Two tests failed under -race on main too: the re-ask state fake's maps were written by a try on the watch's goroutine while
the test read and deleted them, and the operator-ask test read the build store unlocked and waited on the builder while
the build its answer starts could still be counting itself in. The fake's maps are locked and read through it, the store
is read under its lock, and the builder is waited on only once that build has claimed its key.
Builds: design 47 H H-A-290 (the module's tests pass with -race, as its checks ask)
A start asked a publish again for every unreleased done build, older ones whose digests a newer reservation or release
of the same module made dead included, and a live publish queued during a long start pass waited for its end. A done
build a newer reservation or release (the builder's own record) supersedes is now retired at the start and at the pass:
its key dropped, its publish item ended, said in the journal; the start asks only each module's newest. A live enqueue
marks its holder, and a pass about to send a start-class item stops there, lets the sends in flight end and lists again,
so the live one goes next; a start-class enqueue never does.
Builds: design 47 H H-A-291 (1), (2), H-A-290, H-A-275
The tool runner's release waited forty minutes behind old versions: every start asked a publish again for every done
build, and each send relisted and reread the whole queue. A publish whose build ended in this run now goes before any
asked again at a start (the class kept on the key's value, a live key of a build before the start made a start key at
the start); a pass lists a holder's keys once and sends them four at a time; a withdrawn or taken-over build is dropped
with its publish item ended at enqueue, at the pass and at the start; a send that returns unsent says why, and every call
it makes honours its one bound; builder_publish_queue reads its states in pages and answers counts per holder and class,
the oldest and newest queue time, a list capped at fifty and one module's items with their class.
Builds: design 47 H H-A-290 (1)-(5), H-A-269, H-A-271, H-A-275
main.117 judged its end at 17:45:19 and never wrote it: every mover's widens
was read inside the end's compare-and-set loop, 126 movers each cut on
grants-for, so each try took about eleven minutes and every write was refused
as changed until the restart, whose start pass judged no end.
The reads an end needs are made once, before the compare-and-set, shared by
every mover and every retry; the runner's start pass ends each running
pipeline with nothing open before the queue is placed; every end says its
outcome in the journal.
Builds: design 47 H H-A-288, H-A-289, H-A-223, H-A-243
Review round 1: a kept mover's ask ids still came from the pipeline's start
time, so a pin-held ask or a composition took the id of an ask the runner
had just dropped for the old fingerprint, and a restart never matched the
steps placed again; and a kept mover placed later read the branch's held
view, so a lagging view gave the new fingerprint the old pins. Every ask id
of a kept mover's steps (build, tests, validate, pin-held asks,
composition, rerun, restart) now mixes its kept revision into the id's
sum, and every placing of a kept mover (at once, later, at a restart, a
rerun) reads its entries at its kept revision by number, waiting while
that revision cannot be read. The test's starter gets a clock that moves.
Builds: design 47 H H-A-286 (amending H-A-284; with H-A-283, H-A-285)
A module importing packages its own pipeline releases was re-pinned at
each publish; each re-pin revision of the same source took it out of the
running pipeline, and each newer build withdrew the older reservation, so
nothing ever published. A revision of the same source now moves no mover
out of the open pipeline of that source: the pipeline keeps it at its new
fingerprint, ends every step of it superseded (said at the old
fingerprint, its asks dropped, its runs left) and places them again from
the re-pin revision's entry under new asks, the build first, still naming
the pipeline's revision. The revision the mover is kept at rides in the
record's kept-revisions, beside the movers, where the pin check reads its
entry. A revision whose every mover is kept starts no pipeline; one of a
new source moves movers as before.
Builds: design 47 H H-A-283, H-A-284, H-A-285 (with H-A-223, H-A-244, ADR 0682)
Live, a published at a version fourteen reservations share cost three seconds:
the fallback by version read every reservation there, and the journal line
read them all again, so the backlog drained at fifty a minute.
The published now hands the reservations it read (each once, a second pass
taking what the first read) to its journal line. The fallback reads the new
index key of the reservations at a version whose build has not recorded every
digest (kept at reserve and where a digest is recorded, written again at the
start), the only ones a published can be kept on; the line counts the rest from
the index by version without reading them. Before that pass the handler says
in progress to the bus.
Builds: design 47 H H-A-280, H-A-271
Every artifact carries its version from its build (a Go module zip's paths, a
program's stamped version, an image's and a bundle's build environment), so
no digest survives a change of number: the move that kept digests would have
published a binary saying 0.1.37 under another number.
