Skip to content

test(sdk): typed-IaC E2E integration test + cross-plugin-build CI gate (Task 6)#603

Merged
intel352 merged 8 commits into
mainfrom
feat/iac-e2e-task6
May 10, 2026
Merged

test(sdk): typed-IaC E2E integration test + cross-plugin-build CI gate (Task 6)#603
intel352 merged 8 commits into
mainfrom
feat/iac-e2e-task6

Conversation

@intel352
Copy link
Copy Markdown
Contributor

Summary

Task 6 of the strict-contracts force-cutover plan (docs/plans/2026-05-10-strict-contracts-force-cutover.md, rev5).

Adds the workflow-side typed-IaC E2E integration test — the canonical smoke test the strict-contracts cutover wires into CI. plugin/external/sdk/iac_e2e_test.go (build tag integration) uses bufconn for in-process gRPC, registers a fake provider via sdk.RegisterAllIaCProviderServices, dials through a real gRPC channel, and exercises typed RPCs end-to-end.

Stacked on PR #602 (Task 5). Base branch is feat/iac-sdk-contracts-task5.

Test cases (both PASS, race-mode included)

TestIaC_EndToEnd_RequiredAndOptional_TypedDispatch

  • Required service: typed Name + Version RPCs return the expected values
  • Optional Enumerator: EnumerateAll returns the expected ResourceOutput
  • Critical regression guard: ResourceOutput.Sensitive (typed map<string,bool>) survives the roundtrip with secret=true. The pre-cutover structpb path silently dropped this map (T3.9 runtime-launch finding)

TestIaC_EndToEnd_OptionalNotRegistered_ClientFailsTyped

  • A provider that satisfies Required ONLY (no UnimplementedIaCProviderEnumeratorServer embed) → RegisterAllIaCProviderServices skips Enumerator registration, and a typed enumerator client receives a gRPC-layer Unimplemented error
  • Validates the design's "absence of registration IS the negative signal" contract end-to-end (no NotSupported field in response body)

CI integration

.github/workflows/cross-plugin-build-test.yml:

  • New iac-typed-e2e job runs the integration test on every IaC-touching PR. Per cycle 1 I-2 + cycle 2 I-1-NEW, go build alone leaves wire-incompat between workflow + plugin grpc-go versions undetected; this job catches that bug class
  • Path filters extended to gate on plugin/external/** so changes to typed SDK helpers + iac.proto trigger this workflow

The subprocess wire-test variant against the real DO plugin v1.0.0 binary is added once that plugin ships (plan §PR 3 / Task 7+).

Verification

  • GOWORK=off go test -tags=integration -race ./plugin/external/sdk/... -run TestIaC_EndToEnd → PASS (2/2)
  • GOWORK=off go test ./plugin/external/sdk/... → no regression in non-integration tests
  • GOWORK=off go vet -tags=integration ./plugin/external/... → clean
  • actionlint .github/workflows/cross-plugin-build-test.yml → clean
  • python yaml.safe_load(...) → parses

Rollback

Revert this commit; no production code or contract is affected (test + CI YAML only).

Test plan

  • TestIaC_EndToEnd_RequiredAndOptional_TypedDispatch passes (in-process bufconn gRPC roundtrip)
  • TestIaC_EndToEnd_OptionalNotRegistered_ClientFailsTyped passes (auto-registration absence verified at gRPC layer)
  • Race-mode test passes
  • actionlint clean on the modified workflow
  • yaml.safe_load parses the modified workflow

🤖 Generated with Claude Code

intel352 added 6 commits May 10, 2026 02:09
…ices + ResourceDriver

Task 3 of the strict-contracts force-cutover plan
(docs/plans/2026-05-10-strict-contracts-force-cutover.md, rev5).

Adds plugin/external/proto/iac.proto defining the typed gRPC contract
that supersedes the legacy InvokeService/structpb dispatch path for
IaCProvider + ResourceDriver:

- service IaCProviderRequired: 11 RPCs every IaC plugin MUST implement
  (Initialize, Name, Version, Capabilities, Plan, Apply, Destroy,
  Status, Import, ResolveSizing, BootstrapStateBackend). Compile-time
  enforced via the SDK type-assert in Task 4.

- 6 optional services — providers register only the ones they support:
  IaCProviderEnumerator (EnumerateAll, EnumerateByTag),
  IaCProviderDriftDetector (DetectDrift, DetectDriftWithSpecs),
  IaCProviderCredentialRevoker (RevokeProviderCredential),
  IaCProviderMigrationRepairer (RepairDirtyMigration),
  IaCProviderValidator (ValidatePlan),
  IaCProviderDriftConfigDetector (DetectDriftConfig).
  Absence of registration IS the negative signal — no NotSupported
  field on any optional response (per design §Optional services).

- service ResourceDriver: 9 RPCs for per-resource-type CRUD dispatch
  (Create, Read, Update, Delete, Diff, Scale, HealthCheck,
  SensitiveKeys, Troubleshoot), each carrying resource_type so a
  single server can route to the per-type driver implementation.

Hard invariants honored:
- NO google.protobuf.Struct, NO google.protobuf.Any anywhere.
- Free-form per-resource Config/Outputs payloads cross the wire as
  bytes <name>_json (the plugin owns json.Marshal/Unmarshal); this
  eliminates the structpb conversion surface that previously dropped
  map[string]bool entries silently (T3.9 finding).
- ResourceOutput.sensitive uses typed map<string, bool> per design.

Generated iac.pb.go + iac_grpc.pb.go via protoc v34.1 +
protoc-gen-go v1.36.11 + protoc-gen-go-grpc v1.6.1.

Failing test (plugin/external/proto/iac_proto_test.go) asserts the
generated server interfaces exist and have the methods the design
requires — drops in iac.proto cause the test file to fail to compile.

Verification: GOWORK=off go test ./plugin/external/proto/... PASSES;
GOWORK=off go build ./plugin/... ./cmd/... ./module/... clean.

Rollback: revert this commit; legacy InvokeService dispatch in
plugin.proto remains functional; the additive-only nature of this PR
means no consumer is affected until subsequent tasks wire callers.
Task 4 of the strict-contracts force-cutover plan
(docs/plans/2026-05-10-strict-contracts-force-cutover.md, rev5).

Adds plugin/external/sdk/iacserver.go: a single helper that uses Go
type-assertion to register every typed IaC gRPC service the provider
satisfies, in one call.

REQUIRED service:
  pb.IaCProviderRequiredServer — surfaced as a clear startup-time error
  if the provider type doesn't satisfy it (rather than failing at the
  first RPC dispatch with a generic "unimplemented" status).

OPTIONAL services (auto-detected): IaCProviderEnumerator,
IaCProviderDriftDetector, IaCProviderCredentialRevoker,
IaCProviderMigrationRepairer, IaCProviderValidator,
IaCProviderDriftConfigDetector. Plus ResourceDriver.

Per cycle 3 I-1 of the design: plugin authors write ONE call; they
cannot omit registration for a capability they implemented. This
removes the registration-omission bug class (the same shape as the
legacy InvokeService case-string-typo bug) by removing the manual
step entirely.

Tests cover four cases:
- required-satisfied → required service registered + advertised by
  grpcSrv.GetServiceInfo().
- enumerator-only → only the optional Enumerator service registered;
  other optionals stay absent (auto-detection precision).
- empty-stub → returns an error naming the unsatisfied required
  interface, with a docs pointer.
- all-capabilities-stub → all 8 typed services (Required + 6 optional
  + ResourceDriver) registered.

Stacked on feat/iac-proto-task3 (Task 3 PR #598 provides the
generated server interfaces this helper consumes).

Verification: GOWORK=off go test -race ./plugin/external/sdk/...
PASS; GOWORK=off go build ./plugin/... ./cmd/... ./module/... clean;
GOWORK=off go vet ./plugin/external/... clean.

