Bump the alkcall dependency to 0.2 (with the gateway feature) and
converge on the promoted shared pieces:
- The local dispatch spine (gateway/dispatch.rs, 721 lines) is deleted;
GatewayDispatch, schema_disclosure_denial, and DEFAULT_DEADLINE are
re-exported from alkcall::gateway (alkcall ADR-048). The 30 s default
deadline preserves the previous behavior exactly.
- gateway/schema_cache.rs (PublishSchemaCache) is deleted: alkcall CF-003
compiles publish_schema at registration time and exposes
OperationRegistry::publish_validator; the /publish chunk stream
resolves against it. Un-compilable schemas are now rejected at
registration, so the two end-to-end fail-closed tests were reworked
into a registration-rejection test (a stronger guarantee).
- schema_disclosure_denial consumers (to_mcp, routes) use alkcall's
promoted implementation; the alkhttp-local copy is gone (ADR-071
updated: the guard stays as defense-in-depth, the implementation no
longer forks).
- CF-001: from_wss drop monitor and the WS overlay tests use
CallError::connection_closed; the review-001-ws-eof-signal race tests
now assert retryable CONNECTION_CLOSED on both resolution paths (the
tolerated non-retryable INTERNAL write-failure outcome is gone).
- Added CHANGELOG.md (Keep a Changelog), Unreleased section records the
bump and convergence.
Verification: cargo test default 453 ok, wss 470 ok, mcp 526 ok,
all-features 575 ok; clippy -D warnings clean (default + all-features,
all-targets); fmt clean; cargo doc warning-free.
Net: -1093 lines.
- forward_stream build-error arm (forward.rs): wire-level test asserts
one INVALID_INPUT envelope then stream end with zero upstream
contact (a panicking responder counts as the contact guard), plus
the from_jsonschema integration mirror (undeclared key + non-scalar
placeholder, each naming its rejection source)
- PEM read-failure arms (http_client.rs): nonexistent CA path →
CaBundleRead (sync new), nonexistent client-cert path → ClientCertRead
(async reload, prior generation retained)
- Over-cap poll_write rejection leg (byte_adapter.rs): cap+1 write →
InvalidData naming the cap; stream stays usable for an at-cap write
afterwards
- SSE parser edges: CRLF split across feed chunks frames one line;
invalid-UTF8 data lines drop without killing the frame stream
- from_value structural rejects: non-object doc, missing `info`,
missing `paths`, non-object `paths` each name the member
- Connection-failure arms: accept-path ConnectionClosed →
HandlerError::ConnectionClosed via stream_error_to_handler; read-pump
demux-gone break ends the pump when the byte-stream side is dropped
- Delete the caller-less `impl Default for WsTimeouts` (the extension
is constructed explicitly)
cargo llvm-cov --all-features: all named arms covered; TOTAL regions
94.18% (was 93.86%), lines 96.04% (was 95.77%); http_client.rs
86.56% lines (was 81.72%).
docs(tasks): mark review-002-fu-stream-error-coverage done
- corrupt CA bundle section -> CaBundleParse naming the path (the
pure-non-PEM variant parses as zero sections and is accepted by
reqwest, so the arm needs structurally broken PEM)
- garbage client identity -> ClientCertParse naming the cert path,
with the error asserting key material is never echoed
- config() reflects a reloaded CA bundle on the TLS path (FWD-12)
GW-16: empty body / malformed first line / missing header fields /
per-line cap / batch over-cap rejections now route through
call_error_to_http_response_with_identity, mapping INVALID_INPUT to
422 — same status as mid-stream chunk errors. One error class, one
status.
GW-17: invoke_sink wraps the registry sink invoke in the same 30 s
tokio::time::timeout the Once-op invoke uses; a hung sink handler
surfaces as a TIMEOUT (504, retryable) error envelope instead of
holding the HTTP connection forever. The sink wrapper bounds the
whole dispatch (chunk pacing included), matching http-server.md's
deadline contract.
Docs: http-server.md error table documents the 422 triggers and the
sink deadline; http-adapters.md batch cap status corrected.
to_openapi: gateway spec version 1.3.0 -> 1.4.0 (ADR-045 minor):
/publish framing faults and /batch cap reject documented at 422 (the
400 slots moved with the runtime); /publish 400 slot removed; 504
description covers the sink dispatch.
