The versionchange handler was installed only after the asynchronous admission
finished, so another tab's upgrade that arrived in that window waited blocked
until its deadline and failed. The kernel now installs a handler for the
admission window that closes the connection and ends the open as CLOSED.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
verifyIndexedDbDatasetBinding wired oncomplete, onabort, onerror and the
caller abort by hand, the last copy of the transaction state machine outside
the kernel, and missed the kernel's rule for an abort after commit. It now
runs on runIndexedDbTransaction and onIndexedDbRequest with the same verdicts.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The OPFS journal imported mapIndexedDbException from the IndexedDB runtime's
folder, so the two runtimes were not separable as the comments claimed. The
table now lives in browser-file-storage/result.ts next to the DOM table, with
the differences between the two written down; storage/opfs imports nothing
from storage/indexeddb. The storage barrel comment now states the drill's
real scope: it removes the whole bundle, never one folder.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
isObjectIdAllowed could be widened while stored objects, journal rows and
paths keep the fixed opaque id grammar, so such a policy failed puts late.
Resolving the policy now rejects a predicate that admits ids just outside
the grammar, and the prepared-object check uses the policy's own grammar.
The worker and the journal share one stableJson that throws on values JSON
cannot carry instead of writing NaN as null.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The runbook named failure codes the adapter stopped using and promised a
resume and a quarantine reconcile never does. It now states the put's own
code with RECONCILE, the skipped and failed row counts, and what reconcile
really does per phase. The worker never sent EFFECT_UNKNOWN and the client
refused it, so the cleanup effect type and its two adapter branches go.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A put holds the origin-wide mutation lock from BEGIN to FINISH, and the page
waited on its source with no bound, so one stalled source blocked every
tab's writes until their lock wait failed. Each wait for source bytes is now
bounded by putIdleTimeoutMs; the put fails NOT_READABLE (retryable) and its
compensation releases the lock. The port now says writes are one at a time.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
An older tab that met a schema a newer release had upgraded got
MIGRATION_FAILED / READ_ONLY, but could not even read, and operators were sent
to the migration runbook. The VersionError on open now yields UNSUPPORTED /
ONLINE_ONLY with status INCOMPATIBLE, reads and writes give the same answer,
and the OPFS journal answers the same way.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
One corrupt sidecar or receipt stopped every later sweep or prune, so
expired rows stayed on the device and receipts grew until all writes failed
LIMIT_EXCEEDED. Both now leave the corrupt row, count it and report it as one
CORRUPT_DATA observation, and keep a budget they cannot account for as it is.
A malformed budget row in prune now fails as CORRUPT_DATA, not MIGRATION_FAILED.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A LOGOUT or ACCOUNT_DELETION purge removed object directories only; chunk
bytes stayed until a GC 24 h later that might never run for that user.
Once every object is gone the purge now sweeps the partition's unreferenced
chunks at once and reports how the sweep ended. Chunks the journal or a
staged put still references are kept.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
LOGOUT and ACCOUNT_DELETION purged row by row and so needed an open that a
corrupt budget or binding row always rejected, leaving personal rows on the
device. After authority approval, a purge that finds the dataset rejected by
admission now deletes that partition's own database and reports it; healthy
datasets and other partitions are untouched.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A first receipt write that hit the quota left a zero-byte receipt.json that
cleanup and orphan GC could not parse, so reconcile and GC for the whole
authority stopped for good. An empty receipt now counts as no receipt, and
reconcile counts a failed row, observes it and moves on to the rest and GC.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Reconcile cleaned up any incomplete journal row, including a put still
running in another tab, and the put then reported success for deleted files.
Put and delete now hold a Web Lock per transaction from begin to return;
reconcile takes those locks with ifAvailable, re-reads the page under them
and reports the rows it skipped (D-3).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The kernel already treats an undefined signal or timeoutMs as absent, so the
conditional spreads around them were dead weight.
file-transfer gzip: 54484 -> 54468
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
combineAbortSignals moves verbatim from resumable-upload-runtime.ts to
platform/abortable-operation.ts (AbortSignal.any is not available on the
Safari 16 target). Behavior is unchanged: either input aborts the combined
signal, and release only drops listeners. Adds kernel unit tests for it.
gzip: file-transfer 54499 -> 54484, offline-indexeddb 35854 -> 35856,
service-worker-pwa 9500 -> 9499, realtime 39685 unchanged (net -14).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The native-input wait and the system-picker wait no longer add their own
abort listeners; both use createAbortableOperation, and dispose closes the
operation. Results stay ABORTED for caller abort and for dispose. Adds a
characterization test for dispose while the enhanced picker is pending.
Hand-rolled abort ratchet 16 -> 15.
file-transfer gzip: 54500 -> 54499
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
readWithSignal now races reader.read() through createAbortableOperation
instead of its own abort listener, and createAbortScope takes the single
consumer signal its only caller passes instead of looping over an array.
