mxaccessgw/code-reviews/Client.Go/findings.md

Code Review — Client.Go

Field	Value
Module	clients/go
Reviewer	Claude Code
Review date	2026-05-20
Commit reviewed	1cd51bb
Status	Reviewed
Open findings	0

Checklist coverage

#	Category	Result
1	Correctness & logic bugs	Re-review: previous Client.Go-001/003/007 remain resolved. New issue: a dead/no-op test condition in alarms_test.go (Client.Go-011).
2	mxaccessgw conventions	gofmt -l ./... and go vet ./... are clean. No new issues.
3	Concurrency & thread safety	New issue: runGalaxyWatch limit-reached path returns without waiting for the WatchDeployEvents goroutine to drain (Client.Go-013).
4	Error handling & resilience	New issue: direct err == io.EOF comparisons should use errors.Is for chain robustness (Client.Go-014).
5	Security	No issues found — TLS-by-default with TLS 1.2 floor, API key redaction in CLI JSON, no secret logging.
6	Performance & resource management	No issues found — defer client.Close() / defer subscription.Close() consistently applied across CLI and library; bench-read-bulk preallocates latency slice.
7	Design-document adherence	No new issues. The lazy grpc.NewClient + readiness probe migration (Client.Go-005) was applied uniformly to Dial and DialGalaxy.
8	Code organization & conventions	New issue: runWriteBulkVariant's secured parameter is computed but unused (Client.Go-015).
9	Testing coverage	Coverage holes from prior review now filled (Client.Go-008). fakeGalaxyServer.watchSendInterval is declared but never set — minor test cruft (Client.Go-016).
10	Documentation & comments	New issue: the CLI writeUsage line is missing the six bulk and bench subcommands now wired into run (Client.Go-012).

Findings

Client.Go-001

Field	Value
Severity	High
Category	Correctness & logic bugs
Location	clients/go/mxgateway/errors.go:88-93, clients/go/mxgateway/errors.go:117-128
Status	Resolved

Description: MxAccessError.Unwrap returns e.Command directly. EnsureMxAccessSuccess constructs &MxAccessError{Reply: reply} with Command left nil (the HRESULT / failing-MxStatusProxy path). When Command is a nil *CommandError, Unwrap() returns a non-nil error interface wrapping a nil pointer. Consequently errors.As(err, &ce) for *CommandError returns true while setting ce to nil — a caller writing the idiomatic if errors.As(err, &commandErr) { use commandErr.Status } nil-dereferences and panics. Verified empirically; the existing test only exercises the populated-Command path.

Recommendation: Make Unwrap return an untyped nil when Command is nil: if e == nil || e.Command == nil { return nil }; return e.Command. Add a test for the HRESULT-only MxAccessError asserting errors.As(err, &ce) is false.

Resolution: Resolved 2026-05-18: MxAccessError.Unwrap now returns an untyped nil when Command is nil, so errors.As no longer binds a typed-nil *CommandError; added errors_test.go regression coverage for the HRESULT-only and populated-Command paths.

Client.Go-002

Field	Value
Severity	Medium
Category	Error handling & resilience
Location	clients/go/mxgateway/session.go:440-516
Status	Resolved

Description: For the Events/EventsAfter compatibility API (cancelWhenResultBufferFull == true), when the 16-slot results channel is full sendEventResult cancels and returns false; the goroutine returns and close(results) runs — the consumer sees the channel close with no EventResult{Err: ...} ever delivered. A slow consumer cannot distinguish "stream ended normally" from "events were silently dropped." This contradicts the design doc's "libraries should not reorder, coalesce, or drop events by default", and a test currently pins this lossy behaviour.

Recommendation: Before cancelling on a full buffer, deliver a terminal EventResult carrying an explicit error (e.g. ErrEventBufferOverflow). Document the behaviour on Session.Events; steer callers to SubscribeEvents (which blocks instead of dropping).