The try, on a number a registry holds at another digest, withdraws it on the
reservation (a gap) and raises version-held-elsewhere naming both digests,
then asks the build of the same fingerprint again through the build-again
path of a done key whose version is not released. Its reserve never takes a
withdrawn number again, skips every number a registry holds, and the build
records fresh digests and runs its own publish; the takeover ends the old
number's publish item and bound.
Builds: design 47 H H-A-275, H-A-279, H-A-226
A publish item skipped while the registry held the version would stay skipped
for ever once its published event was lost; and node-tools 0.1.37, held by the
artifact registry at another digest than the build's, was asked again for ever
and never released.
When every missing artifact is held at its version with the built digest the
builder marks it published from the registry's record and says
package-published, as a heard published would; the release follows from the
catalog. Before reserving, the registry seats' versions of the module's
artifacts are read and every number one holds is skipped (version-held-elsewhere,
scope module); a read that fails lets the reserve proceed. A done build whose
number a registry holds at another digest moves to the next free number, its
digests kept, the old number withdrawn as a gap; its old item and bound end and
its publish is asked under the new number. The finding clears at the module's
version-released.
Builds: design 47 H H-A-274, H-A-275, H-A-245, H-A-226
The publish item's try asked the registry to publish node-tools 0.1.37 again
every few minutes while the registry had published it and the builder's own
consumer stood thousands of messages behind, adding events to the backlog.
Before each try the builder reads its durable on the events stream (pending
plus unacknowledged): at a hundred or more the try asks nothing. Below it the
registry seats' own record (their verb versions) is read: every missing
artifact held at the version with its built digest asks nothing either. A
skipped try is said in the journal, written again with its pause and spends
no try; a count or record that cannot be read holds no ask.
Builds: design 47 H H-A-269 (2), H-A-272
The builder's consumer of the registry seats' published stood thousands behind:
each published, and each one that marked nothing, read every reservation.
An index beside the reservations (three-token keys in the same state, which the
reservations' two-token filter never lists) names the reservation keys holding
a built digest and those standing at a version; it is kept at reserve, at a
digest recorded, at release and at drop, and written again once at the start.
A published reads its digest's entry and that reservation; a digest no
reservation holds is judged from its version's entry, never a scan.
The test stores and the tools' read store now read a key filter token by token
as the bus does, which showed the publish queue's resume listing three-token
keys with the two-token filter: it lists them with its own filter now.
Builds: design 47 H H-A-269 (1), H-A-271
Review round 2. The siblings watched main, which is also the default
branch, so no test held the alike rule; they now watch stable and the
entering module is asserted to watch it.
Builds: design 47 H H-A-265, H-A-266
Review round 1. A pipeline that ended (failed, cancelled) with a mover
still waiting left the finding for ever. It is now judged at every
pipeline's end and in Waiting: cleared when the registration is held, or
when no running pipeline still has an unplaced mover of the module.
Builds: design 47 H H-A-262, H-A-265
Review round 1. A module with a register proposal pending is skipped (the
operator's Approve registers it); a module the seed left out keeps
seed-left-out and is not registered; the branch is the first watched
branch the repository's registered modules hold alike, else its default
branch read through the seat git, and a module whose default branch is not
read is skipped and said, registered at the next kept trunk revision. One
listing of the registrations finds the modules with none, so a revision
whose modules are all registered reads no registration.
Builds: design 47 H H-A-262, H-A-263, H-A-264, H-A-265, H-A-266
A mover with no registration had its wait said only in the journal. The
runner now raises the finding registration-missing (scope module, id the
module) through its finding writer as it places a pipeline whose mover has
no registration, and clears it when the module's registration arrives or
the registrations held name it.
Builds: design 47 H H-A-262, G A-327, ADR 0666 §2
Four modules added to a repository after the registrations' seed were never
registered, so a pipeline's steps for them waited on a registration nobody
asked. A kept trunk revision of any kind now registers each of its modules
with no registration, before the revision is said, with no ask: origin its
repository, path its manifest's, the first watched branch, the steps its
repository's other modules hold alike (else the seed's) and a per-artifact
status only where every sibling naming that artifact holds it identically,
said registration-changed. The start does the same once for the current
trunk graph. With no registration held the seed stays the only writer; a
module that leaves the graph keeps its registration.