Adds a characterization test for a consumer-only abort during a pending
read (ABORTED, owner signal untouched, body cancelled).
file-transfer gzip: 54546 -> 54500
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The parser now finds CR/LF with Uint8Array.indexOf before decoding, so a
line's UTF-8 byte count comes from byte offsets instead of re-encoding the
text. Each complete line is decoded once by a streaming fatal TextDecoder;
a line that continues into the next read is stream-decoded, and the bytes
of a character split across reads count once it completes, as before.
Behavior kept (the push() input was already Uint8Array; no seam change in
fetch-sse-connection.ts, public API unchanged):
- CR / LF / CRLF, CR and LF split across reads, multi-line data
- one leading BOM stripped (now done by the parser, decoder uses
ignoreBOM: true, so a BOM at a later line start stays text)
- line / event / incomplete-buffer byte limits, same failure point
- drain() and the per-call item cap; kept bytes are copied
- a pending CR or the item cap waits while only part of a character is
left, as the text-based parser did
- error codes (MALFORMED_EVENT / EVENT_TOO_LARGE / QUEUE_OVERFLOW / CLOSED)
Differences only on invalid UTF-8: the old parser decoded a whole chunk
first, so malformed bytes won over another error later in that chunk and
malformed bytes past the item cap failed in push(). Now the first
problem in stream order wins and those bytes fail in drain(). Both
still fail closed.
Fixed: Node's streaming TextDecoder (ignoreBOM: false) drops a U+FEFF
right after the stream's first character when that character is split
across reads, e.g. [EF] [BB BF EF BB BF] decoded to "" instead of
"". The old parser inherited that; the new one keeps the text.
Tests: existing parser and connection tests unchanged and passing. Added
a split-BOM case and a differential test against the round-2 parser
(tests/unit/realtime/sse-parser-reference.ts) over 2x1,500 seeded random
ASCII / multibyte streams with random read splits, random small limits,
truncated streams and skipped drains. 300,000 seeds also matched in a
manual run.
Benchmark (manual, not committed; Node v24.14.0, one push of one chunk,
best of 150 runs, host load average ~11-14 so absolute ms are noisy):
| input (one chunk) | original e59994a | round 2 | new | new vs original | new vs round 2 |
|--------------------------------|------------------|---------|---------|-----------------|----------------|
| ASCII, 249 KB, 209 events | 103.72 ms | 0.48 ms | 0.30 ms | 348x | 1.6x |
| Korean/emoji, 249 KB, 209 ev. | 23.30 ms | 0.85 ms | 0.44 ms | 53x | 1.9x |
| ASCII, 234 KB, 4 x 60 KB lines | 176.45 ms | 0.61 ms | 0.34 ms | 513x | 1.8x |
| Korean/emoji, 234 KB, 4 x 60KB | 25.28 ms | 1.10 ms | 0.65 ms | 39x | 1.7x |
Across five runs, multibyte new vs original stayed between 19x and 70x
(target: 10x or more).
Realtime bundle: 39903 / 40000 gzip bytes (baseline 39685, +218).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The resumable-upload fetch transport had its own parser that handed any
non-integer value to Date.parse, so "1.5", "2025-01-01" or an RFC 850 date
became a zero-delay retry. It now uses platform/retry-after.ts, which only
accepts delta-seconds or an IMF-fixdate.
file-transfer gzip: 54406 -> 54546
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The capability-verification deadline, the probe wait and runtime close now
use createAbortableOperation/snapshotAbortTimers instead of three local
helpers and a default scheduler wrapper. Results, error codes and the
scheduler TypeError message are unchanged. Hand-rolled abort ratchet 17 -> 16.
file-transfer gzip: 54562 -> 54406
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- client: the final V2 classification step maps a 408/425 response with a
valid error envelope to SERVER_FAILURE (via gatewayFailureOr), keeping
the backend code and request/trace ids. contracts/errors.ts is unchanged.
- client: Retry-After is also parsed (kernel parser) on 503, so the retry
waits at least that long and the 30s ceiling applies. 502/504 keep the
local backoff.
RED (before the fix, 14 failing in tests/integration/http-client.test.ts):
- safe GET on a 408/425 error envelope: kind UNKNOWN_CLIENT_FAILURE,
fetched once (expected SERVER_FAILURE, 3 fetches, code UPSTREAM_BUSY)
- none/keyed command on a 408/425 error envelope: kind
UNKNOWN_CLIENT_FAILURE; keyed fetched once instead of 3 times
- 408/425 error envelope with Retry-After: 2: not retried
- 503 (gateway body and envelope) with Retry-After: 2: slept 125ms
instead of 2000ms
- 408/425 envelope and 503 with Retry-After: 31: retryAfterMs missing
Pins that passed before and after: a non-idempotent command is sent once
on 408/425/503 gateway bodies and on a 503 envelope; 502/504 ignore
Retry-After (125ms).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- retry-policy: a SERVER_FAILURE with status 408/425 follows the
502/503/504 rule (safe or keyed only; never idempotency "none" or
retry "never"). Retry-After is honored and its 30s ceiling applies to
any failure that carries it, not only RATE_LIMITED.