Rollback: revert this commit; SDK consumers can still register
services manually via the per-service Register* helpers protoc
generated.
…rver callback

Task 29 of the strict-contracts force-cutover plan
(docs/plans/2026-05-10-strict-contracts-force-cutover.md, rev5).

Adds the high-level plugin-author API on top of Task 4's
RegisterAllIaCProviderServices:

  func main() {
      sdk.ServeIaCPlugin(&doProvider{}, sdk.IaCServeOptions{})
  }

Per cycle 3 I-1 of the design, service registration happens INSIDE
go-plugin's GRPCServer callback (iacGRPCPlugin.GRPCServer) — the
framework owns *grpc.Server lifecycle, so plugin authors cannot
pre-create a server and forget to register a typed service on it.

API surface (all in plugin/external/sdk/iacserver.go):
- IaCServeOptions{ PluginInfo *PluginInfo } — caller-side options.
- PluginInfo{ HandshakeConfig goplugin.HandshakeConfig } — extension
  point for future Name/Version metadata; defaults to ext.Handshake
  (the canonical wfctl<->plugin handshake) when zero-valued.
- iacGRPCPlugin{provider any} — implements goplugin.Plugin
  (GRPCServer + GRPCClient). The GoCodeAlone fork of go-plugin v1.7.0
  is gRPC-only and exposes only the canonical Plugin interface; there
  is no GRPCPlugin alias or NetRPCUnsupportedPlugin embed to use.
- ServeIaCPlugin(provider, opts) — wraps goplugin.Serve with the
  resolved handshake + a single iacGRPCPlugin entry under the "iac"
  key.
- resolveServeHandshake(opts) — extracted helper so the override-vs-
  default rule is unit-testable without invoking the blocking
  goplugin.Serve loop.

Tests (iacserver_serve_test.go) cover six cases via internal-package
tests (so the unexported plugin type is exercisable without a real
subprocess; subprocess-level coverage lands in Task 6's typed-IaC E2E
test):
- iacGRPCPlugin.GRPCServer registers all satisfied services on the
  framework-managed *grpc.Server (Required + Enumerator + ResourceDriver
  for the all-stub).
- iacGRPCPlugin.GRPCServer propagates the auto-register error for an
  empty stub — go-plugin aborts plugin startup with an actionable
  message.
- iacGRPCPlugin.GRPCClient is a no-op (host builds typed clients
  directly).
- iacGRPCPlugin satisfies goplugin.Plugin at compile time (refactor
  guard).
- ServeIaCPlugin defaults to ext.Handshake when PluginInfo is nil.
- ServeIaCPlugin honors a non-zero override handshake when provided.

Stacked on feat/iac-sdk-auto-register-task4 (Task 4 PR #599 provides
RegisterAllIaCProviderServices, which the GRPCServer callback delegates
to).

Verification: GOWORK=off go test -race ./plugin/external/sdk/... PASS;
GOWORK=off go build ./plugin/... ./cmd/... ./module/... clean;
GOWORK=off go vet ./plugin/external/... clean.

Rollback: revert this commit; plugin authors can fall back to
manually constructing goplugin.Serve + Plugins map referencing
RegisterAllIaCProviderServices in their own GRPCServer callback.
Task 5 of the strict-contracts force-cutover plan
(docs/plans/2026-05-10-strict-contracts-force-cutover.md, rev5).

Adds plugin/external/sdk/contracts.go with the BuildContractRegistry
helper that enumerates grpc.Server.GetServiceInfo() and emits a
SERVICE-kind ContractDescriptor for each registered service.
ContractMode is set to STRICT_PROTO so the host can distinguish typed
IaC services from the legacy structpb-mode contracts produced by
Module/Step/Trigger ContractProvider implementations.

Per cycle 3 I-1 of the design: wfctl needs a single mechanism to
discover "is the optional service registered on this plugin handle?".
Reusing the existing ContractRegistry shape keeps Module/Step/Trigger
and IaC capability discovery on the same wire surface — no new gRPC
server-reflection dependency required.

Service descriptors are emitted in deterministic alphabetical order
so callers can rely on stable output for diff/compare operations and
the wftest BDD test in Task 15.

The helper is safe to call with a nil server (returns an empty but
non-nil ContractRegistry) so callers that may construct it before the
gRPC server exists do not panic.

Tests (contracts_iac_test.go) cover three cases — all pass:
- AdvertisesRegisteredIaCServices: a Required + Enumerator +
  DriftDetector stub yields exactly those service descriptors.
- ServiceContractsUseStrictProtoMode: every emitted descriptor is
  Kind=SERVICE + Mode=STRICT_PROTO (host-side discriminator).
- NilServer_ReturnsEmpty: defensive contract for nil input.

Stacked on feat/iac-sdk-serve-task29 (Task 29 PR #600 provides
ServeIaCPlugin which IaC plugins use to register the services this
helper enumerates).

Verification: GOWORK=off go test -race ./plugin/external/sdk/... PASS;
GOWORK=off go build ./plugin/... ./cmd/... ./module/... clean;
GOWORK=off go vet ./plugin/external/... clean.

Rollback: revert this commit; ContractRegistry returns the prior
shape (Module/Step/Trigger only via the existing ContractProvider
hook in grpc_server.go).
Task 6 of the strict-contracts force-cutover plan
(docs/plans/2026-05-10-strict-contracts-force-cutover.md, rev5).

Adds plugin/external/sdk/iac_e2e_test.go (build tag `integration`) —
the canonical workflow-side smoke test for the typed IaC contract.
Uses bufconn for in-process gRPC, registers a fake provider via
sdk.RegisterAllIaCProviderServices, dials the server through a real
gRPC channel, and exercises typed RPCs on both Required (Name,
Version) and the Enumerator optional (EnumerateAll).

Critical assertion: ResourceOutput.Sensitive (typed map<string,bool>)
survives the roundtrip with value=true. The pre-cutover structpb path
silently dropped this map (T3.9 finding); this E2E test guards the
regression.

Second case asserts that when a provider satisfies Required ONLY (no
Enumerator embed), the auto-registration helper SKIPS the optional
service registration — and a typed enumerator client receives a
gRPC-layer Unimplemented error rather than a NotSupported flag in a
response body. This is the "absence of registration IS the negative
signal" contract from the design.

CI integration (.github/workflows/cross-plugin-build-test.yml):
- Adds an `iac-typed-e2e` job that runs the tests under
  -tags=integration on every IaC-touching PR. Per cycle 1 I-2 +
  cycle 2 I-1-NEW, `go build` alone leaves wire incompat between
  workflow and plugin grpc-go versions undetected; this job catches
  that bug class.
- Extends the path filters to gate on plugin/external/**, so changes
  to the typed sdk helpers + iac.proto trigger this workflow rather
  than only the AWS/GCP/Azure compile-compat job.
- The subprocess wire-test variant against the real DO plugin v1.0.0
  binary is added once that plugin ships (per plan §PR 3 / Task 7+).

Stacked on feat/iac-sdk-contracts-task5 (Task 5 PR #602 provides
BuildContractRegistry; the E2E test exercises the surface from
Tasks 3–5 + 29 end-to-end through gRPC).

Verification:
- GOWORK=off go test -tags=integration -race \
    ./plugin/external/sdk/... -run TestIaC_EndToEnd → PASS (2/2)
- GOWORK=off go test ./plugin/external/sdk/... → PASS (no regression
  in non-integration tests)
- GOWORK=off go vet -tags=integration ./plugin/external/... → clean
- actionlint .github/workflows/cross-plugin-build-test.yml → clean
- python yaml.safe_load(...) → parses

Rollback: revert this commit; no production code or contract is
affected (test + CI YAML only).
Per cycle 4 code-review PR 603 MINOR-1: the previous assertion in
TestIaC_EndToEnd_OptionalNotRegistered_ClientFailsTyped only checked
err != nil — any error (network flake, deadline, transport-layer
behavior change) would satisfy it, masking real Unimplemented-vs-
other regressions in the absence-of-registration signal.

