer releases a version once every other artifact is published,
naming each upstream artifact with the pin its manifest gives, and refuses the
release loudly when an upstream artifact names no pin.
Builds: design 47 H2 H-A-486, H-A-486 (addendum)
The drop without a second store judged from the parts the holder holds,
and distribution's Holds answers per layer: a publish cut before its
store, whose layers were held, was dropped and never said as published.
The drop now asks the holder's own HoldsWhole (the file at its name and
version); nil never drops, and the reopen finishes as usual.
Builds: design 47 H2 H-A-481 second addendum
A reopen that lacked nothing finished from the held set without reading
the chunks back, so a finish cut after its chunks were dropped, or after
its store, stored wrong bytes or stored twice. Now each part taken is
read back and hashed against its digest as a take does, a failing part is
lacked again, and a transfer whose every part the holder already holds is
dropped without a second store. The takeover bound is the receiver's
field TakeOver, two minutes by default, as its sibling bounds.
Builds: design 47 H2 H-A-481 addendum; review round 1 findings (1), (2)
The verb output failed "holds no header" on an empty object, since the
runtime's object read answers an absent object as empty data without an
error: an empty object is now no object, said "the ask kept no output".
A failed build kept nothing for the verb output to answer: it now keeps
its error text, as the build module bounded it, as the ask's output
object with the header a test step's carries.
Builds: design 47 H2 H-A-482, H-A-482 addendum (a)
An opener that opened a transfer again with nothing lacked was answered
"nothing lacked" and the transfer never finished; another caller with the
same digest was refused; and a transfer lacking nothing at its open was
finished without its idle bound, so a restart during the finish left it
open. Now a reopen with nothing lacked finishes it, by its opener or by
another caller; a transfer that lacks a part is taken over by another
caller once its last part is older than two minutes; and the idle bound
is armed before every finish at an open.
Builds: design 47 H2 H-A-481
The assigner took a tier progress for `pipeline.<branch>.<revision>` only (two dots), so
the progress of a pipeline started again at the head, `<branch>.<revision>.again.<n>`,
was never moved and read as an empty end: its later tiers were never wanted. One parse
of the runner's seat's key forms (mesh-delivery-sdk pipelinekey, over the contract's key
parsers) now reads a pipeline's key for the assigner's tier progress and the runner's
mover keys; a pipeline-ended whose key is of no form is refused loudly and nothing of
it is kept.
Builds: design 47 H2 H-A-485
A version-released whose tag the seat git's holder refused was handed back at once,
spent its five deliveries in seconds and was dropped by the bus: its reservation stood
and its publish item asked again every 30 s for ever. At the start the builder now holds
every built, unreleased reservation against the catalog's verb versions and takes a
released one in as its version-released would (tag, release, bound and item ended,
findings cleared); a refused tag is handed back with the publish pause. The tag check
itself is the forge module's, already accepting a nested package path on the catalog's
trunk.
Builds: design 47 H2 H-A-484, H-A-484 addendum
The verb's description and its answer's class and list say what the
builder now does: one consumer per holder, items listed in stream order,
no class pulled first. Description only, regenerated with seatgen; the
seat's format stays, as at B4's description changes of this seat.
Builds: design 47 H2 H-A-483 third addendum, §793
With one consumer per holder an item has no class to move ahead of, so
builder_publish_first, its mode and its act go. The verb publish-queue
lists the items in stream order, the order the one consumer gives them,
and its description says so. A holder is a module name, so one ending in
-live or -again (an earlier release's class durable) is refused at the
learn, said, instead of having another holder's durable deleted.
Builds: design 47 H2 H-A-483 third addendum, §793
The two class consumers (live, again) could not order live before again
under every scheduling: the idle-dispatch row failed on the build node
after four review rounds. One consumer per learnt holder (publishes-<holder>)
gives items in order; a failed send is handed back with its pause, so an
item put meanwhile goes first by construction. The builder retires the
holder's class durables of the earlier release before making the one,
since a work queue refuses overlapping filters; their unacknowledged items
reach the one consumer. The gap read, the held again items and the live
in-flight check go.