Verification: scripts/verify.sh OK (397 tests); cargo test
--all-features OK (513 tests); clippy --all-features --all-targets -D
warnings OK; cargo fmt --check OK.
- new tests/retry_policy_wire.rs: POST+500 = 1 upstream hit (method
gate), GET 500/503/200 = 3 hits (retry-to-success), budget exhaustion
stops retries under a 50-attempt cap (wall-time + hit-count bounds)
- forward_stream terminal arms on the wire: oversized SSE line ->
one INTERNAL terminal envelope; mid-stream transport abort (staged
via a notify gate so the abort is genuinely mid-stream) -> terminal
envelope after the delivered frame; pending event EOF-flush; dead
port through both forward() and forward_stream()
- Once-path decode arms: malformed application/json 200 -> INTERNAL
decode envelope; application/octet-stream 200 -> byte-array envelope
- COV-12 credential arms: ApiKey and Basic malformed values fail
loudly without echoing secret material; declared header-param
invalid name/value rejections
- new tests/client_config_reload.rs: config() reflects a reloaded
config (FWD-12 atomicity half)
Wire-level round trip through the real rmcp server: a scalar input
reaches the remote tool as {"value": <input>}, matching the
value_to_json_object wrap.
peer.call_tool through the real rmcp streamable-HTTP transport for
schema/call/batch/unknown — the production ServerHandler::call_tool
routing shell, previously only exercised via the invoke_tool bypass
for search.
Verification: cargo test --features 'mcp test-support' --test full_surface
A logical request's Retry-After waits are now bounded by
max_total_retry_duration, and a retry storm can no longer re-arm a
full ceiling per attempt:
- BudgetClock anchored per logical request by RetryGateMiddleware
(into the request Extensions, shared across retry attempts) and
read by the inner RetryAfterMiddleware every attempt; the monotonic
anchor projects the hard stop through wall-clock steps.
- maybe_sleep_for truncates the sleep to the remaining budget;
a spent budget skips the sleep entirely.
- record() keeps the EARLIEST deadline per URL (retry storms cannot
extend the first-seen deadline); a refresh that cannot make it
under the budget hard stop drops the entry so the next attempt
starts immediately instead of parking.
- Middleware without a budget anchor (budget = 0) keeps the prior
semantics; the shared client now wires
HttpClientConfig.max_total_retry_duration into the Retry-After
middleware.
- Wire tests (tests/retry_after_budget.rs): always-429 responder with
a 300 s Retry-After is bounded by budget + one attempt; separate
logical requests still honor the recorded throttle window.
- FWD-12 atomic reload pairing and the FWD-15 stream-client split
untouched (stack built in build_client_with_pems for both).
- WsTimeouts { idle, write } request extension mirrors ChannelsPolicy:
a deployment layers it on a WS route (bare-registry routes included)
to set the pump knobs per route
- precedence: extension present replaces the router state entirely;
absent falls back to SessionState (adapter-configured idle) and the
crate default write window — a Default impl never clobbers the
adapter-configured idle knob
- upgrade.rs module + handler docs now state the real defaults for
bare-registry routes (60 s idle + 60 s write, 64-session semaphore,
handler-private WsSessions) and the extension surface
- split_ws_to_bytes_idle_with_write exposes the WS-18 write window to
run_channels_session; acceptance test drives a bare-registry route
with a 150 ms extension idle window (1001 eviction observed)
Verification: scripts/verify.sh OK (343 passed), test-support suite
ok, clippy -D warnings clean, fmt clean
Mirrors the WS-13 progress semantics onto the axum upgrade path with
three integration tests over the real HttpAdapter surface (also the
COV-11b dark-knob gate — with_ws_idle_timeout had no test caller):
the forever-dribble client is evicted with 1001 despite arriving
messages; a productive session round-trips calls across many windows
without eviction; the None knob never evicts a dribbling client.
Replace rmcp's unbounded Peer::list_all_tools with a bounded walk over
list_tools: hard cap of 100 pages (MCP_MAX_TOOLS_LIST_PAGES) plus a 60 s
overall deadline (MCP_TOOLS_LIST_DEADLINE). Tripping either budget fails
loudly with AdapterError::DiscoveryFailed naming pages fetched and tools
accumulated — no silent truncation, no partial registration (import
fails closed, as before). A slow or hung page is cut off by a per-page
tokio timeout sized to the remaining budget.