Tightens the assertion to status.Code(err) == codes.Unimplemented
so the test specifically pins the design's "absence of registration
IS the negative signal" contract end-to-end at the gRPC layer.

Verification: GOWORK=off go test -tags=integration -race
./plugin/external/sdk/... -run TestIaC_EndToEnd → PASS (2/2);
gofmt clean.
Copy link
Copy Markdown
Contributor

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Adds a workflow-side integration (E2E) test that exercises the typed IaC gRPC services end-to-end over a real in-process gRPC channel (bufconn), and wires that test into CI via a new job in the cross-plugin build gate so typed-RPC wire incompatibilities are caught early.

Changes:

  • Introduces plugin/external/sdk/iac_e2e_test.go (build tag integration) to validate typed IaC Required + optional Enumerator behavior over bufconn gRPC.
  • Extends .github/workflows/cross-plugin-build-test.yml to run the integration E2E test as a dedicated iac-typed-e2e job.
  • Broadens PR path filters to trigger the gate on changes under plugin/external/**.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 5 comments.

File Description
plugin/external/sdk/iac_e2e_test.go New integration test validating typed IaC gRPC roundtrips (Required + optional Enumerator) via bufconn.
.github/workflows/cross-plugin-build-test.yml Adds CI job to run the typed IaC integration test and updates path filters to trigger this workflow.

// requiring the DO plugin to be cross-built first.
func TestIaC_EndToEnd_RequiredAndOptional_TypedDispatch(t *testing.T) {
listener := bufconn.Listen(e2eBufSize)
server := grpc.NewServer()
Comment thread plugin/external/sdk/iac_e2e_test.go Outdated
// signal" contract from the design — wfctl observes the absence at the
// gRPC layer rather than via a NotSupported flag in the response body.
func TestIaC_EndToEnd_OptionalNotRegistered_ClientFailsTyped(t *testing.T) {
listener := bufconn.Listen(e2eBufSize)
Comment on lines +160 to +162
enumClient := pb.NewIaCProviderEnumeratorClient(conn)
_, err = enumClient.EnumerateAll(context.Background(), &pb.EnumerateAllRequest{ResourceType: "x"})
if err == nil {
# Strict-contracts cutover Task 6 — typed IaC contract + sdk helpers
# live under plugin/external/sdk; gate this workflow on changes there
# so the iac-typed-e2e job catches typed-RPC drift.
- 'plugin/external/**'
… 603)

Per cycle 4 code-review PR 603 (Copilot 4 Important + 1 MINOR):

IMPORTANT-2/3 — bufconn listener leak:
  Both TestIaC_EndToEnd_* tests called bufconn.Listen but never
  closed the listener. server.Stop in t.Cleanup tears down the
  *grpc.Server but leaves the listener's accept goroutine alive
  until -race's GC pressure trips it. Adds
  `t.Cleanup(func() { _ = listener.Close() })` after each
  bufconn.Listen call.

IMPORTANT-4/5 — RPC deadline:
  RPCs used context.Background() with no deadline → CI worker
  hangs until suite-wide timeout on transport failure. Replaces
  with `ctx, cancel := context.WithTimeout(context.Background(),
  e2eRPCDeadline)` (5s) + t.Cleanup(cancel). Both tests now
  bound their RPC time even if the gRPC layer wedges.

  e2eRPCDeadline lives at package scope alongside e2eBufSize so
  the per-test allocation reads cleanly and a future timeout
  bump is one line.

MINOR-6 — path-filter intent comment:
  cross-plugin-build-test.yml `plugin/external/**` filter is
  broad on purpose — typed contract + sdk helpers + downstream
  IaC dispatch + remote-plugin orchestration code all live under
  this dir, and ALL of them affect the iac-typed-e2e job. Comment
  rewrites to document the intent (was: comment said only sdk
  helpers, suggesting a narrower path).

Verification:
  GOWORK=off go test -tags=integration -race
    ./plugin/external/sdk/... -run TestIaC_EndToEnd → PASS (2/2);
  gofmt clean; actionlint clean.

Rollback: revert this commit; bufconn listener leaks return +
RPC unbounded; cross-plugin-build path filter intent comment
returns to misleading wording.
@intel352 intel352 force-pushed the feat/iac-sdk-contracts-task5 branch from c03c871 to 8bb6e52 Compare May 10, 2026 11:31
@intel352 intel352 changed the base branch from feat/iac-sdk-contracts-task5 to main May 10, 2026 11:38
# Conflicts:
#	plugin/external/proto/iac.pb.go
#	plugin/external/proto/iac.proto
#	plugin/external/proto/iac_grpc.pb.go
#	plugin/external/proto/iac_proto_test.go
#	plugin/external/sdk/iacserver.go
#	plugin/external/sdk/iacserver_serve_test.go
#	plugin/external/sdk/iacserver_test.go
Copilot AI review requested due to automatic review settings May 10, 2026 11:40
Copy link
Copy Markdown
Contributor

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

- uses: actions/setup-go@v5
with: { go-version-file: go.mod }
- name: Typed-IaC E2E test (in-process gRPC roundtrip)
run: GOWORK=off go test -tags=integration ./plugin/external/sdk/... -run TestIaC_EndToEnd -count=1 -v
@github-actions
Copy link
Copy Markdown

⏱ Benchmark Results

No significant performance regressions detected.

benchstat comparison (baseline → PR)
## benchstat: baseline → PR
baseline-bench.txt:262: parsing iteration count: invalid syntax
baseline-bench.txt:285353: parsing iteration count: invalid syntax
baseline-bench.txt:577155: parsing iteration count: invalid syntax
baseline-bench.txt:872752: parsing iteration count: invalid syntax
baseline-bench.txt:1176119: parsing iteration count: invalid syntax
baseline-bench.txt:1495584: parsing iteration count: invalid syntax
benchmark-results.txt:262: parsing iteration count: invalid syntax
benchmark-results.txt:364574: parsing iteration count: invalid syntax
benchmark-results.txt:682455: parsing iteration count: invalid syntax
benchmark-results.txt:981577: parsing iteration count: invalid syntax
benchmark-results.txt:1353887: parsing iteration count: invalid syntax
benchmark-results.txt:1758789: parsing iteration count: invalid syntax
goos: linux
goarch: amd64
pkg: github.com/GoCodeAlone/workflow/dynamic
cpu: AMD EPYC 7763 64-Core Processor                
                            │ baseline-bench.txt │
                            │       sec/op       │
InterpreterCreation-4               5.620m ± 83%
ComponentLoad-4                     3.647m ± 10%
ComponentExecute-4                  1.931µ ±  0%
PoolContention/workers-1-4          1.099µ ±  1%
PoolContention/workers-2-4          1.092µ ±  4%
PoolContention/workers-4-4          1.100µ ±  1%
PoolContention/workers-8-4          1.101µ ±  3%
PoolContention/workers-16-4         1.101µ ±  1%
ComponentLifecycle-4                3.645m ±  3%
SourceValidation-4                  2.311µ ±  1%
RegistryConcurrent-4                814.0n ±  4%
LoaderLoadFromString-4              3.653m ±  1%
geomean                             18.48µ