Resolution: Resolved 2026-05-18: confirmed against source — on a full bounded buffer the compatibility path cancelled and closed results with no terminal result. Added the exported sentinel ErrEventBufferOverflow (errors.go); sendEventResult now, on a full buffer, cancels the stream then calls the new deliverTerminalResult helper, which evicts one of the oldest buffered events to make room and places EventResult{Err: ErrEventBufferOverflow} so it becomes the consumer's last item before the channel closes. The previously lossy regression test (TestEventsAfterCancelsStreamWhenCompatibilityChannelIsAbandoned) was re-pointed to assert the terminal ErrEventBufferOverflow result is delivered. clients/go/README.md now documents the bounded-buffer/overflow behaviour and steers no-loss callers to SubscribeEvents.

Client.Go-003

Field	Value
Severity	Medium
Category	Correctness & logic bugs
Location	clients/go/cmd/mxgw-go/main.go:517-532
Status	Resolved

Description: parseInt32List calls panic(err) when an item-handles token fails to parse as an int32. The CLI is a documented user-facing tool; a typo like -item-handles 1,foo crashes the process with an unrecovered panic and stack trace instead of returning a clean error and exit code 2 like every other validation path in main.go.

Recommendation: Change parseInt32List to return ([]int32, error) and have runUnsubscribeBulk propagate the error, matching parseValue's pattern.

Resolution: Resolved 2026-05-18: confirmed against source — parseInt32List called panic(err) on a malformed token. It now returns ([]int32, error), wrapping the bad token (invalid item handle %q: %w); runUnsubscribeBulk parses item handles before dialing and returns the error, so a typo flows through runWithIO to os.Exit(2) like other validation paths. Regression tests TestParseInt32ListParsesValidTokens and TestParseInt32ListReturnsErrorOnMalformedToken added to cmd/mxgw-go/main_test.go.

Client.Go-004

Field	Value
Severity	Low
Category	mxaccessgw conventions
Location	clients/go/mxgateway/alarms_test.go:153-154, clients/go/mxgateway/galaxy_test.go:58-59
Status	Resolved

Description: gofmt -l flags alarms_test.go and galaxy_test.go for misaligned struct-literal field padding. The Go client README lists gofmt as part of the workflow and the repo enforces style; unformatted committed code breaks gofmt-gated checks and CI.

Recommendation: Run gofmt -w mxgateway/alarms_test.go mxgateway/galaxy_test.go.

Resolution: Resolved 2026-05-18: confirmed gofmt -l . flagged both files for misaligned struct-literal padding. Ran gofmt -w on mxgateway/alarms_test.go and mxgateway/galaxy_test.go; gofmt -l . is now clean for the whole module.

Client.Go-005

Field	Value
Severity	Low
Category	Design-document adherence
Location	clients/go/mxgateway/client.go:64,68, clients/go/mxgateway/galaxy.go:83,87
Status	Resolved

Description: The client uses grpc.DialContext with grpc.WithBlock(). In current grpc-go both are deprecated in favour of grpc.NewClient (lazy connection). WithBlock also changes failure semantics: a transient gateway-unavailable at dial time becomes a hard Dial error rather than a connection that recovers when the gateway comes up, working against the design doc's resilience intent.

Recommendation: Migrate to grpc.NewClient; if a fail-fast connect probe is still wanted, do an explicit readiness wait bounded by DialTimeout, and update the doc comment.

Resolution: Resolved 2026-05-18: confirmed Dial/DialGalaxy used the deprecated grpc.DialContext + grpc.WithBlock pair. Migrated both to the shared dial(ctx, opts) helper, which now builds a lazy connection with grpc.NewClient and runs an explicit waitForReady readiness probe (Connect + WaitForStateChange until connectivity.Ready) bounded by DialTimeout — preserving fail-fast behavior while letting an otherwise lazy connection recover when the gateway is briefly down. Note: grpc.NewClient defaults the target scheme to dns, so the bufconn test harnesses (client_session_test.go, alarms_test.go, galaxy_test.go) were updated to use passthrough:///bufnet so the fake target reaches the context dialer. New tests TestDialFailsFastWhenGatewayUnreachable and TestDialReadinessProbeReachesReady cover the probe; go vet reports no deprecation. clients/go/README.md documents the lazy-connect + readiness-probe semantics.

