Use *string for the RequestId field in ErrorResponse so that an empty
string is serialized (not omitted by omitempty), while nil — used for
fault events — stays omitted. Fixes test_lambda_runtime_error snapshot
mismatch where requestId: "" was expected but absent.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

- Move Init Duration after Max Memory Used in REPORT line (matches AWS)
- Add Status: timeout to REPORT line on invoke timeout
- Fix timeout error message format to "RequestId: <id> Error: Task timed out after N.00 seconds"
- Add ErrorType: "Sandbox.Timedout" to timeout error response
- Track init start time and emit Init Duration on first non-retry invocation
- Add is-init-retry field to InvokeRequest to suppress Init Duration on retry invokes

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

- Close response bodies in SendStatus/SendLogs/SendResult so idle
  connections are released instead of leaked.
- Use errors.New instead of fmt.Errorf with no format arguments.
- Document the single-invoke assumption behind the unsynchronized
  initStart/warmStart fields.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Resolve the fault request ID in the events API: prefer an explicit ID,
then the current invoke ID so a mid-invocation runtime crash reports the
actual request, and only synthesize a UUID as a fallback for init-phase
faults where no invocation has been dispatched yet. Previously the
supervisor always passed a random UUID, masking the real invoke ID.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

…pidcore

Move the Lambda init-phase timeout retry into the RIE and replace the custom
supervisor with rapidcore's existing init-failure machinery.

- rapidcore: add AwaitInitializedWithDetails (structured init outcome) and
  AwaitInitializedWithTimeout (timer-aware; does NOT consume the init-failures
  channel on timeout, so the invoke path's Reserve() can still drive suppressed
  init). Refactor awaitInitialized into interpretInitFailure.
- main.go: on init-phase timeout (LOCALSTACK_INIT_PHASE_TIMEOUT, default 10s),
  emit INIT_REPORT, signal ready, and reset the in-progress init so the first
  invoke re-runs it (suppressed init) under the function timeout. Genuine init
  failures are reported via SendInitError.
- custom_interop.go: SendInitError crash-path fallback with an initErrorForwarded
  dedup guard, formatted as AWS's "RequestId: <id> Error: <msg>"; ReportInitTimeout
  + initTimedOut-driven Init Duration suppression.
- Remove the custom LocalStackSupervisor/LocalStackEventsAPI; drop unused
  IsInitRetry from lsapi.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

The synthetic START line was written eagerly when LocalStack dispatched /invoke,
i.e. before the suppressed init re-runs the function's static code. AWS emits
START upon the Invoke event reaching the runtime, which rapidcore sequences after
any inline (suppressed) init (doInvoke -> sendInvokeStartLogEvent).

Emit START from a minimal LocalStackEventsAPI.SendInvokeStart override (riding
rapidcore's correctly-placed invoke-start event) and drop the eager write in the
/invoke handler. Correct for warm, cold, and suppressed-init invocations; fixes
test_lambda_timeout_init_phase against the unmodified AWS snapshot.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…t it

Follow the established LocalStack-env strategy: capture the value in InitLsOpts
(before UnsetLsEnvs runs) and add it to the UnsetLsEnvs list. Previously it was
read inline with os.Getenv after UnsetLsEnvs, so the variable was never unset and
leaked into the function's environment (forwarded via os.Environ in InitHandler),
breaking AWS parity.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…ckTrace)

SendInitErrorResponse round-tripped the runtime's /init/error payload through a
typed struct whose stackTrace used omitempty, dropping an empty-but-present
"stackTrace": [] (as AWS emits for Runtime.HandlerNotFound). Decode into a map and
only inject requestId, forwarding the runtime's fields verbatim.