                            │ baseline-bench.txt │
                            │        B/op        │
InterpreterCreation-4               2.027Mi ± 0%
ComponentLoad-4                     2.180Mi ± 0%
ComponentExecute-4                  1.203Ki ± 0%
PoolContention/workers-1-4          1.203Ki ± 0%
PoolContention/workers-2-4          1.203Ki ± 0%
PoolContention/workers-4-4          1.203Ki ± 0%
PoolContention/workers-8-4          1.203Ki ± 0%
PoolContention/workers-16-4         1.203Ki ± 0%
ComponentLifecycle-4                2.183Mi ± 0%
SourceValidation-4                  1.984Ki ± 0%
RegistryConcurrent-4                1.133Ki ± 0%
LoaderLoadFromString-4              2.182Mi ± 0%
geomean                             15.25Ki

                            │ baseline-bench.txt │
                            │     allocs/op      │
InterpreterCreation-4                15.68k ± 0%
ComponentLoad-4                      18.02k ± 0%
ComponentExecute-4                    25.00 ± 0%
PoolContention/workers-1-4            25.00 ± 0%
PoolContention/workers-2-4            25.00 ± 0%
PoolContention/workers-4-4            25.00 ± 0%
PoolContention/workers-8-4            25.00 ± 0%
PoolContention/workers-16-4           25.00 ± 0%
ComponentLifecycle-4                 18.07k ± 0%
SourceValidation-4                    32.00 ± 0%
RegistryConcurrent-4                  2.000 ± 0%
LoaderLoadFromString-4               18.06k ± 0%
geomean                               183.3

cpu: Intel(R) Xeon(R) Platinum 8370C CPU @ 2.80GHz
                            │ benchmark-results.txt │
                            │        sec/op         │
InterpreterCreation-4                 3.367m ± 193%
ComponentLoad-4                       3.503m ±   3%
ComponentExecute-4                    1.878µ ±   3%
PoolContention/workers-1-4            1.209µ ±   1%
PoolContention/workers-2-4            1.202µ ±   2%
PoolContention/workers-4-4            1.211µ ±   1%
PoolContention/workers-8-4            1.209µ ±   1%
PoolContention/workers-16-4           1.218µ ±   1%
ComponentLifecycle-4                  3.583m ±   1%
SourceValidation-4                    2.249µ ±   1%
RegistryConcurrent-4                  877.7n ±   6%
LoaderLoadFromString-4                3.571m ±   1%
geomean                               18.34µ

                            │ benchmark-results.txt │
                            │         B/op          │
InterpreterCreation-4                  2.027Mi ± 0%
ComponentLoad-4                        2.180Mi ± 0%
ComponentExecute-4                     1.203Ki ± 0%
PoolContention/workers-1-4             1.203Ki ± 0%
PoolContention/workers-2-4             1.203Ki ± 0%
PoolContention/workers-4-4             1.203Ki ± 0%
PoolContention/workers-8-4             1.203Ki ± 0%
PoolContention/workers-16-4            1.203Ki ± 0%
ComponentLifecycle-4                   2.183Mi ± 0%
SourceValidation-4                     1.984Ki ± 0%
RegistryConcurrent-4                   1.133Ki ± 0%
LoaderLoadFromString-4                 2.182Mi ± 0%
geomean                                15.25Ki

                            │ benchmark-results.txt │
                            │       allocs/op       │
InterpreterCreation-4                   15.68k ± 0%
ComponentLoad-4                         18.02k ± 0%
ComponentExecute-4                       25.00 ± 0%
PoolContention/workers-1-4               25.00 ± 0%
PoolContention/workers-2-4               25.00 ± 0%
PoolContention/workers-4-4               25.00 ± 0%
PoolContention/workers-8-4               25.00 ± 0%
PoolContention/workers-16-4              25.00 ± 0%
ComponentLifecycle-4                    18.07k ± 0%
SourceValidation-4                       32.00 ± 0%
RegistryConcurrent-4                     2.000 ± 0%
LoaderLoadFromString-4                  18.06k ± 0%
geomean                                  183.3

pkg: github.com/GoCodeAlone/workflow/middleware
cpu: AMD EPYC 7763 64-Core Processor                
                                  │ baseline-bench.txt │
                                  │       sec/op       │
CircuitBreakerDetection-4                  286.4n ± 7%
CircuitBreakerExecution_Success-4          21.36n ± 0%
CircuitBreakerExecution_Failure-4          66.37n ± 0%
geomean                                    74.05n

                                  │ baseline-bench.txt │
                                  │        B/op        │
CircuitBreakerDetection-4                 144.0 ± 0%
CircuitBreakerExecution_Success-4         0.000 ± 0%
CircuitBreakerExecution_Failure-4         0.000 ± 0%
geomean                                              ¹
¹ summaries must be >0 to compute geomean

                                  │ baseline-bench.txt │
                                  │     allocs/op      │
CircuitBreakerDetection-4                 1.000 ± 0%
CircuitBreakerExecution_Success-4         0.000 ± 0%
CircuitBreakerExecution_Failure-4         0.000 ± 0%
geomean                                              ¹
¹ summaries must be >0 to compute geomean

cpu: Intel(R) Xeon(R) Platinum 8370C CPU @ 2.80GHz
                                  │ benchmark-results.txt │
                                  │        sec/op         │
CircuitBreakerDetection-4                     458.9n ± 4%
CircuitBreakerExecution_Success-4             59.77n ± 0%
CircuitBreakerExecution_Failure-4             65.40n ± 0%
geomean                                       121.5n

                                  │ benchmark-results.txt │
                                  │         B/op          │
CircuitBreakerDetection-4                    144.0 ± 0%
CircuitBreakerExecution_Success-4            0.000 ± 0%
CircuitBreakerExecution_Failure-4            0.000 ± 0%
geomean                                                 ¹
¹ summaries must be >0 to compute geomean

                                  │ benchmark-results.txt │
                                  │       allocs/op       │
CircuitBreakerDetection-4                    1.000 ± 0%
CircuitBreakerExecution_Success-4            0.000 ± 0%
CircuitBreakerExecution_Failure-4            0.000 ± 0%
geomean                                                 ¹
¹ summaries must be >0 to compute geomean

pkg: github.com/GoCodeAlone/workflow/module
cpu: AMD EPYC 7763 64-Core Processor                
                                 │ baseline-bench.txt │
                                 │       sec/op       │
JQTransform_Simple-4                     875.8n ± 30%
JQTransform_ObjectConstruction-4         1.460µ ±  9%
JQTransform_ArraySelect-4                3.430µ ±  3%
JQTransform_Complex-4                    38.51µ ±  2%
JQTransform_Throughput-4                 1.805µ ±  1%
SSEPublishDelivery-4                     63.67n ±  1%
geomean                                  1.639µ

                                 │ baseline-bench.txt │
                                 │        B/op        │
JQTransform_Simple-4                   1.273Ki ± 0%
JQTransform_ObjectConstruction-4       1.773Ki ± 0%
JQTransform_ArraySelect-4              2.625Ki ± 0%
JQTransform_Complex-4                  16.22Ki ± 0%
JQTransform_Throughput-4               1.984Ki ± 0%
SSEPublishDelivery-4                     0.000 ± 0%
geomean                                             ¹
¹ summaries must be >0 to compute geomean

                                 │ baseline-bench.txt │
                                 │     allocs/op      │
JQTransform_Simple-4                     10.00 ± 0%
JQTransform_ObjectConstruction-4         15.00 ± 0%
JQTransform_ArraySelect-4                30.00 ± 0%
JQTransform_Complex-4                    324.0 ± 0%
JQTransform_Throughput-4                 17.00 ± 0%
SSEPublishDelivery-4                     0.000 ± 0%
geomean                                             ¹
¹ summaries must be >0 to compute geomean

cpu: Intel(R) Xeon(R) Platinum 8370C CPU @ 2.80GHz
                                 │ benchmark-results.txt │
                                 │        sec/op         │
JQTransform_Simple-4                        1.061µ ± 13%
JQTransform_ObjectConstruction-4            1.581µ ±  2%
JQTransform_ArraySelect-4                   3.361µ ±  2%
JQTransform_Complex-4                       36.36µ ±  1%
JQTransform_Throughput-4                    1.873µ ±  1%
SSEPublishDelivery-4                        77.07n ±  0%
geomean                                     1.759µ