- client: Retry-After is parsed by platform/retry-after.ts for
408/425/429, on every failure path (gateway bodies included).
The local parseRetryAfterHeader is removed.
RED (before the fix, 15 failing):
- 408/425 gateway body on a safe GET: expected 3 fetches, got 1
- keyed command on 408/425: expected 3 fetches, got 1
- 408/425 with Retry-After: 2: expected ok after a 2000ms wait, got the
first failure
- 408/425 with Retry-After: 31: retryAfterMs missing on the failure
- 429 Retry-After "1.5": slept 1500ms instead of the 125ms backoff;
"Thursday, 01-Jan-70 00:00:05 GMT": slept 5000ms; "1e3", "0x10",
"-1", "2099-01-01T00:00:00Z": accepted by the old parser, so the
request was not retried with the 125ms backoff
- unit shouldRetry(safe, SERVER_FAILURE 408): false
Non-idempotent command on 408/425 stays 1 fetch (pinned, passed before
and after).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The demo adapter now reports "demo-user" as its subject while
authenticated (null otherwise), so an authenticated notification for the
same demo user no longer fences the server-state scope. Sign-out and any
state change still fence.
The scope runtime now treats an empty-string subject like a missing one
(fail-closed), so an owner that returns "" for everyone cannot disable
fencing between users.
Pinned: state change fences, subject change fences, an external owner
without readSubject fences on every authenticated notification.
RED (before the fix, tests/unit/server-state-scope-runtime.test.ts):
x does not fence when the demo session re-announces the same user
expected [ { kind: 'FENCED', ... } ] to deeply equal []
x still fences when an authenticated session names an empty subject
expected true to be false
Tests 2 failed | 12 passed (14)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- browser-file-picker: the native and enhanced pickers carried two copies of
the same selection observation (#finishObservation / #observe); one
module-level observeSelection replaces both.
- indexeddb-checkpoint-store: read / compareAndSwap / remove /
removeUnreadable each repeated acquire -> transaction -> get -> onerror;
withCheckpointRow does that once and hands the row to the operation, still
inside the same transaction. deleteCheckpointRow shares the delete tail.
- resumable-upload-runtime: the four checkpoint removals that remap to
UPLOAD_RECONCILE go through removeCheckpoint.
- image-cdn-runtime: the verification deadline reuses
combineImageRuntimeAbortSignals instead of a second listener loop; the
verifier-coverage check throws its TypeError from one place; the registry
id pattern is one constant instead of three inline literals.
- presigned executor / capability provider: reuse checkpoint-schema's
SHA256_HEX and isSafeUploadReceiptToken instead of equal local patterns.
file-transfer reference runtime: 54754 -> 54562 gzip bytes (budget 54600).
Other reference runtimes unchanged.
Verified: check:architecture, check:types:app, check:types:test, lint,
test:unit, test:component, check:adapter-inventory, build,
check:optional-recipes (file-transfer 54562/54600, offline-indexeddb
35854/36000, realtime 39685/40000, service-worker-pwa 9500/10000),
test:recipes, check:types:recipes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- image-header-metadata: `=== undefined` / `?? 0` / `?? null` guards on bytes
and boxes whose index was bounds-checked a few lines earlier (PNG IHDR,
JPEG marker, WebP VP8X/VP8L, AVIF ftyp/meta/iprp/ipma, ascii,
readUint24LittleEndian). A differential fuzz of the old and new parser over
720k valid and mutated PNG/JPEG/WebP/AVIF inputs found no difference.
- incremental-sha256: `?? 0` on in-range Uint32Array reads and `>>> 0` on
Uint32Array stores, which already reduce modulo 2^32. Checked against
node:crypto for 0..1 MiB inputs, one-shot and streamed.
- p256 verifier: the alphabet index can't be negative after the base64url
pattern check, and the decoded signature is already an owned, exactly sized
buffer, so it is no longer copied again.
- browser-image-probe: once the bitmap matches the header dimensions, which
are positive and inside the decode budget, the repeated positivity and
budget checks on the bitmap cannot fail.
file-transfer reference runtime: 54894 -> 54754 gzip bytes. Other reference
runtimes unchanged.
Verified: check:architecture, check:types:app, check:types:test, lint,
test:unit, test:component, check:adapter-inventory.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- cancelBody / cancelReader move from presigned/http-common.ts into
platform/abortable-operation.ts next to compensateLateHandle; the probe's
own cancelResponseBody/cancelReader copies are dropped and every caller
imports the kernel pair.
- fetch-json-transport uses the kernel operation directly: the FetchAttempt
projection only renamed terminal() (ABORT/TIMEOUT/CLOSED), and only the
CALLER_ABORT case was ever read.
- browser-image-probe uses the kernel operation directly (awaitOwned /
unanswered helpers replace the ProbeAbortScope object), and the body read
goes through the same race instead of the hand-written readOrAbort.