Fixes test_lambda_handler_not_found.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

AWS folds a failed cold-start init into the first invocation (suppressed init),
reporting it as a failed invoke with the full INIT_REPORT(phase=invoke)/START/END/
REPORT envelope rather than a separate init error. Match this for on-demand:

- main.go: on init failure for on-demand, signal ready and keep the process alive
  instead of SendInitError+exit, so the first invoke surfaces the cached init error.
- custom_interop: skip /status/error forwarding for on-demand (cache only, so the
  invoke carries the error); emit INIT_REPORT Phase:invoke Status:error Error Type
  before START and add Status/Error Type to the REPORT, using rapidcore's scrubbed
  fatal error type. PC/SnapStart/Managed Instances keep the provisioning-time model.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

The runtime emits multi-line records (e.g. an unhandled-init traceback) as a single
log frame with internal newlines replaced by bare carriage returns. AWS renders
these back as line feeds, so convert bare CR to LF in the assembled log output while
preserving genuine CRLF endings (which AWS keeps, e.g. the LAMBDA_WARNING line).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…elper to its own file

The init-phase timeout / suppressed-init support previously edited the vendored
upstream internal/lambda/rapidcore/server.go (exported InitCompletionResponse,
extracted interpretInitFailure, added AwaitInitializedWithDetails and
AwaitInitializedWithTimeout). Modifying vendored upstream files causes rebase
conflicts when syncing with aws-lambda-runtime-interface-emulator.

Revert server.go to byte-identical upstream and move the only load-bearing
addition (AwaitInitializedWithTimeout, plus a local InitCompletionResponse and a
duplicated interpretInitFailure) into a new same-package file
server_localstack.go. It must stay in package rapidcore because it needs the
unexported getInitFailuresChan()/Release()/setRuntimeState() helpers, but as a
standalone file it never conflicts on rebase. AwaitInitializedWithDetails was
unused by cmd/localstack and is dropped.

No behavior change: RIE go test ./... and TestLambdaErrors both green.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Commit 9b54e66 rewrote every bare carriage return in the captured runtime output to
a line feed, intending to render multi-line init tracebacks across lines. But AWS
keeps a bare CR inside a single CloudWatch log event (records split on LF only), and
LocalStack's log ingestion likewise splits on "\n". The conversion therefore wrongly
split any record containing a bare CR into multiple events, breaking the AWS-validated
TestCloudwatchLogs::test_multi_line_prints (a user `print("a\rb")` was emitted as two
events "a" and "b" instead of one event "a\rb").

Emit the runtime output verbatim. Verified: test_multi_line_prints and the full
TestLambdaErrors suite (incl. the runtime-exit/segfault error-reporting tests) are
both green.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

… rapid's init duration

Reporting fixes for the on-demand folded-into-invoke (suppressed init) path:

- The invoke handler no longer unconditionally forces error status from the
  recorded init failure: the REPORT Status/Error Type lines and the /error
  routing now apply only when the invocation actually failed, and the
  recorded failure is cleared once an invocation succeeds. Previously one
  init failure permanently tainted a recovered environment - every later
  (successful) invocation was posted to /error with an error REPORT line.
  Persistent failures keep re-emitting the full failure envelope per invoke.

- RecordInitError captures the failure type detected by rapidcore for
  runtimes that crashed WITHOUT calling /init/error (sys.exit, segfault,
  invalid entrypoint), so those failures render the same
  INIT_REPORT(phase=invoke) and REPORT Status/Error Type lines as the
  /init/error-reported flavor instead of a healthy-looking envelope.

- REPORT's Init Duration now uses rapid's authoritative Init-phase
  measurement captured from the INIT_REPORT(phase=init) lifecycle event,
  instead of wall-clock time.Since(initStart) at invoke arrival, which
  wrongly included the idle gap between init completion and the first
  invoke (minutes for provisioned concurrency). It is also omitted for
  non-on-demand invokes, matching AWS, which reports provisioned
  concurrency init in separate provisioning-time log streams.

- Document that SnapStart environments intentionally use the on-demand
  error model in LocalStack (they initialize lazily at the first invoke,
  not at version publish).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

… the suppressed init's real error

After the timeout path resets the in-progress init, awaitInitCompletion is
parked sending a ResetReceived failure on the unbuffered initFailures
channel. The first invoke's Reserve()/awaitInitialized() consumed it and
cached a generic placeholder error (Sandbox.Failure with an EMPTY payload,
see the ErrInitResetReceived handling in Invoke), which took precedence over
the real failure when the suppressed init re-run crashed without calling
/init/error: the invocation returned an empty error payload instead of e.g.
Runtime.ExitError "Runtime exited with error: exit status 1" (AWS-validated
by test_lambda_init_timeout_then_crash in localstack-pro).