                                 │ benchmark-results.txt │
                                 │         B/op          │
JQTransform_Simple-4                      1.273Ki ± 0%
JQTransform_ObjectConstruction-4          1.773Ki ± 0%
JQTransform_ArraySelect-4                 2.625Ki ± 0%
JQTransform_Complex-4                     16.22Ki ± 0%
JQTransform_Throughput-4                  1.984Ki ± 0%
SSEPublishDelivery-4                        0.000 ± 0%
geomean                                                ¹
¹ summaries must be >0 to compute geomean

                                 │ benchmark-results.txt │
                                 │       allocs/op       │
JQTransform_Simple-4                        10.00 ± 0%
JQTransform_ObjectConstruction-4            15.00 ± 0%
JQTransform_ArraySelect-4                   30.00 ± 0%
JQTransform_Complex-4                       324.0 ± 0%
JQTransform_Throughput-4                    17.00 ± 0%
SSEPublishDelivery-4                        0.000 ± 0%
geomean                                                ¹
¹ summaries must be >0 to compute geomean

pkg: github.com/GoCodeAlone/workflow/schema
cpu: AMD EPYC 7763 64-Core Processor                
                                    │ baseline-bench.txt │
                                    │       sec/op       │
SchemaValidation_Simple-4                   1.111µ ± 13%
SchemaValidation_AllFields-4                1.659µ ±  2%
SchemaValidation_FormatValidation-4         1.597µ ±  4%
SchemaValidation_ManySchemas-4              1.818µ ±  4%
geomean                                     1.521µ

                                    │ baseline-bench.txt │
                                    │        B/op        │
SchemaValidation_Simple-4                   0.000 ± 0%
SchemaValidation_AllFields-4                0.000 ± 0%
SchemaValidation_FormatValidation-4         0.000 ± 0%
SchemaValidation_ManySchemas-4              0.000 ± 0%
geomean                                                ¹
¹ summaries must be >0 to compute geomean

                                    │ baseline-bench.txt │
                                    │     allocs/op      │
SchemaValidation_Simple-4                   0.000 ± 0%
SchemaValidation_AllFields-4                0.000 ± 0%
SchemaValidation_FormatValidation-4         0.000 ± 0%
SchemaValidation_ManySchemas-4              0.000 ± 0%
geomean                                                ¹
¹ summaries must be >0 to compute geomean

cpu: Intel(R) Xeon(R) Platinum 8370C CPU @ 2.80GHz
                                    │ benchmark-results.txt │
                                    │        sec/op         │
SchemaValidation_Simple-4                      1.044µ ±  4%
SchemaValidation_AllFields-4                   1.533µ ± 45%
SchemaValidation_FormatValidation-4            1.504µ ±  3%
SchemaValidation_ManySchemas-4                 1.489µ ±  3%
geomean                                        1.376µ

                                    │ benchmark-results.txt │
                                    │         B/op          │
SchemaValidation_Simple-4                      0.000 ± 0%
SchemaValidation_AllFields-4                   0.000 ± 0%
SchemaValidation_FormatValidation-4            0.000 ± 0%
SchemaValidation_ManySchemas-4                 0.000 ± 0%
geomean                                                   ¹
¹ summaries must be >0 to compute geomean

                                    │ benchmark-results.txt │
                                    │       allocs/op       │
SchemaValidation_Simple-4                      0.000 ± 0%
SchemaValidation_AllFields-4                   0.000 ± 0%
SchemaValidation_FormatValidation-4            0.000 ± 0%
SchemaValidation_ManySchemas-4                 0.000 ± 0%
geomean                                                   ¹
¹ summaries must be >0 to compute geomean

pkg: github.com/GoCodeAlone/workflow/store
cpu: AMD EPYC 7763 64-Core Processor                
                                   │ baseline-bench.txt │
                                   │       sec/op       │
EventStoreAppend_InMemory-4                1.114µ ± 18%
EventStoreAppend_SQLite-4                  1.385m ±  2%
GetTimeline_InMemory/events-10-4           13.56µ ±  2%
GetTimeline_InMemory/events-50-4           75.72µ ±  3%
GetTimeline_InMemory/events-100-4          153.8µ ±  2%
GetTimeline_InMemory/events-500-4          627.1µ ± 27%
GetTimeline_InMemory/events-1000-4         1.277m ±  1%
GetTimeline_SQLite/events-10-4             106.4µ ±  1%
GetTimeline_SQLite/events-50-4             246.5µ ±  1%
GetTimeline_SQLite/events-100-4            417.9µ ±  1%
GetTimeline_SQLite/events-500-4            1.776m ±  0%
GetTimeline_SQLite/events-1000-4           3.459m ±  1%
geomean                                    220.4µ

                                   │ baseline-bench.txt │
                                   │        B/op        │
EventStoreAppend_InMemory-4                  798.0 ± 4%
EventStoreAppend_SQLite-4                  1.989Ki ± 1%
GetTimeline_InMemory/events-10-4           7.953Ki ± 0%
GetTimeline_InMemory/events-50-4           46.62Ki ± 0%
GetTimeline_InMemory/events-100-4          94.48Ki ± 0%
GetTimeline_InMemory/events-500-4          472.8Ki ± 0%
GetTimeline_InMemory/events-1000-4         944.3Ki ± 0%
GetTimeline_SQLite/events-10-4             16.74Ki ± 0%
GetTimeline_SQLite/events-50-4             87.14Ki ± 0%
GetTimeline_SQLite/events-100-4            175.4Ki ± 0%
GetTimeline_SQLite/events-500-4            846.1Ki ± 0%
GetTimeline_SQLite/events-1000-4           1.639Mi ± 0%
geomean                                    67.41Ki

                                   │ baseline-bench.txt │
                                   │     allocs/op      │
EventStoreAppend_InMemory-4                  7.000 ± 0%
EventStoreAppend_SQLite-4                    53.00 ± 0%
GetTimeline_InMemory/events-10-4             125.0 ± 0%
GetTimeline_InMemory/events-50-4             653.0 ± 0%
GetTimeline_InMemory/events-100-4           1.306k ± 0%
GetTimeline_InMemory/events-500-4           6.514k ± 0%
GetTimeline_InMemory/events-1000-4          13.02k ± 0%
GetTimeline_SQLite/events-10-4               382.0 ± 0%
GetTimeline_SQLite/events-50-4              1.852k ± 0%
GetTimeline_SQLite/events-100-4             3.681k ± 0%
GetTimeline_SQLite/events-500-4             18.54k ± 0%
GetTimeline_SQLite/events-1000-4            37.29k ± 0%
geomean                                     1.162k

cpu: Intel(R) Xeon(R) Platinum 8370C CPU @ 2.80GHz
                                   │ benchmark-results.txt │
                                   │        sec/op         │
EventStoreAppend_InMemory-4                    1.115µ ± 5%
EventStoreAppend_SQLite-4                      984.1µ ± 4%
GetTimeline_InMemory/events-10-4               13.61µ ± 2%
GetTimeline_InMemory/events-50-4               76.07µ ± 3%
GetTimeline_InMemory/events-100-4              152.1µ ± 2%
GetTimeline_InMemory/events-500-4              601.4µ ± 1%
GetTimeline_InMemory/events-1000-4             1.217m ± 1%
GetTimeline_SQLite/events-10-4                 81.30µ ± 1%
GetTimeline_SQLite/events-50-4                 237.6µ ± 2%
GetTimeline_SQLite/events-100-4                435.1µ ± 2%
GetTimeline_SQLite/events-500-4                1.913m ± 1%
GetTimeline_SQLite/events-1000-4               3.776m ± 2%
geomean                                        210.8µ