Builds: design 47 H H-A-262, H-A-263, H-A-264, G A-327, ADR 0666 §2
Review round 1: the start handed a removal-only keeping to the background
while passes wrote wants with no revision kept, so their wanted was never
said. A pass now writes no want until the first keeping keeps a revision
(that keeping wakes the pass), the start's removal folds into the first
pass's keeping, and the pass judges the store's fill itself, through a
handle of its own, before it writes. The integration bed makes the
revisions' object store, as the grants part does from the manifest.
Builds: design 47 H H-A-259, H-A-261; ADR 0682
A pass that kept a fresh assignment graph revision sat six to eight minutes
keeping it and removing the revisions not current, so the wakes of
pipeline-ended and running waited for it. The keeping now runs off the pass
loop, one at a time and folded; the removal takes each entry's module and
fingerprint from the store's listing instead of reading every entry object;
each pass and each keeping says its duration.
Builds: design 47 H H-A-259; ADR 0682
The start read the second bound with a get, which says not found both
for a key never armed and for one the holder's ask-taken deleted, so
every later start asked an answered step once more. The first read
keeps each key's last entry; a first bound's expiry is taken only when
its second bound left no entry at all, and with both passed only the
second is taken, ending the step without asking it again.
Builds: design 47 H H-A-238 (5)
The bounds were watched for updates only, so an expiry marker written
while the runner was down was never heard and that ask had no bound.
At the start each ask key's last entry is read; an expiry marker is
taken then. A running heard before the seat-holders view is read is
heard again after a delay instead of at once.
Builds: design 47 H H-A-238 (5), H-A-243 (1)
The rerun put only the named mover's validate under the new ask; the
others stayed failed under the old one, so the pipeline failed again
and a second rerun made a second validate ask, breaking the pipeline's
one validate ask. The compositions its failed validate ended not asked
now wait again, asked once validate passes as at the placing.
Builds: design 47 H H-A-51, H-A-236
The pipeline's one validate ask names no mover, but its end names the
mover whose body the holder took; the runner compared that mover with
the first step carrying the ask and dropped a passed validate, so the
pipeline held its slot (main.88).
Builds: design 47 H H-A-236, H-A-252
Only a newer pipeline that waited took movers out of older waiting ones;
one placed at once left them, so main.84 was placed nine hours later
with a mover a newer revision had moved and built a stale fingerprint.
Builds: design 47 H H-A-223, H-A-244 (1)
mesh-builds and mesh-wants say nothing before a step's end, so asking
them again at the short bound would double a running build. Their ask
has one bound, ask-silent-bound, and is put again when the node-engine
says the holder running, since the holder takes asks on a plain
subscription and loses those put while it is down. An ask asked again
on a build seat carries a second id derived in the accepted form.
Builds: design 47 H H-A-235, H-A-238, H-A-241, H-A-243
An ask whose holder lost it (the seat mesh-wants' holder restarted) held
its pipeline's turn for nine hours. Every ask but a test run's now arms
a key of the state bounds before it is put; its expiry asks once more,
said in the journal, and the second ends the step could not run with
the finding call-unanswered, so the pipeline ends and its turn frees.
Builds: design 47 H H-A-235, H-A-238 (1)-(5)
The rerun verb looked a step's ask up by the mover's id, but the validate
ask is one per pipeline and names no mover, so every validate rerun, and
a composition rerun, was refused "placed no ask now".
Builds: design 47 H H-A-236
The tool runner's approval findings of call-unanswered are of scope machine;
the indexer's entry now names scope call, so no approval key is cleared even
where an id form would match. The test's negative ids take the approval form.
Builds: design 47 H H-A-249, G §580
The daemon's call-unanswered keys from before the rename would be seeded and
stated by the tool runner's writer, which owns that kind, and never cleared.
The retired list now names call-unanswered with the indexer's own id forms
(the seed's node-build-<kind>.steps, the floor's
mesh-catalog.versions-floor.<module>, the versions fill's
mesh-delivery-builder.builder_versions), so those keys clear once and the
tool runner's approval findings stay.
Builds: design 47 H H-A-247 (3), H-A-249, G §580
The daemon's writer is handed to the process reader (event-unhandled) and the
heard writer (family-not-heard), kinds its list lacked; and call-unanswered
belongs to the tool runner's writer over the same state. The daemon's
unanswered calls are now indexer-call-unanswered (same scope and id), its list
holds the SDK parts' kinds, and merge-untaken is retired, so its live key and
condition clear at the next start. The test reads the SDK parts handed the
writer and refuses a retired kind still raised.