- storage-manager-adapter's inspection wait is the kernel race: a caller
abort ends the wait and a later settlement, including a rejection, is
consumed, exactly as the hand-written listener did. The hand-rolled abort
ratchet drops 19 -> 18 for this file.
file-transfer reference runtime: 55154 -> 54894 gzip bytes.
offline-indexeddb: 35951 -> 35854. service-worker-pwa: 9581 -> 9500.
realtime unchanged.
Verified: check:architecture, check:types:app, check:types:test, lint,
test:unit, test:component, check:adapter-inventory (hand-rolled abort
wiring 17/18).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Identical private copies lived in both file adapters and in five
browser-transfer modules. One copy each now lives next to isValidByteLength
in the browser-file-storage port (outside every reference-runtime source
root, so it adds no namespace export):
- byteBucket (browser-files/file-observer, presigned/http-common) ->
browserDataByteBucket; both modules re-export it under the old name
- positiveSafeInteger / nonNegativeSafeInteger (checkpoint-schema,
http-control-plane-adapter, runtime-policy, upload-byte-source,
image-cdn-policy, browser-image-probe, file-policy's isPositiveSafeInteger)
-> isPositiveSafeInteger / isNonNegativeSafeInteger
- hasControlCharacters (image-cdn-policy) and
hasForbiddenHeaderValueCharacter (fetch-json-transport) -> hasControlCharacter
Same predicates, same results for every input.
file-transfer reference runtime: 55306 -> 55154 gzip bytes.
offline-indexeddb: 35957 -> 35951. realtime, service-worker-pwa unchanged.
Verified: check:architecture, check:types:app, check:types:test, lint,
test:unit, test:component, check:adapter-inventory.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The reference-runtime measurement exposes every export of every catalog
module through a namespace object, so an export nobody imports still costs
bytes. Made module-private (only used inside their own file, not by src,
tests or scripts):
- download-input-validation.ts: SHA256, snapshotDownloadSource,
validPresignedCapabilityForInput, safeOpaqueId
- checkpoint-schema.ts: RECEIPT_TOKEN, isUploadPartDescriptor,
partMatchesFingerprint
Barrel (index.ts) exports are untouched.
file-transfer reference runtime: 55382 -> 55306 gzip bytes. Other reference
runtimes unchanged.
Verified: check:architecture, check:types:app, check:types:test, lint,
test:unit, test:component, check:adapter-inventory.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The resumable-upload runtime and fetch transport carried private copies of
helpers that already exist once in the same adapter or its port:
- the 18-code failure list -> isBrowserDataFailureCode (same list, same order)
- observeTerminal/transferByteBucket -> observeTransferResult, moved from the
presigned executor into presigned/http-common.ts and also used by the
capability provider
- cancelResponseBody/cancelReader in the fetch transport -> http-common copies
- the loose isAbortSignal guard (transport + runtime) -> one export in
checkpoint-schema.ts
- upload-response-validation's exactKeys -> checkpoint-schema's exactRecord
file-transfer reference runtime: 55704 -> 55382 gzip bytes. Other reference
runtimes unchanged.
Verified: check:architecture, check:types:app, check:types:test, lint,
test:unit, test:component, check:adapter-inventory.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The application port (best-effort-telemetry) and the feature port
(feature-telemetry pipeline) each posted to TELEMETRY_ENDPOINT on their
own: 20+20 sends per minute, cookies on the application side, no
pagehide flush or dispose abort on the feature side, and no deadline on
any fetch, so a hung collector held activeFlush forever (6.3-10).
New src/adapters/telemetry/telemetry-collector.ts is the one transport:
shared send-rate window (createTelemetrySendRate, now in the platform
pipeline and injectable via `sendRate`), credentials "omit", cache
"no-store", a per-request deadline through the kernel
createAbortableOperation, and one pagehide fan-out. runtime-adapters
builds one collector and passes it to both ports. Body projections are
unchanged and pinned by characterization tests first.
Verified: new tests in tests/unit/telemetry.test.ts (5 RED before
wiring: credentials/cache, hung collector timeout, shared rate,
feature pagehide flush, feature dispose abort); all gates pass except
the known tests/unit/removal-fixture.test.ts (artifacts/ absent);
check:adapter-inventory 137/137; FE-RB-004 drill PASS.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- bounded-poll: an executor that throws becomes OFFLINE retryable:true,
but OFFLINE is not a poll-retryable kind, so the run ended through
pollFailure(kind) with retryable:false. OFFLINE now keeps its
retryable flag on that exit, like the lifecycle OFFLINE exits.
- sse-parser: TextDecoder(ignoreBOM: false) already strips the one
leading BOM the stream grammar allows; the parser stripped a second one.
The manual strip is removed.
- event-codec: the raw event was UTF-8 encoded twice; now once.
- fetch-sse-connection: one module TextEncoder instead of a new one per
event and per cursor check.
- websocket-protocol: the R-05 admission JSDoc sat on utf8ByteLength; it
now documents exceedsUtf8ByteLimit, and utf8ByteLength reuses one
encoder.