Client.Go-006

Field	Value
Severity	Low
Category	Error handling & resilience
Location	clients/go/mxgateway/errors.go:9-130
Status	Resolved

Description: docs/ClientLibrariesDesign.md recommends a high-level error taxonomy (TransportError, AuthenticationError, TimeoutError, etc.). The Go client collapses all transport/gRPC failures into a single GatewayError with no way to classify transient (Unavailable, DeadlineExceeded) vs permanent (Unauthenticated, InvalidArgument) without manually unwrapping and calling status.Code.

Recommendation: Add a helper (e.g. IsTransient(err) bool) or expose the gRPC codes.Code on GatewayError, so retry/timeout/auth handling can be written without re-parsing the wrapped error.

Resolution: Resolved 2026-05-18: implemented the recommended classification surface in errors.go rather than a full parallel type hierarchy (the existing GatewayError/CommandError/MxAccessError chain already separates transport from protocol from MXAccess failures). Added GatewayError.Code() (returns the wrapped gRPC codes.Code, OK for nil, Unknown for a non-status error) and the free function IsTransient(err error) bool, which unwraps through *GatewayError and any gRPC-status chain and reports true for Unavailable, DeadlineExceeded, ResourceExhausted, and Aborted. Tests TestGatewayErrorCode and TestIsTransient cover the matrix; clients/go/README.md documents both for retry/timeout/auth handling.

Client.Go-007

Field	Value
Severity	Low
Category	Correctness & logic bugs
Location	clients/go/mxgateway/session.go:526-532
Status	Resolved

Description: newCorrelationID returns an empty string when crypto/rand.Read fails, silently producing an MxCommandRequest with no correlation id. rand.Read failure is rare, but the failure mode (untraceable command, no error surfaced) is worse than failing loud, and the empty-id path is untested.

Recommendation: Either propagate the error up through invokeCommand, or fall back to a time/counter-based id rather than an empty string.

Resolution: Resolved 2026-05-18: confirmed newCorrelationID returned "" on a rand.Read failure. It now falls back to a non-empty "fallback-<unixnano>-<counter>" id built from time.Now().UnixNano() and a process-wide atomic.Uint64 monotonic counter, so every command stays traceable even without entropy. The crypto/rand call was routed through a randRead package variable so the failure path is testable; TestNewCorrelationIDFallsBackOnRandFailure simulates a rand.Read failure and asserts the fallback id is non-empty, fallback- prefixed, and unique, and TestNewCorrelationIDUsesRandEntropy pins the happy path.

Client.Go-008

Field	Value
Severity	Low
Category	Testing coverage
Location	clients/go/mxgateway/ (test files)
Status	Resolved

Description: Several critical paths are untested: TLS credential resolution in resolveTransportCredentials (only the Plaintext path is exercised); the callContext deadline-shortening logic (client.go:198-204) including the negative-timeout disable case; and NativeValue/NativeArray for the array, raw-bytes, null, and unsupported-kind branches.

Recommendation: Add unit tests for resolveTransportCredentials precedence, callContext deadline arithmetic, and NativeValue/NativeArray round-trips for every kind.