Builds: design 47 H H-A-242, H-A-247 (1) (2) (3), H-A-82, G §580
The daemon's writer owned every kind while the tool runner's writer shares the
state `findings`; under G §580 each bundle states only its own kinds. The
daemon's writer is now made with FindingKinds, the ten kinds its code raises,
and a unit test reads the daemon's code for every kind it gives the writer and
fails on one the list does not declare.
Builds: design 47 H H-A-240, H-A-242 (2), H-A-233, G §580
A pipeline placed long after its take-in found its old build done but holding no
number, and was told to reserve one; the series had moved to a newer fingerprint.
The step now says it is stale and names that fingerprint and its version.
Builds: design 47 H H-A-244 (2), H-A-226
Review round 1. A branch's take-in runs through the dry indexer, which
shares the live packages state, so a branch cut before main added a
module would have deleted main's key: publish drops a kept key only on
the real indexer at the trunk's merge. And a repository whose every
asked directory is gone (one key, or all gone) returned the whole error
and raised packages-not-published at every take-in: the seat's trees of
the commit now tell a gone directory from a commit with no tree.
Builds: design 47 H H-A-232, H-A-233, H-A-234; issue 1585
A Go module mesh-catalog removed stayed in the packages state, so every
publish of mesh-catalog asked the forge for its directory, the seat git
answered not-found, and the one-directory fallback (issue 1585) ran only
on an answer without the directory: nothing of the repository was
published and its packages' digests froze unsaid.
filesVia takes the seat's not-found as a failure of the tree, so the
fallback runs; publish drops the kept key of a package whose directory
the trunk's merge no longer holds, said in the journal; a listing that
fails for another reason raises the finding packages-not-published
(scope merge, id the repository), cleared at the next publish that lists.
Builds: design 47 H H-A-232, H-A-233, H-A-234; issue 1585
A fill read on every chunk made the tool runner list the registry's
incoming store per chunk. The kept transfer now holds the fill read at its
open and the bytes taken since; each chunk tells their sum, written with
the put already made per chunk, and the store is read again only at the
finish, a drop, or when a part's chunks are dropped.
Builds: design 47 H H-A-227, H-A-228
The validate step's run ask built the manifest validator from the indexer's
tree with no pins, so the Go build module refused it for a requirement of
another repository (main.74's feh validate, main.67's mesh-sdk validate);
the composition step's run ask built the composer from the node-engine's
tree the same way. The seat mesh-wants' holder builds both asks: each now
carries the pins of the entry its program is built from, the indexer's
entry at the step's own revision and the node-engine entry chosen per
machine, those with a version, in the form the entry keeps and the
builder sends.
Builds: design 47 H H-A-203, H-A-192, H-A-205
The series key's Withdrawn grew at every take-over and was never pruned. At each write it now keeps only the
withdrawn versions above the catalog's highest released (those at or below it are refused by the plain rule anyway),
carried across a renumbering of the same fingerprint too; a dropped number is never reused (H-A-226 withdraws
H-A-225's freeing). The comment of the rebuild's reservation now states A-490b as H-A-225 amends it.
Builds: design 47 H H-A-225, H-A-226
After a builder restart, the run already building a module carried its reserved number while a re-run of the same
fingerprint reserved the next, so its publish was refused "has no version reserved or released" (main.61: the builder
0.1.111 vs 0.1.112, mesh-grants 0.1.52 vs 0.1.54, and others reserved twice within seconds). A take-over by another
fingerprint dropped the older reservation but let the older build run on and publish a number that no longer existed.
Reserve now answers the standing reservation of the same fingerprint while it is unreleased and above the catalog's
highest, a build again or a re-run alike (A-490b amended so). A take-over keeps the dropped version in Withdrawn and
withdraws the older build (ended superseded, ADR 0664 §4); a late publish of it is refused as withdrawn, naming the
fingerprint that took it over.
Builds: design 47 H H-A-225; G A-490b as amended; ADR 0664 §4
The watchman's go.mod names buildrun in a replace directive with no require
line, and the walk of the module graph followed require lines only, so
buildrun's requirement of the module assigner's client was never pinned and
every test step of the watchman failed "no pin". The walk now takes the
same-repository modules a go.mod replaces by a directory as required, with
the same same-repository test, so their go.mod files are read and their
other-repository requirements pinned, transitively.