Bounds are documented in the module doc. Integration test added: a
cycling-cursor server (next_cursor always Some("a")) terminates with
the clean budget error inside an outer 10 s guard; the existing 3-page
pagination test is unchanged and passes.
Verification: cargo test (304 pass), cargo test --features mcp,
cargo test --all-features (412 pass), clippy -D warnings both default
and --all-features --all-targets, cargo fmt --check, cargo doc --no-deps.
- /search, /schema: document the envelope wrapper and the real item/spec
fields (PRJ-01/02); /search drops unreachable 401/403, documents 404
(PRJ-15)
- error statuses: 422 for dispatch-path INVALID_INPUT /
INVALID_OPERATION_TYPE; extractor 400s documented as the plain-text
gap they are (PRJ-03); operation-declared errors projected by
http_status with x-runtime-behavior: 500 on non-HTTP_* codes (PRJ-04
project-honest decision) — no runtime changes
- /subscribe: 200+SSE only; event:error terminal contract documented
(PRJ-05, GW-12)
- components for requests/responses; CallRequest no longer inlined
per-path (PRJ-14); all library expect() paths removed (PRJ-11)
- error projections folded into BTreeMaps: same registry =>
byte-identical doc, sorted enums (PRJ-12)
- components.securitySchemes.bearerAuth + top-level security (PRJ-15)
- info.version 1.1.0 -> 1.2.0 (ADR-045 minor: additive documentation of
the settled runtime contract)
- 31 unit tests incl. golden print-level assertions mirroring the routes
tests' actual bodies and a determinism test
Verification: cargo test (288), clippy --all-targets -D warnings, fmt
--check, cargo doc --no-deps, cargo test --all-features (all green in a
clean worktree at HEAD; shared tree carries parallel agents' edits).
Full-surface integration suite (tests/full_surface.rs, mcp feature):
- one HttpAdapter over real TCP (ProtocolHandler::handle path) serving
gateway endpoints, /openapi.json, /mcp, and the WS channels session
- gateway: search/schema/call/subscribe/batch/publish presence,
envelope shapes, error fidelity end-to-end
- from_openapi import -> Internal-by-default invisible from the wire ->
External facade composes it via env.invoke -> upstream HTTP API
called end-to-end (ADR-015 composition model exercised)
- to_openapi 6-path doc validated against openapiv3 over the wire
- to_mcp: MCP client connects to /mcp on the served adapter, lists the
4 gateway tools, search returns ACL-filtered ops (Sub excluded)
Production fix: the WS upgrade route was reserved but never wired into
HttpAdapter's router (the ws-upgrade-session tests built their own
router). Now wired with ws_bearer_auth (401 without a resolvable
token) around ws_upgrade_handler.
Docs sync: all 28 'Port notes' sections/blockquotes stripped from
ported ADRs/specs; OQ-01/OQ-02 statuses corrected to resolved in
overview.md, websocket.md, and the README table (open-questions.md was
already current).
Publish prep: cargo publish --dry-run --allow-dirty succeeds;
cargo doc --no-deps warning-free (ADR link targets fixed); feature
combinations (default / test-support / mcp / wss / all) compile
warning-free under clippy -D warnings.
Verified: cargo test (182 lib default), --all-features (227 lib + 29
integration), clippy -D warnings x3 feature sets, fmt, doc,
publish --dry-run.
Ported the alknet-http overlay verification to the channels-over-WS
session (tests/ws_overlay_ops.rs, test-support feature, 8 tests):
- overlay mechanism: browser-registered ops land in the connection's
Layer 2 overlay (register_imported), exposed via overlay_env() —
no PeerIds (browsers are not peers); PeerRef::Specific to a browser
id routes to nothing (NOT_FOUND)
- hub→browser call through compose_root_env's attached overlay
- AccessControl on browser ops gates hub calls (scope match allows,
missing scope FORBIDDEN)
- overlay dies with the connection; no leak between connections;
in-flight calls to browser ops resolve on close
- wire-level: 10 interleaved concurrent calls across two WS sessions
— no cross-correlation, no deadlock; disconnect mid-call resolves
and a fresh session works (no listener wedge)
byte_adapter: read_eof Notify now gated to the wss feature (its only
consumer is from_wss) so a test-support-only build is warning-free.
Verified: cargo test (182 lib), --all-features (227 lib + 5 MCP + 8
overlay + 10 WS integration), clippy -D warnings (default,
test-support, all-features), fmt.
glm-5.3-flash