RED: a throwing executor returned retryable:false; a stream starting with
two BOMs before "data: x" produced an event. Both pass now; the
encode/JSDoc items are covered by the existing codec, SSE and WS suites.
Gates: architecture, types app/test, lint, adapter-inventory, test:unit,
test:component pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
bufferProbeValue compared the ceilings against acceptedEvents /
acceptedBytes, counters that only grew. Activation drains the buffer
while the probe keeps receiving, so once a probe had accepted
maxProbeBufferedEvents values in total, the next write during activation
failed closed and left the coordinator CLOSED even with an almost empty
buffer. The check now uses buffer.length and bufferedBytes, which the
drain already decrements; the two cumulative counters are removed.
RED: new test (limit 2, write two, activate, write a third while the
first drains) got QUEUE_OVERFLOW before the fix; it now buffers, and a
fourth write that really exceeds the current buffer still fails closed.
Gates: architecture, types app/test, lint, adapter-inventory, test:unit,
test:component pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
reconnect-coordinator run() repeated the same attempt teardown (settle
the barrier gate, close the opened session, clear `session`, abort and
dispose the phase) at about fifteen exits. It is now one endAttempt(gate,
session) closure; each exit keeps its original order (the one site that
closed a session before settling the gate still does so before calling
endAttempt). The interrupted-attempt exit also goes through endAttempt;
its extra phase abort is a no-op because the phase was already aborted.
Exits that drain before disposing, or that already closed the session,
are left as they were so no session is closed twice.
websocket-connection failure()/connectionFailure() ignored their 4th
`_recovery` argument, which 40 call sites passed. The parameter and the
arguments are removed; both helpers are module-private, not a port.
No behavior change, so the existing reconnect and WebSocket suites
are the safety net. Gates: architecture, types app/test, lint,
adapter-inventory, test:unit, test:component pass. Realtime bundle
39433 -> 39354 gzip.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Realtime kept three DataSnapshot readers and about ten hand-written
abort/timeout races, and two parsers had drifted copies in other adapters.
Snapshot: reconnect-coordinator and bounded-poll now use result.ts
captureRealtimeDataSnapshot (+ `source`, variadic hasExactSnapshotKeys).
Kept the result.ts rules, the stricter ones: Object.hasOwn instead of
`"value" in` (prototype pollution), enumerable fields only, keys read from
one descriptor snapshot, frozen computed at snapshot time instead of a
later Object.isFrozen. Poll therefore now rejects non-enumerable fields;
it also accepts null-prototype records, which the other two copies
already did (pure data, no inherited accessors). Characterization test
added for both in bounded-poll-coordinator.test.ts.
Race: one raceRealtimeTask(task, signal, deadline) in result.ts, built on
kernel createAbortableOperation, replaces fetch-sse timed/cancelReader,
reconnect race/drain (the drain keeps a host timer via timerDeadline,
since the clock itself may hang), poll awaitTaskOrAbort/awaitWithinLease,
live-poll checkpoint + two quiescence races (now one), stream-coordinator
deadline/recovery-wait/close-drain, and the two WS inline races. Kept
semantics: deadline armed first and wins a tie (poll lease), task still
starts when the clock throws (reader cancel), collaborator rejections
still reach stream-coordinator callers, scheduler install failure still
fails closed as a timeout. Small change: nothing starts once the signal is
already aborted (WS apply / SSE onEvent / reader.read could run just
before being reported ABORTED). The mapping adds one microtask per race,
so the reconnect test flush() helper gets 16 turns instead of 8.
Kernel: parseRetryAfterDelay moves to platform/retry-after.ts (realtime
re-exports it; http retry-policy parseRetryAfter now delegates, so it
rejects "", " ", "1e3", "1.5", "0x10" and non-IMF dates it used to
accept). json-member-scanner moves to platform/; web-push push-codec uses
it instead of its top-level-only duplicate-key check, so nested
duplicates are rejected too. INVENTORY 136 -> 137, detail tables updated.
Tests: RED for http strict Retry-After, web-push nested duplicate, poll
snapshot rules; new direct raceRealtimeTask tests. Gates: architecture,
types app/test (+ service-worker/web-worker), lint, adapter-inventory
(abort ratchet 21/21), test:unit, test:component, realtime-boundaries,
integration/platform all pass. Realtime bundle 40346 -> 39433 gzip.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
isReconnectAttemptResetEligible returned true on either a valid
heartbeat/event or stableOpenMs elapsed. A server or proxy that sends one
`:` comment and drops kept attemptIndex at 0, so the delay stayed in
[0, baseDelay) and maxAttempts/maxElapsed were never reached (many
clients reconnecting in lockstep). Both conditions are now required. The
reconnect coordinator uses this helper at both close paths, so no
coordinator change is needed. The two design docs that stated "or" now
say "and".