Resolution: Resolved 2026-05-18: added clients/go/mxgateway/coverage_test.go. TestResolveTransportCredentialsPrecedence exercises every branch (explicit TransportCredentials, Plaintext, missing CACertFile error, TLSConfig + ServerNameOverride, default TLS floor) and TestResolveTransportCredentialsDoesNotMutateTLSConfig confirms the supplied *tls.Config is cloned. TestCallContextDeadlineArithmetic covers zero/default, negative-disable, positive timeout, caller-deadline-sooner-kept, and caller-deadline-later-shortened. TestNativeValueEdgeKinds, TestNativeArrayEdgeKinds, and TestNativeValueUnsupportedKind cover the null, raw-bytes (including the no-alias copy), array, timestamp-with-nil, and unsupported-kind branches.

Client.Go-009

Field	Value
Severity	Low
Category	Code organization & conventions
Location	clients/go/mxgateway/galaxy.go:60-93,241-256, clients/go/mxgateway/client.go:41-74,190-205
Status	Resolved

Description: DialGalaxy/Dial and GalaxyClient.callContext/Client.callContext are near-identical duplicates (dial-context setup, credential resolution, dial-option assembly, deadline arithmetic). A fix to one (e.g. the Client.Go-005 dial migration) must be applied twice and can drift.

Recommendation: Extract a shared unexported dial(ctx, opts) and a free callContext(opts, ctx) function, and have both client constructors call them.

Resolution: Resolved 2026-05-18: extracted the shared unexported dial(ctx, opts) (*grpc.ClientConn, error) (credential resolution, dial-option assembly, grpc.NewClient, readiness probe) and the free callContext(ctx, callTimeout) (context.Context, context.CancelFunc) into client.go. Dial/DialGalaxy and both (*Client).callContext/(*GalaxyClient).callContext methods now delegate to them; the duplicated dial and deadline code in galaxy.go was removed (its now-unused errors import dropped). This was done together with the Client.Go-005 migration so the grpc.NewClient change lives in exactly one place.

Client.Go-010

Field	Value
Severity	Low
Category	Documentation & comments
Location	clients/go/mxgateway/client.go:39-40
Status	Resolved

Description: The Dial doc comment states it configures "blocking dial cancellation from ctx." This describes the deprecated WithBlock behaviour; once Client.Go-005 is addressed the comment is misleading about how connection establishment and cancellation work.

Recommendation: Reword to describe the actual connect/timeout semantics after resolving Client.Go-005, and clarify that DialTimeout bounds the initial connect attempt.

Resolution: Resolved 2026-05-18: alongside the Client.Go-005 migration, the Dial doc comment was rewritten to describe the lazy grpc.NewClient connection, the DialTimeout-bounded (default 10s, or ctx deadline when sooner) readiness probe, that a briefly-unavailable gateway recovers instead of producing a hard error, and that cancelling ctx aborts the probe. DialGalaxy and the new dial/waitForReady/callContext helpers carry matching doc comments.

Client.Go-011

Field	Value
Severity	Low
Category	Correctness & logic bugs
Location	clients/go/mxgateway/alarms_test.go:66-73
Status	Resolved

Description: TestAcknowledgeAlarmRejectsNilRequest contains a no-op if with an empty body whose intent is documented in a comment ("Accept either: the helper returned the literal sentinel, or the generic transport error — both prove nil was rejected"). The condition

if err == nil || !errors.Is(err, errors.Unwrap(err)) && err.Error() != "mxgateway: acknowledge alarm request is required" {
    // ...
}

evaluates expressions for side effects only and asserts nothing — Go's && binds tighter than ||, the body is empty, and the actual nil check happens on the very next if err == nil. The block is effectively dead code masquerading as a check. It also evaluates errors.Unwrap(err) regardless of err's shape, and would call err.Error() even when err might be a wrapped status error whose message wording the gateway is free to change — making the apparent assertion brittle on top of being dead.

Recommendation: Drop the empty-body if entirely (the subsequent if err == nil { t.Fatalf(...) } already enforces the contract), or, if the intent is to additionally pin the literal error message for the sentinel path, replace it with a real assertion (if err.Error() != "mxgateway: acknowledge alarm request is required" { t.Fatalf(...) }) and remove the spurious errors.Is(err, errors.Unwrap(err)) clause.