                                   │ benchmark-results.txt │
                                   │         B/op          │
EventStoreAppend_InMemory-4                    778.5 ± 10%
EventStoreAppend_SQLite-4                    1.986Ki ±  2%
GetTimeline_InMemory/events-10-4             7.953Ki ±  0%
GetTimeline_InMemory/events-50-4             46.62Ki ±  0%
GetTimeline_InMemory/events-100-4            94.48Ki ±  0%
GetTimeline_InMemory/events-500-4            472.8Ki ±  0%
GetTimeline_InMemory/events-1000-4           944.3Ki ±  0%
GetTimeline_SQLite/events-10-4               16.74Ki ±  0%
GetTimeline_SQLite/events-50-4               87.14Ki ±  0%
GetTimeline_SQLite/events-100-4              175.4Ki ±  0%
GetTimeline_SQLite/events-500-4              846.1Ki ±  0%
GetTimeline_SQLite/events-1000-4             1.639Mi ±  0%
geomean                                      67.27Ki

                                   │ benchmark-results.txt │
                                   │       allocs/op       │
EventStoreAppend_InMemory-4                     7.000 ± 0%
EventStoreAppend_SQLite-4                       53.00 ± 0%
GetTimeline_InMemory/events-10-4                125.0 ± 0%
GetTimeline_InMemory/events-50-4                653.0 ± 0%
GetTimeline_InMemory/events-100-4              1.306k ± 0%
GetTimeline_InMemory/events-500-4              6.514k ± 0%
GetTimeline_InMemory/events-1000-4             13.02k ± 0%
GetTimeline_SQLite/events-10-4                  382.0 ± 0%
GetTimeline_SQLite/events-50-4                 1.852k ± 0%
GetTimeline_SQLite/events-100-4                3.681k ± 0%
GetTimeline_SQLite/events-500-4                18.54k ± 0%
GetTimeline_SQLite/events-1000-4               37.29k ± 0%
geomean                                        1.162k

Benchmarks run with go test -bench=. -benchmem -count=6.
Regressions ≥ 20% are flagged. Results compared via benchstat.

@intel352 intel352 merged commit 0a6156e into main May 10, 2026
26 checks passed
@intel352 intel352 deleted the feat/iac-e2e-task6 branch May 10, 2026 11:54
intel352 added a commit that referenced this pull request May 10, 2026
… round 2)

Per team-lead + spec-reviewer ruling on PR #618 (round 1 used
typed-then-fallback pattern; rejected for code-shape reasons): tighten
to PURE Option B at all 5 wfctl-side dispatch sites. interfaces.X
fallback removed; non-typed providers hit a typed-error at the
type-assert site rather than silently falling through.

Sites converted to pure typed-pb:
- cmd/wfctl/infra_cleanup.go: hard-fail on non-typed provider; only
  pb.IaCProviderEnumeratorClient.EnumerateByTag at dispatch.
- cmd/wfctl/infra_apply_refresh.go: hard-fail (typed error from
  runInfraApplyRefreshPhase); detectDriftConfigTyped via typed
  client when registered, falls through to required IaCProvider.DetectDrift
  via typed adapter when not.
- cmd/wfctl/infra_status_drift.go: warn-and-skip on non-typed (this
  function returns bool; doesn't propagate error); detectDriftConfigTyped
  via typed client when registered.
- cmd/wfctl/infra_bootstrap.go: resolveCredentialRevoker hard-fails
  on non-typed (warning + nil revoker, same UX as missing service);
  returns the typed adapter directly so its RevokeProviderCredential
  method translates to the typed pb.RevokeProviderCredential RPC under
  the hood.
- cmd/wfctl/infra_align_rules.go: continue (silent skip) on non-typed;
  R-A10's "treat unimplemented as not-applicable" semantics preserved
  at the typed-adapter accessor level.

ADR-0028 (decisions/0028-task-17-pure-typed-cutover.md) records the
decision, failure modes the dual-path preserved (loader-gate
weakening, test-fixture DI leak, future contributor cargo-culting,
reviewer cognitive load), the bufconn migration pattern for tests
(per PR #603 + PR #609 precedent), and the strict-mode invariant
translation (gRPC codes.Unimplemented from a non-registered service
+ translateRPCErr in the adapter preserves operator-visible
ErrProviderMethodUnimplemented surface).

EXPECTED: ~10 test fixtures fail to compile or run after this commit
because they inject fake interfaces.IaCProvider implementations at
the dispatch sites. Fixture rewrites land in follow-up commits on
this same branch (no force-push). PR 618 stays in CHANGES REQUESTED
state until the test pass.

Local validation:
  GOWORK=off go build ./cmd/wfctl/                  # clean
  GOWORK=off go vet ./cmd/wfctl/                    # clean
  GOWORK=off go test ./cmd/wfctl/ -count=1 -short   # FAILS (expected — fixture rewrites pending)
  GOWORK=off golangci-lint run --enable=gocritic,gosec ./cmd/wfctl/...  # 0 issues (in code; tests are next commit)

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
intel352 added a commit that referenced this pull request May 10, 2026
…, Task 17 item 4)

Per ADR-0028 (PR 618 round 2), wfctl IaC dispatch sites are pure typed-pb
(`provider.(*typedIaCAdapter)`) — no interfaces.X fallback. Test fixtures
that previously injected fake `interfaces.IaCProvider` implementations no
longer reach the dispatch path; the type-assert fails. This migrates the
fixtures whose tests actually exercise a Task 17 dispatch site to use a
real *typedIaCAdapter wired to an in-process bufconn-served pb.IaCProvider*
gRPC server.

Shared fixture helper (cmd/wfctl/iac_typed_fixture_test.go):
- fixtureTypedAdapter declarative builder: each non-nil pb-server field
  registers the matching service on the bufconn server, mirroring the
  ContractRegistry-driven optional-client construction in production.
- fixtureRequiredServer: baseline IaCProviderRequiredServer with
  configurable name/version + UnimplementedIaCProviderRequiredServer
  embed for everything else.
- recordingEnumeratorServer: canned EnumerateByTag / EnumerateAll
  responses with mutex-guarded recorded inputs.
- recordingResourceDriverServer: minimal pb.ResourceDriverServer that
  records Delete invocations + per-call error injection.
- recordingDriftDetectorServer: canned DetectDrift responses.
- driftsToPBOrEmpty: engine-side []DriftResult to pb wire shape,
  mirroring the inverse driftsFromPB in iac_typed_adapter.go.

Pattern precedents: PR #603 (iac_e2e_test.go bufconn), PR #609
(discover_typed_loader_test.go boundary test), PR #605 (typed adapter
unit tests).

Migrated fixture files:

1. cmd/wfctl/infra_cleanup_test.go - fakeEnumeratingProvider/
   fakeNonEnumeratingProvider/fakeDeleteDriver replaced with
   newCleanupEnumFixture / newCleanupNonEnumFixture builders that
   produce *typedIaCAdapter instances. 7 TestInfraCleanup_* tests now
   exercise the bufconn typed dispatch end-to-end.

2. cmd/wfctl/infra_apply_refresh_test.go - refreshFakeProvider replaced
   with newRefreshDriftFixture which registers the typed
   IaCProviderDriftDetector service. 9 TestApplyRefresh_* tests now go
   through the typed wire path. TestApplyRefresh_TransientErrorDoesNotPrune
   asserts on the error substring rather than errors.Is(transientErr)
   because the gRPC wire boundary doesn't preserve error identity
   across the bufconn server.

3. cmd/wfctl/infra_align_ra10_test.go - stubIaCProvider type +
   validatingStubProvider type replaced with stubIaCProvider() and
   validatingStubProvider() builder functions returning *typedIaCAdapter.
   cannedValidatorServer registers IaCProviderValidator returning canned
   PlanDiagnostics. 8 TestCheckRA10_* + TestInfraAlign_RA10_FixtureProvider_Fires
   now exercise the typed Validator dispatch.