Drain the notification right after the synchronous reset, so the invoke's
awaitInitialized() observes the closed channel, treats the init outcome as
pending, and the suppressed init's own result stays authoritative. This also
unparks the awaitInitCompletion goroutine for environments that time out
their init but never receive an invoke.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…drop rapidcore additions

Simplify the init error-reporting and init-timeout implementation with no
behavior change (TestLambdaErrors 12/12, timeout/PC/log-envelope tests, and
full test_lambda.py green against the rebuilt RIE):

* Delete internal/lambda/rapidcore/server_localstack.go: the timeout-bounded
  await is now a goroutine running the exported AwaitInitialized() plus a
  select in main.go. On timeout, receiving the goroutine's result after the
  reset doubles as the init-failure drain and orders its cleanup before the
  ready signal, so it can never cancel the first invoke's fresh reservation.
  No LocalStack-owned code remains under internal/.

* Render all INIT_REPORT flavors (cold-start failure, timeout, suppressed
  init) in LocalStackEventsAPI from rapid's native InitStart/InitRuntimeDone/
  InitReport events instead of three hand-rolled call sites. The suppressed
  init's Phase: invoke line now reports that init's measured duration after
  its own logs, matching AWS more closely.

* Record the scrubbed fatal error type natively from SendInitRuntimeDone
  (which fires even for crashes that never call /init/error), replacing the
  initErrorType mirroring; the cold-start Init Duration becomes take-once
  state, replacing warmStart/initTimedOut; wall-clock fallbacks are dropped
  because rapid's InitReport event always fires.

* Collapse /status/error reporting into a single post site
  (ReportInitFailure: forward the stashed /init/error payload, synthesize
  otherwise), removing the initErrorForwarded duplicate-send guard.

* Add unit tests for the events API rendering matrix.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

…tError

The recorded init error type was sticky across init attempts and cleared
cross-component by the invoke handler after a successful invocation.
Stickiness is unnecessary: every invocation into a failed-init
environment starts a fresh suppressed Init phase (rapidcore shuts the
runtime down after an init failure so the next FastInvoke re-inits), so
each failing attempt re-records its error type via its own
SendInitReport. Resetting the field in SendInitStart therefore keeps the
same behavior with one mechanism instead of two.

This also fixes a mislabeling bug: previously, when a suppressed init
re-run recovered but the invocation itself then died fatally
(ErrInvokeDoneFailed), the stale sticky error type tainted that
invocation's REPORT line with the original init failure.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…erver

The adapter was hoisted into main.go "so it can be shared with the
interop server", but nothing in main.go uses it after the constructor
call — main reaches the adapter through interopServer.localStackAdapter.
Revert to in-constructor creation and drop the extra parameter. Also fix
a stale comment in events.go referencing a CustomInteropServer.onDemand
field that does not exist.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Runtime-reported error payloads include a blank "requestId" (and
possibly an empty "stackTrace") on AWS, while platform-synthesized
payloads (Sandbox.Timedout, Runtime.ExitError) carry neither — both
validated by localstack-pro's AWS snapshots. Spell this out on
lsapi.ErrorResponse and adaptInitErrorPayload so the asymmetry between
the two ReportInitFailure paths is not mistaken for a bug and unified.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

PrintEndReports took pre-formatted "Init Duration: ..." and
"Status: ...\tError Type: ..." string fragments, splitting knowledge of
the REPORT line format between custom_interop.go and awsutil.go. Pass
the structured values (duration, status, error type) instead and render
the line in PrintEndReports only. Output is byte-identical.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

On-demand cold-start inits fold their duration into the first invocation's
REPORT line, but provisioned-concurrency / Managed Instances environments
initialize ahead of time and their invokes omit "Init Duration". AWS instead
emits a standalone INIT_REPORT line in the invoke-serving log stream carrying
only the duration -- no Phase, Status, or Error Type (those appear only on
failed or timed-out init lines).

SendInitReport previously emitted nothing for these successful non-on-demand
inits, so LocalStack's log stream diverged from AWS. Add the bare INIT_REPORT
line for that case and update the unit test accordingly.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>