Builds: design 47 H-A-224, H-A-201
A moved mover's steps ended superseded, but their asks kept running on the
build nodes. Each such step holding its own ask on a node-build seat, one
reading no shared run, now has the seat's drop asked with the newer
pipeline's why, through the drop the runner already bounds and keeps. The
seat mesh-builds serves no drop, so a build ask there runs to its end,
which the runner ignores: the step is no longer open.
Builds: design 47 H H-A-223
No test failed when Start stopped calling the supersede of running
pipelines, the path the live runner takes. Two rows now start a newer
pipeline through Start, a take-in's and a release revision's, with the
supersede wired to a hearer over the same store, and hold that the older
running pipeline's moved mover left it. The supersede lists only the keys
of the newer pipeline's branch instead of every pipeline's.
Builds: design 47 H H-A-223
With the SDK's guard of one pull per durable (pullOnce reads the consumer's
CachedInfo and releases on the consume's Closed), the fake of a branch's
durable embedded a nil consumer and a nil consume context, so
TestABranchDurableNotThereIsMadeAndHeard panicked. The fake now answers a
ConsumerInfo with its stream, name and a created time of its own, and its
consume context closes when stopped. No other fake in mesh-delivery embeds
jetstream.Consumer; the assigner's, runner's, builder's, watchman's and
grants' unit suites pass with the new SDK unchanged.
Builds: design 47 H-A-196, H §774
A release re-pinned node-tools while main.54 still built it at its old
fingerprint; the build failed and failed the pipeline for its other 119
movers, because supersession took movers out of waiting pipelines only.
Every start of a newer revision's pipeline now also walks the older running
pipelines of its branch: each mover whose fingerprint it moved leaves, its
open steps ending superseded with a why naming the newer pipeline, a step
reading a test run leaving it as a cancel's does, and the mover kept in the
pipeline's superseded. The end judgement counts a left mover's steps as
neither holding nor failed, so the pipeline ends by its other movers alone;
one every mover left ends superseded. A run on request's pipeline is never
touched.
Builds: design 47 H H-A-223, H-A-163, H-A-213
Review round 1: an ask names the revision and the source of one revision, so a
kept revision read by number whose source differs from the ask's is not the
ask's; the check is restored, naming both sources. And a kept revision naming
another source or other fingerprints is a mismatch no wait can mend: the step
fails at once, loud, instead of waiting out revision-wait-minutes; only a
revision the catalog does not answer is waited for.
Builds: design 47 H-A-222 (2), H-A-218
An ask of the very source the picture holds, whose entries the picture no
longer names (a re-pin keeps a later revision of the same source), waited for
the picture and failed past its bound, because the kept revision was read by
number only for a source below the picture's. Now an ask naming a revision is
read by number whenever the picture does not name its entries, and is built
when the catalog keeps it and it names every fingerprint asked, whatever the
source says; the source is compared only for an ask naming no revision.
Builds: design 47 H-A-222 (2), H-A-205, H-A-213
The tools bundle answered the catalog seat's revision verb from its view of the
trunk in memory, which ran one or more revisions behind what the indexer kept
until it heard the event, so the runner and the builder waited on revisions the
bus held all along. The verb now reads which revision is newest from the store
at the time of the ask (a listing and one object) and takes the view's entries
only when the view holds that very revision.
Builds: design 47 H-A-222 (1), H-A-205, H-A-213
indexer_take_in_again wrote TakeInAsk{Again} and DecodeTakeIn refused it: its
hand-kept list of act fields lacked Again, so the handler never reached its
Again branch. Every field of TakeInAsk now says by its tag takein whether it is
an act, an argument or of the ask; DecodeTakeIn reads the acts from the tags,
and a table test builds every hand tool's value by its own Run path, decodes it
and acts on it through takenIn, and refuses a field left unclassed.
Builds: design 47 H §774, H-A-208, H-A-210
The runner crashed at every start on a live bus with "concurrent map
writes" in MoverStore.Get: the waiting pipelines of each branch are placed
from goroutines of their own, and the store's last reads and inline marks
were written under each pipeline's own lock alone. The store's mutex now
guards the lock table, the last reads and the inline marks; behaviour is
unchanged. The run test starts the runner over six kept pipelines in the
old form, one of 437 steps, so the start path runs under the race detector.
Builds: design 47 H H-A-187, H §774