Existing tests encoded the old "either" rule (policy unit test and the
coordinator "resets a spent retry" test) and are rewritten. RED: the
policy test (heartbeat-only returned true) and the coordinator test
(heartbeat on an instant close still reset: 3 attempts, expected 2)
failed before the fix. Gates: architecture, types app/test, lint,
adapter-inventory, test:unit, test:component pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A full dedupe window (2,048 IDs / 4 MiB) used to trigger a full
DEDUPE_OVERFLOW recovery on every event past capacity, so a steady
3.4 events/s stream re-fetched snapshots and reconnected every window.
The coordinator now drops the oldest entry (Map insertion order equals
apply order) and keeps applying; a replay pushed out of the window still
drops as STALE_EVENT through the sequence check. The design doc row and
the maxDedupeBytes contract comment now say eviction instead of recovery.
The semantic fingerprint is now a 16-char 64-bit FNV-1a hash of the
canonical envelope instead of the canonical text itself (up to 64 KiB
each). The canonical-size EVENT_TOO_LARGE check only bounded that stored
text, so it goes too; the wire-size check stays. pruneDedupe stops at the
first unexpired entry instead of scanning every entry.
The old "recovers instead of evicting" coordinator test encoded the bug
and is replaced by an eviction test. RED: the eviction test (got
RECOVERED/DEDUPE_OVERFLOW) and a fixed-size fingerprint codec test (got
3326 vs 333 bytes) failed before the change. Added a TTL prune
characterization test (passes before and after). Gates: architecture,
types app/test, lint, adapter-inventory, test:unit, test:component pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A single reader chunk with more than maxItemsPerChunk (256) items used to
fail the stream with non-retryable QUEUE_OVERFLOW, killing normal replay
bursts even though maxChunkBytes already bounds memory. The parser now
stops at the cap, keeps the remaining decoded text, and returns it through
a new additive drain() call; fetch-sse-connection drains after every push
before reading again. Pushing or finishing with undrained text still fails
closed (QUEUE_OVERFLOW), so buffered text stays bounded by one chunk. A
lone CR followed by another line now re-checks the cap before continuing.
COMMENT and empty item lists reuse frozen constants.
The old parser test that expected QUEUE_OVERFLOW for 3 comments with a
cap of 2 encoded the bug and is replaced.
RED: new parser drain test and a 300-event single-chunk connection test
failed with QUEUE_OVERFLOW before the change. Gates: check:architecture,
check:types:app/test, lint, check:adapter-inventory, test:unit,
test:component all pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Composition checked authProfileId against the module-global installed
registry, the V3 executor used its injected registry, and the V2 client
resolved against the raw REST_AUTH_PROFILES record (or an injected raw
record) that was never installed. A BEARER_HEADER profile with
`credentials: "same-origin"` is refused by installation but resolved
fine.
- resolveRestSecurityProfiles / validateRestProfileBindings take
`RestAuthProfileSource` (installed registry or raw record) and default
to INSTALLED_REST_AUTH_PROFILES. A raw record is installed (validated)
before lookup; lookups go through the registry, and the CSRF record
lookup uses Object.hasOwn.
- composeContractContributions takes the registry as an optional
second parameter (default INSTALLED_REST_AUTH_PROFILES).
- src/adapters/http/client.ts: only the V2 resolve call site (and its
import) now falls back to INSTALLED_REST_AUTH_PROFILES.
Verified: 3 new cases in tests/unit/rest-profile-contract.test.ts were
RED before (incoherent raw record accepted, installed registry not
resolvable, composition ignored the given registry); gates green except
the known removal-fixture test.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
V2 checked the content type and decoded the body before looking at the
status. An nginx 502 with an HTML page became CONTENT_TYPE_MISMATCH and
an empty or broken 503/504 JSON body became MALFORMED_JSON: both
non-retryable and shown as a contract breach. For a 5xx/408/425 status
these branches, and the invalid-envelope branch, now return
SERVER_FAILURE with the httpStatus, the same R-3.3 rule V3 uses.
Retry follows V2's existing SERVER_FAILURE rule in retry-policy.ts, so
502/503/504 retry and 408/425/500 do not; SERVER_FAILURE was already in
retryKinds, so retry-policy.ts is unchanged.
The existing "non-JSON response" test used a 502 and pinned the old
CONTENT_TYPE_MISMATCH classification; it now uses a 200, which still
covers the non-JSON rejection and body redaction.
Verified: four new http-client cases (502 HTML, 503 empty JSON, 504
broken JSON, 408 text) failed RED with CONTENT_TYPE_MISMATCH /
MALFORMED_JSON and now pass with the expected fetch counts. All lane
gates pass except the known removal-fixture test; test:reference-feature
timed out once under load and passed on rerun.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- cursor pagination re-serialized every accumulated item on each page
(O(pages x total)) and used items.push(...page.items), which throws
RangeError for a large page, so loadAll rejected instead of returning
a Result. Each page is now measured once and summed (errs high by at
most one byte per page), the item count is checked before copying,
and items are appended one by one.
- conditional-validator-store accepted `W/"` as a weak ETag because its
opening quote also satisfied endsWith('"'). The closing quote must now
come after the opening one.