Resolution: 2026-05-20 — Removed the empty-body if in TestAcknowledgeAlarmRejectsNilRequest; the subsequent if err == nil { t.Fatalf(...) } already enforces the nil-rejection contract without the dead, brittle compound predicate.

Client.Go-012

Field	Value
Severity	Low
Category	Documentation & comments
Location	clients/go/cmd/mxgw-go/main.go:1063-1065, clients/go/cmd/mxgw-go/main.go:88-104
Status	Resolved

Description: writeUsage lists the available subcommands as version|open-session|close-session|register|add-item|advise|subscribe-bulk|unsubscribe-bulk|write|stream-events|smoke|galaxy-test-connection|galaxy-last-deploy|galaxy-discover|galaxy-watch. Six subcommands wired into run are missing from this list: read-bulk, write-bulk, write2-bulk, write-secured-bulk, write-secured2-bulk, and bench-read-bulk. A user invoking mxgw-go with no args or an unknown command (the two paths that print this banner) sees an incomplete CLI surface and may believe the bulk-write / read-bulk families are not implemented. The README does document them, but the inline usage banner is the first source of truth a CLI user consults.

Recommendation: Extend the usage string to include every command registered in the switch args[0] in run, or generate it from a single source-of-truth slice keyed on command name → handler so the two cannot drift again.

Resolution: 2026-05-20 — writeUsage now lists the previously missing read-bulk, write-bulk, write2-bulk, write-secured-bulk, write-secured2-bulk, and bench-read-bulk subcommands alongside the original surface, so the no-args / unknown-command banner reflects every command wired into run.

Client.Go-013

Field	Value
Severity	Low
Category	Concurrency & thread safety
Location	clients/go/cmd/mxgw-go/main.go:1246-1249, clients/go/cmd/mxgw-go/main.go:1257-1262
Status	Resolved

Description: In runGalaxyWatch, the signal-cancellation branch carefully drains the buffered events channel after cancelStream() so the WatchDeployEvents goroutine can exit (for range events { }). The limit-reached branch (if *limit > 0 && count >= *limit { cancelStream(); return nil }) skips that drain and returns immediately. After the function returns, defer client.Close() runs and tears down the gRPC connection; in the gap before the connection close propagates, the WatchDeployEvents goroutine may still be blocked on case events <- event: (the channel is buffered to 16 but a slow producer can refill it) — the goroutine then exits via <-ctx.Done() because streamCtx was cancelled, so it isn't a permanent leak, but the two cancellation paths behave inconsistently and the limit-reached path can briefly hold a goroutine plus the gRPC stream while the client tears down underneath it.

Recommendation: Factor the drain into a helper and use it from both branches, e.g. after cancelStream() always for range events { } (and let the surrounding select/for re-evaluate <-errs if a terminal error was already buffered). Alternatively, drop the explicit drain in both branches and rely on defer cancelStream() plus defer client.Close() — but pick one model and apply it consistently.

Resolution: 2026-05-20 — The limit-reached branch in runGalaxyWatch now drains the buffered events channel (for range events { }) after cancelStream(), matching the signal-cancel branch. Both cancellation paths now wait for the WatchDeployEvents goroutine to exit before defer client.Close() tears the gRPC connection down.

Client.Go-014

Field	Value
Severity	Low
Category	Error handling & resilience
Location	clients/go/mxgateway/session.go:602, clients/go/mxgateway/galaxy.go:189
Status	Resolved

Description: Two stream Recv loops compare end-of-stream with err == io.EOF directly:

• session.go:602 — if err == io.EOF || status.Code(err) == codes.Canceled || streamCtx.Err() != nil { return }
• galaxy.go:189 — if recvErr == io.EOF { return }

gRPC's generated Recv() does return the io.EOF sentinel directly today, so the comparisons work in practice. However, the Go idiom (and the project's docs/style-guides/GoStyleGuide.md) is to use errors.Is(err, io.EOF) so future wrapping (e.g. an interceptor decorating Recv errors) does not silently flip the loop from "stream finished normally" to "stream produced an error". The mxgateway client itself wraps non-EOF Recv errors in *GatewayError, which errors.Is already supports — using errors.Is keeps both paths consistent.