4. cmd/wfctl/infra_strict_mode_test.go -
   TestInfraCleanup_MultiProvider_ContinuesPastUnimplemented updated
   to use the migrated cleanup fixtures. Provider A (no Enumerator
   service registered) -> adapter.Enumerator() returns nil -> cleanup
   skips with "skipped fake-a: provider does not implement Enumerator"
   log line, preserving the multi-provider continue-on-skip semantics
   in their typed-shape form.

Scope notes:

ADR-0028 lists 10 fixture file paths. Of those:
- cmd/wfctl/infra_status_drift_test.go does not exist (the related
  drift test logic lives in infra_destroy_test.go's
  TestDriftInfraModules_NoDrift; it currently passes silently because
  the dispatch warns "not a typed IaC adapter" + returns false. A
  follow-up PR can migrate that test to harden the silent-pass case.)
- cmd/wfctl/infra_bootstrap_force_rotate_test.go uses stubProviderRevoker
  (interfaces.ProviderCredentialRevoker) rather than IaCProvider; tests
  call bootstrapSecrets directly, bypassing the resolveCredentialRevoker
  dispatch. No migration needed.
- cmd/wfctl/infra_rotate_and_prune_test.go uses fakeProviderEnumerableDriver
  (a custom test interface), not interfaces.IaCProvider.
- cmd/wfctl/infra_audit_keys_test.go's fakeIaCProviderForAuditKeys goes
  through `p.(interfaces.EnumeratorAll)` dispatch which is NOT a Task 17
  dispatch site (different from the 5 sites converted).
- cmd/wfctl/dryrun_test.go and cmd/wfctl/infra_provider_dispatch_test.go
  use iactest.NoopProvider via the resolveIaCProvider seam; the tests
  exercise the plan path, which doesn't type-assert to *typedIaCAdapter.

The 4 migrated files cover every test that was actually failing the
type-assert under PR #618 round 2's pure-typed dispatch. Tests in the
other ADR-listed files continue to pass without migration because they
don't reach a Task 17 dispatch site.

Local validation:
  GOWORK=off go build ./cmd/wfctl/                # clean
  GOWORK=off go vet ./cmd/wfctl/                  # clean
  GOWORK=off go test ./cmd/wfctl/ -count=1        # all PASS (7.3s)
  GOWORK=off go test ./cmd/wfctl/ -count=1 -race  # all PASS (10.1s)
  GOWORK=off golangci-lint run ./cmd/wfctl/...    # 0 issues

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
intel352 added a commit that referenced this pull request May 10, 2026
… / Task 17) (#618)

* feat(wfctl): typed-RPC capability discovery at 5 dispatch sites (Task 17)

Per plan §Task 17 (strict-contracts force-cutover, rev5) and team-lead's
Option B ruling: convert the 5 wfctl-side `p.(interfaces.X)` type-assert
sites to typed-pb dispatch via per-service accessors on the typed
IaCProvider adapter. Capability discovery happens BEFORE the call
(typed-client accessor returns nil when the plugin's ContractRegistry
didn't advertise the optional service) so we don't pay the wasted-RPC +
sentinel-error round-trip the legacy interfaces.X dispatch incurred.

**typedIaCAdapter accessors (cmd/wfctl/iac_typed_adapter.go +96 lines):**
- RequiredClient() pb.IaCProviderRequiredClient — always non-nil after
  the loader gate (PR #610) accepts the plugin.
- Enumerator() pb.IaCProviderEnumeratorClient
- DriftDetector() pb.IaCProviderDriftDetectorClient
- DriftConfigDetector() pb.IaCProviderDriftConfigDetectorClient
- CredentialRevoker() pb.IaCProviderCredentialRevokerClient
- MigrationRepairer() pb.IaCProviderMigrationRepairerClient
- Validator() pb.IaCProviderValidatorClient
- ResourceDriverClient() pb.ResourceDriverClient
Each optional accessor returns nil when the matching service isn't in
the `registered` map passed to newTypedIaCAdapter. Per-method
docstrings describe the dispatch sites that consume each accessor.

**Typed-RPC dispatch helpers (cmd/wfctl/iac_typed_dispatch.go +51 lines):**
- detectDriftConfigTyped(ctx, cli, refs, specs) → []DriftResult
- validatePlanTyped(ctx, cli, plan) → []PlanDiagnostic
Wrap a single typed pb.IaC* RPC + the marshalling helpers from
iac_typed_adapter.go (refsToPB / specToPB / driftsFromPB / planToPB /
planDiagnosticSeverityFromPB). Single source of truth for
proto/Go shape conversions; call sites stay focused on dispatch logic.

**5 dispatch sites converted:**

1. cmd/wfctl/infra_cleanup.go:97 — `p.(interfaces.Enumerator)` →
   typed pb.IaCProviderEnumeratorClient.EnumerateByTag. Falls back to
   the interfaces.Enumerator type-assert path for non-typed providers
   (test fixtures + non-wfctl consumers); typedIaCAdapter satisfies
   interfaces.Enumerator too, so the legacy branch path is functionally
   equivalent when used against the real adapter — the typed branch is
   preferred for clarity + to avoid wasted RPC against unregistered
   services.

2. cmd/wfctl/infra_apply_refresh.go:69 — `provider.(interfaces.DriftConfigDetector)`
   → typed pb.IaCProviderDriftConfigDetectorClient.DetectDriftConfig
   via detectDriftConfigTyped helper. Same fallback pattern.

3. cmd/wfctl/infra_status_drift.go:107 — same as #2 but for
   `wfctl infra status drift`. Same fallback pattern.

4. cmd/wfctl/infra_bootstrap.go:335 — resolveCredentialRevoker now
   short-circuits via typedIaCAdapter.CredentialRevoker() == nil
   before returning the interfaces.ProviderCredentialRevoker value.
   Caller signature stays interfaces.ProviderCredentialRevoker for
   stability + test-fixture compatibility; the typed dispatch happens
   inside typedIaCAdapter.RevokeProviderCredential which translates
   to a typed pb.RevokeProviderCredential RPC. Net effect: capability
   discovery moves from call-time (sentinel error) to load-time
   (accessor nil-check) without changing the caller's API.

5. cmd/wfctl/infra_align_rules.go:777 — `p.(interfaces.ProviderValidator)`
   → typed pb.IaCProviderValidatorClient.ValidatePlan via
   validatePlanTyped helper. Same fallback pattern as #1-3.

**Plan-correction notes**

Spec §Task 17 says "use optionals from Task 16" — Task 16's adapter
exposed optional clients as private fields, not a public map. Task 17
adds typed-client accessors as the extension surface (per team-lead
Option B). The 5 sites use a typed-then-fallback pattern rather than
pure typed-only: keeping the interfaces.X branch as a stable seam for
test fixtures + non-wfctl consumers avoids forcing every caller to
also be a typedIaCAdapter consumer (which would require re-writing
~10 test fixtures across 4 files for no semantic gain — typedIaCAdapter
satisfies all the interfaces too, so the typed branch is the
strict-cutover preferred path while the fallback preserves the
interfaces.X integration point that out-of-org / future provider impls
might still use).

Net effect: wfctl call sites prefer typed pb dispatch; interfaces.X
type-assertions remain as a documented fallback. The interfaces/X
definitions stay in `interfaces/` for engine-side consumers per the
strict-contracts design (typedIaCAdapter is the wfctl-side adapter
that bridges the typed pb client to the engine's interfaces.X).