- browser-rpc-runtime's validTransportFailure had no callers, duplicated
decodeTransportFailure and read properties directly, which RPC-RR-04
forbids. Deleted (approved for this template).
Verified: new tests failed RED (18 serialized items instead of 6 because
the accumulated array was passed each time, RangeError for a 1,000,000
item page, `W/"` installed) and pass now. All lane gates pass except the
known removal-fixture test.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The V2 client only attempted 401 session recovery while the physical
attempt count was under maxRetryAttempts + 1. With MAX_RETRY_ATTEMPTS=0,
or when a 503 retry had already used the budget, a recoverable 401 went
straight to onUnauthenticated() and logged the user out. Recovery is
already bounded to one replay by recoveryUsed and by the deadline, so
the budget check is removed.
The comments in runtime-adapters.ts and adapters/http/index.ts said V2
works as a rollback path for V3 once its four slots are injected. That
is wrong: V2 validates the {success,data,meta} envelope and V3 the bare
DTO, so the same backend fails every V2 response with
ENVELOPE_MISMATCH. The comments and the client.ts ledger row in
docs/reviews/adapters/01-network-and-state.md now say a rollback also
needs a V2-envelope server. No docs/architecture page made the claim.
V2 code is kept.
Verified: two new auth-recovery cases (retries disabled; 503 then 401)
failed RED with ok:false and now recover. All lane gates pass except
the known removal-fixture test; verify:documentation passes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Three gaps in the unary/stream runtime:
- A KEYED_COMMAND/NON_REPLAYABLE unary that failed after the transport
was invoked (timeout, throw, UNAVAILABLE, ...) came back retryable
with action "retry", although the server may have applied it. Every
failure of a COMMAND after send now carries effect MAYBE_APPLIED
(retryable false, contact-support). Pre-send rejections and queries
are unchanged.
- A retry did not re-check the scope generation, so after a logout the
previous user's command could be sent again. Each retry now checks
safeIsCurrent first and ends with SCOPE_GENERATION_CHANGED.
- A stream whose consumer aborted but never called return() stayed
suspended at a yield; its finally never ran and the operation stayed
RPC_STREAM_DRAINING for good. The lease is now released (cancel +
positive-close subscription) from the stream's own abort listener;
finally triggers the same path through its abort.
Verified: five new tests failed RED (no effect, second attempt sent,
cancel never called) and pass now; a pin that pre-send command failures
and query failures stay effect-free passes before and after. All lane
gates pass except the known removal-fixture test.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
TanStack 5 invalidateQueries resolves only after the active refetches
finish. The coordinator awaited each namespace in a for loop, so one
invalidation (and lease.release(), which waits on it) took the sum of
every namespace's refetch time. The namespaces now run through
Promise.allSettled; each rejected one is still reported once as
"invalidate".
Verified: new test holds the first namespace's refetch open and failed
RED (second invalidateQueries never started); it now passes and checks
that a failure is reported once. All lane gates pass except the known
removal-fixture test; provider-guardian-transaction timed out once under
load and passed on rerun.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A retried command reported only its last attempt's effect. An IDEMPOTENT
command whose first attempt was lost (network error or 429) and whose
retry got a NOT_APPLIED problem ended as "not applied, retry", although
the first attempt may have applied. A 5xx with an HTML or empty body is
MAYBE_APPLIED for a command, yet the executor still replayed it.
The executor now records what each retried attempt settled on and joins
it into the final outcome's effect. An admission whose own certainty is
MAYBE_APPLIED (or APPLIED_CONFIRMED) is not retried; a 429 still is,
because its admission reason is RATE_LIMITED rather than an effect
verdict. The logical accumulator is not reused because it is raised at
dispatch and cannot tell a rejected attempt from a lost one.
Verified: three new tests (lost response then 400, 429 then 400, 504
empty body) failed RED with NOT_APPLIED / a second attempt and now pass.
All lane gates pass except the known removal-fixture test; two component
tests timed out once under load and passed on rerun.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The executor decided the terminal outcome of an ended wait in six places
(credentials, lease, dispatch, transport backoff, admission, retry
backoff). The two backoff copies only checked the caller, so a scope
fence during a retry sleep ended as TRANSPORT_FAILURE{TIMEOUT} and the
user saw "timed out, retry" after logging out.
All six sites now call one terminalOutcomeFor(phase) helper: the caller
cancels, the deadline times out, a scope fence during a wait is
ABORTED_BY_SCOPE and during admission stays the SCOPE_FENCED violation.
Verified: a new owner x phase table in http-execution-v3.test.ts pins
the 16 existing combinations and failed RED only for the two backoff
scope-fence cases (TIMEOUT instead of ABORTED_BY_SCOPE). All lane gates
pass except the known removal-fixture test (missing artifacts/).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The exact media-type regex lived in object-url-lease.ts,
browser-file-policy-registry.ts and the download validation;
safeOpaqueValue (vault) and safeGeneratedToken (download) were the same
regex; and five modules each defined their own positive-safe-integer check
(isPositiveByteLimit included). file-policy.ts now exports one copy of each:
EXACT_MEDIA_TYPE, isSafeGeneratedToken, isPositiveSafeInteger.