Recommendation: Replace recvErr == io.EOF / err == io.EOF with errors.Is(err, io.EOF) (the errors package is already imported in both files).

Resolution: 2026-05-20 — Both stream Recv loops now use errors.Is(err, io.EOF): session.go already imported errors, and galaxy.go gained the missing errors import alongside the recvErr == io.EOF → errors.Is(recvErr, io.EOF) change, keeping EOF detection robust against any future Recv-error wrapping.

Client.Go-015

Field	Value
Severity	Low
Category	Code organization & conventions
Location	clients/go/cmd/mxgw-go/main.go:410-512
Status	Resolved

Description: runWriteBulkVariant(ctx, args, stdout, stderr, command, withTimestamp, secured bool) accepts secured but never uses it — the routing is keyed on command (the string "write-bulk" / "write2-bulk" / "write-secured-bulk" / "write-secured2-bulk"). The function ends with _ = secured // currently only used for routing above; reserved for future per-variant validation, which is misleading because secured is not in fact used for routing. The four wrapper functions (runWriteBulk, runWrite2Bulk, runWriteSecuredBulk, runWriteSecured2Bulk) all pass a secured argument that has no effect. The four CLI options -current-user-id, -verifier-user-id are unconditionally registered on every variant, including the non-secured ones, so a write-bulk invocation that passes -current-user-id 42 silently does nothing. Either remove secured and the dead _ = secured comment, or use it to gate the registration of secured-only flags so wrong combinations are rejected with a clean error.

Recommendation: Drop the secured parameter (the command switch already distinguishes the four variants) and the misleading _ = secured line; or, if validation is the goal, branch flag registration on secured so secured-only flags are unavailable for the non-secured variants and emit a clean usage error if they appear.

Resolution: 2026-05-20 — Dropped the unused secured parameter from runWriteBulkVariant (the command switch already distinguishes the four variants) and removed the misleading _ = secured line. The variant is now derived locally from command and used to gate flag registration: -current-user-id / -verifier-user-id are only registered for the secured variants and -user-id only for Write/Write2, so a wrong-variant flag now fails with a clean flag provided but not defined usage error instead of silently no-op'ing. The four runWrite*Bulk wrappers were updated to match the new signature.

Client.Go-016

Field	Value
Severity	Low
Category	Testing coverage
Location	clients/go/mxgateway/galaxy_test.go:382-429
Status	Resolved

Description: fakeGalaxyServer.watchSendInterval is declared on the test fake and consulted inside WatchDeployEvents (if s.watchSendInterval > 0 { ... }) but no test in the package sets a non-zero value. The dead field plus its branch were presumably added to support a backpressure / pacing test that was never landed, and now the only effect is reader confusion ("which test uses this?") and a pointlessly larger fake. Backpressure on the bootstrap-plus-events sequence is also genuinely worth testing, given that WatchDeployEvents writes to a 16-deep buffered channel.

Recommendation: Either delete the unused watchSendInterval field and its branch in WatchDeployEvents, or add the test it was added for — e.g. one that pumps more than 16 events with a small interval and asserts the consumer keeps up without losing or reordering events. Linking the field to a // for TestX comment if it stays would also help.

Resolution: 2026-05-20 — Removed the unused watchSendInterval field from fakeGalaxyServer and the corresponding if s.watchSendInterval > 0 { ... } branch in WatchDeployEvents; no test set the field, so the dead code path is gone and the fake is leaner. gofmt -w reflowed the struct to drop the no-longer-needed field-name padding.