Builds: design 47 H2 H-A-483, its addendum and second addendum, §793
A rerun that would reopen an ended pipeline while another open delivering
pipeline carries one of its movers at the same revision is refused with
"its pipeline <key> is running", nothing asked, so one entry is never built
side by side. Tests now cover the two paths review round 2 found untested:
the indexer keeps a revision's gathering while another pipeline's
reservation is left and removes it with the last, and a test step's rerun
reopening an ended pipeline says pipeline-started before its run is started
again.
Builds: design 47 H2 H-A-479 third addendum, H-A-479 second addendum, H-A-479 addendum (2); review round 2
A pipeline ending not passed withdrew every reservation of its revision, so
an again pipeline's failure could withdraw the head's take-in pipeline's
reservations and the other way round. The indexer now withdraws only the
reservations of the ended pipeline's own movers, read from the movers its
pipeline-ended carries, and withdraws the revision's gathering only when
none is left. A start with modules is refused for a module whose head entry
a pipeline still runs ("its pipeline <key> is running"), and takes the next
n without a bound. A rerun reopening an ended pipeline says its
pipeline-started before the step is asked again, so the revision is pinned
before anything reads it. Tests now cover the again form through the
branch's key forms and a release written into a passed again pipeline.
Builds: design 47 H2 H-A-479 second addendum, H-A-479 addendum (2); review round 1
A revision's kept grants carry the digest of the inputs they were
composed with; an ask whose inputs differ composes again, so grants-for
never lists a change the candidate did not cause. TestGrantsForKeepsWhatItMay
now proves that a revision dropped beyond the bound has no grants kept
either (review round 1); the Logf comment says the tools program's
journal too.
Builds: design 47 H2 H-A-478 second addendum
The indexer pins a pipeline's revision from pipeline-started to
pipeline-ended, so a rerun of a step in an ended pipeline (main.258's
composition at 14:41, after its end at 14:15) asked a step against a
revision the catalog had already removed. A rerun of an ended delivering
pipeline now reads its revision fresh through the catalog seat's verb
revision first and is refused by name, nothing asked, when it is gone
("its revision <r> is gone; start the mover again at the head"); once
reopened it says its pipeline-started again, so the indexer pins the
revision until its next end.
Builds: design 47 H2 H-A-479, H-A-479 addendum (2)
A module whose pipeline failed at an unchanged entry was never carried again
until its fingerprint moved, since start refused the head whose pipeline
existed. The start verb now takes modules: the named modules whose newest
entry has no passed pipeline are started at the head's revision, at the
take-in's tiers, keyed <branch>.<revision>.again.<n> (n from 1, written on
absence); a module whose newest entry passed, or that the head holds no
entry of, is refused by name, nothing started. Exactly one of module or
modules is given. The new key form is read by every key-form list (the
sweep's, a release's, the mover keys'). The runner seat's form is 8; the
contract is regenerated.
Builds: design 47 H2 H-A-479, H-A-479 addendum (1)
A cold grants-for (one composing a revision's grants the grants part did
not keep yet) took 3 to 4 s against the 5 s bound and was cut three times
on 2026-11-22; the bound becomes 30 s in the manifest and the code.
Builds: design 47 H2 H-A-478
grants-for composed every module twice per ask (3 to 4 s against the
runner's 5 s bound, cut three times on 2026-11-22). The tools program
that serves it now keeps each kept revision's composed grants beside the
revision (the newest's and the candidate's), bounded by
grants-for-revisions-per-branch, so a second ask for a kept revision
composes nothing; it reads its inputs at each ask and journals each
stage's time on its standard error.
Builds: design 47 H2 H-A-478, H-A-478 addendum
TestAPassWritesWantsWhileAKeepingIsBlocked read the graph's need right after
two passes, while the keeping takes it on its own goroutine; on a loaded build
node it read too early (pipeline main.251). The blocked object bucket now says
when a keeping meets its gate, and the test waits on that, bounded by the
level bound, as it does on the pass and the wake.
Builds: design 47 H2 H-A-476