ObjectUrlLeaseRegistry.create also loses an inner revoke branch that could
never run (it required a non-empty string inside a "not a non-empty string"
guard).
cache-storage/public-cache-policy.ts keeps its private positiveSafeInteger:
adapters may not import another adapter's module.
Verified: a characterization test (unsafe reference -> UNAVAILABLE, 128-char
token accepted, empty object URL -> no revoke, fractional/zero limits and a
wildcard download media type rejected) passes on the code before and after.
Lane gates pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
public-response-cache-adapter.ts (1986 lines) and download-delivery-adapter.ts
(1491 lines) mixed the facade with codecs and validators. Moved verbatim:
- cache-storage/public-cache-codec.ts: manifest normalization and canonical
digest input, marker/active-pointer validation, bounded control JSON,
owned cache names, CacheValidationFailure -> Result mapping.
- cache-storage/public-cache-response-validation.ts: response admission
(status, directives, Content-Type, Vary, bounded body, digest).
- browser-files/download-input-validation.ts: request snapshot, input/limit
checks, presigned and browser-managed capability validation, handoff target.
The adapters drop to 1164 and 1157 lines.
Two non-verbatim edits: validateDownloadInput takes a structural
DownloadInputValidationOptions instead of DownloadDeliveryAdapterOptions (no
import cycle), and readBodyBounded races reads on the kernel
createAbortableOperation instead of its own abort listener. Moving it out of
a primitive-importing file would otherwise have raised the hand-rolled abort
ratchet to 20/19; the abort outcome (ABORTED, reader cancelled) is unchanged.
Verified: characterization tests added first and passing before the move
(cache manifest/response/control-JSON failure codes, lookup abort during a
pending body read; download input validation codes); a line diff confirms the
moved code is verbatim apart from the two edits above. Inventory 139/139,
lane gates, check:browser-file-storage-boundaries pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
sanitizeSuggestedFileName dropped one code point at a time and re-encoded the
whole name each round, so a 100k-character server-supplied name took ~10^10
steps. truncateUtf8 now walks the code points once, counting UTF-8 bytes, and
cuts at the longest fitting prefix (trailing whitespace trimmed, as before).
The fallback base name uses the same helper. deliver() also sanitized the
name twice (validation, then again for use); validateDownloadInput now
returns the sanitized name and deliverActive uses it.
Verified: new unit test pins results for ASCII, Hangul, emoji and the
whitespace-trim case, and asserts a 5k-character name needs <10
TextEncoder.encode calls (RED before: 4827). Lane gates pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
createBrowserCrossContextInvalidationFromHost returned undefined without any
diagnostic when the cache epoch was invalid or secure random was missing, and
bootstrap built the epoch as `release.${releaseId}`, so any release ID with
`+`, `/`, `@` or over the length limit silently turned cross-tab
invalidation off.
- The host factory now reports OPEN/FAILED/NONE with the new reasons
INVALID_CACHE_EPOCH or SECURE_RANDOM_UNAVAILABLE before returning
undefined (the zero-topic build still probes nothing and reports nothing).
- New crossContextCacheEpochForRelease() derives `release.<fnv1a-64 hex>`
from the release ID; bootstrap uses it (one call-site change in
src/bootstrap/runtime-adapters.ts).
- The observation reason union gains two members (additive); the ops runbook
lists them.
Verified: two new unit tests failed before (no observation; missing epoch
helper) and pass now. Lane gates, reference-feature and V3 integration tests,
and check:diagnostics pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
NativeInputFilePicker installed the window-focus fallback on every select,
even in engines that dispatch a native `cancel`. A provider that returns
focus first and dispatches `change` only after fetching the file (Android
cloud providers) was reported as DISMISSED after the grace window. The picker
now remembers that it observed a native `cancel` and stops installing the
focus fallback for later selections.
Verified: new unit test (cancel seen, then focus + fallback timer + late
change) returned DISMISSED before and SELECTED now. Lane gates pass
(tests/component/router.test.tsx timed out once under full-gate load and
passed on rerun).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
download-delivery-adapter and browser-file-vault each carried their own
combineAbortSignals/awaitWithSignal copies. Two gaps followed:
- a presigned lease that resolved after the abort but before the delivery's
finally was seen by neither compensateLateSource (holder not yet closed) nor
closeHeldSource (lease never adopted), so it leaked;
- a writable from handle.createWritable() that arrived after the abort was
never aborted (Chrome keeps a .crswap file).
Each delivery and each vault stream now owns a createAbortableOperation joined
to the caller signal; dispose() closes the live owners. Every await goes
through owner.race, and late leases/writables are compensated by the race.
The hand-rolled abort ratchet drops 21 -> 19.
Verified: two new download tests (same-turn late lease, late writable) failed
before (close 0 times, no abort) and pass now; a new vault characterization
test (caller abort / dispose during a pending read) passes before and after.
All lane gates pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>