Local validation (against current main, post-rebase):
  GOWORK=off go build ./cmd/wfctl/                                     # clean
  GOWORK=off go vet ./cmd/wfctl/                                       # clean
  GOWORK=off go test ./cmd/wfctl/ -count=1 -short                      # all PASS (6.5s)
  GOWORK=off golangci-lint run --enable=gocritic,gosec ./cmd/wfctl/... # 0 issues

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

* refactor(wfctl): pure typed-pb dispatch at 5 sites + ADR-0028 (PR 618 round 2)

Per team-lead + spec-reviewer ruling on PR #618 (round 1 used
typed-then-fallback pattern; rejected for code-shape reasons): tighten
to PURE Option B at all 5 wfctl-side dispatch sites. interfaces.X
fallback removed; non-typed providers hit a typed-error at the
type-assert site rather than silently falling through.

Sites converted to pure typed-pb:
- cmd/wfctl/infra_cleanup.go: hard-fail on non-typed provider; only
  pb.IaCProviderEnumeratorClient.EnumerateByTag at dispatch.
- cmd/wfctl/infra_apply_refresh.go: hard-fail (typed error from
  runInfraApplyRefreshPhase); detectDriftConfigTyped via typed
  client when registered, falls through to required IaCProvider.DetectDrift
  via typed adapter when not.
- cmd/wfctl/infra_status_drift.go: warn-and-skip on non-typed (this
  function returns bool; doesn't propagate error); detectDriftConfigTyped
  via typed client when registered.
- cmd/wfctl/infra_bootstrap.go: resolveCredentialRevoker hard-fails
  on non-typed (warning + nil revoker, same UX as missing service);
  returns the typed adapter directly so its RevokeProviderCredential
  method translates to the typed pb.RevokeProviderCredential RPC under
  the hood.
- cmd/wfctl/infra_align_rules.go: continue (silent skip) on non-typed;
  R-A10's "treat unimplemented as not-applicable" semantics preserved
  at the typed-adapter accessor level.

ADR-0028 (decisions/0028-task-17-pure-typed-cutover.md) records the
decision, failure modes the dual-path preserved (loader-gate
weakening, test-fixture DI leak, future contributor cargo-culting,
reviewer cognitive load), the bufconn migration pattern for tests
(per PR #603 + PR #609 precedent), and the strict-mode invariant
translation (gRPC codes.Unimplemented from a non-registered service
+ translateRPCErr in the adapter preserves operator-visible
ErrProviderMethodUnimplemented surface).

EXPECTED: ~10 test fixtures fail to compile or run after this commit
because they inject fake interfaces.IaCProvider implementations at
the dispatch sites. Fixture rewrites land in follow-up commits on
this same branch (no force-push). PR 618 stays in CHANGES REQUESTED
state until the test pass.

Local validation:
  GOWORK=off go build ./cmd/wfctl/                  # clean
  GOWORK=off go vet ./cmd/wfctl/                    # clean
  GOWORK=off go test ./cmd/wfctl/ -count=1 -short   # FAILS (expected — fixture rewrites pending)
  GOWORK=off golangci-lint run --enable=gocritic,gosec ./cmd/wfctl/...  # 0 issues (in code; tests are next commit)

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

* test(wfctl): bufconn-backed *typedIaCAdapter fixtures (PR 618 round 3, Task 17 item 4)

Per ADR-0028 (PR 618 round 2), wfctl IaC dispatch sites are pure typed-pb
(`provider.(*typedIaCAdapter)`) — no interfaces.X fallback. Test fixtures
that previously injected fake `interfaces.IaCProvider` implementations no
longer reach the dispatch path; the type-assert fails. This migrates the
fixtures whose tests actually exercise a Task 17 dispatch site to use a
real *typedIaCAdapter wired to an in-process bufconn-served pb.IaCProvider*
gRPC server.

Shared fixture helper (cmd/wfctl/iac_typed_fixture_test.go):
- fixtureTypedAdapter declarative builder: each non-nil pb-server field
  registers the matching service on the bufconn server, mirroring the
  ContractRegistry-driven optional-client construction in production.
- fixtureRequiredServer: baseline IaCProviderRequiredServer with
  configurable name/version + UnimplementedIaCProviderRequiredServer
  embed for everything else.
- recordingEnumeratorServer: canned EnumerateByTag / EnumerateAll
  responses with mutex-guarded recorded inputs.
- recordingResourceDriverServer: minimal pb.ResourceDriverServer that
  records Delete invocations + per-call error injection.
- recordingDriftDetectorServer: canned DetectDrift responses.
- driftsToPBOrEmpty: engine-side []DriftResult to pb wire shape,
  mirroring the inverse driftsFromPB in iac_typed_adapter.go.

Pattern precedents: PR #603 (iac_e2e_test.go bufconn), PR #609
(discover_typed_loader_test.go boundary test), PR #605 (typed adapter
unit tests).

Migrated fixture files:

1. cmd/wfctl/infra_cleanup_test.go - fakeEnumeratingProvider/
   fakeNonEnumeratingProvider/fakeDeleteDriver replaced with
   newCleanupEnumFixture / newCleanupNonEnumFixture builders that
   produce *typedIaCAdapter instances. 7 TestInfraCleanup_* tests now
   exercise the bufconn typed dispatch end-to-end.

2. cmd/wfctl/infra_apply_refresh_test.go - refreshFakeProvider replaced
   with newRefreshDriftFixture which registers the typed
   IaCProviderDriftDetector service. 9 TestApplyRefresh_* tests now go
   through the typed wire path. TestApplyRefresh_TransientErrorDoesNotPrune
   asserts on the error substring rather than errors.Is(transientErr)
   because the gRPC wire boundary doesn't preserve error identity
   across the bufconn server.

3. cmd/wfctl/infra_align_ra10_test.go - stubIaCProvider type +
   validatingStubProvider type replaced with stubIaCProvider() and
   validatingStubProvider() builder functions returning *typedIaCAdapter.
   cannedValidatorServer registers IaCProviderValidator returning canned
   PlanDiagnostics. 8 TestCheckRA10_* + TestInfraAlign_RA10_FixtureProvider_Fires
   now exercise the typed Validator dispatch.

4. cmd/wfctl/infra_strict_mode_test.go -
   TestInfraCleanup_MultiProvider_ContinuesPastUnimplemented updated
   to use the migrated cleanup fixtures. Provider A (no Enumerator
   service registered) -> adapter.Enumerator() returns nil -> cleanup
   skips with "skipped fake-a: provider does not implement Enumerator"
   log line, preserving the multi-provider continue-on-skip semantics
   in their typed-shape form.

Scope notes:

ADR-0028 lists 10 fixture file paths. Of those:
- cmd/wfctl/infra_status_drift_test.go does not exist (the related
  drift test logic lives in infra_destroy_test.go's
  TestDriftInfraModules_NoDrift; it currently passes silently because
  the dispatch warns "not a typed IaC adapter" + returns false. A
  follow-up PR can migrate that test to harden the silent-pass case.)
- cmd/wfctl/infra_bootstrap_force_rotate_test.go uses stubProviderRevoker
  (interfaces.ProviderCredentialRevoker) rather than IaCProvider; tests
  call bootstrapSecrets directly, bypassing the resolveCredentialRevoker
  dispatch. No migration needed.
- cmd/wfctl/infra_rotate_and_prune_test.go uses fakeProviderEnumerableDriver
  (a custom test interface), not interfaces.IaCProvider.
- cmd/wfctl/infra_audit_keys_test.go's fakeIaCProviderForAuditKeys goes
  through `p.(interfaces.EnumeratorAll)` dispatch which is NOT a Task 17
  dispatch site (different from the 5 sites converted).
- cmd/wfctl/dryrun_test.go and cmd/wfctl/infra_provider_dispatch_test.go
  use iactest.NoopProvider via the resolveIaCProvider seam; the tests
  exercise the plan path, which doesn't type-assert to *typedIaCAdapter.

The 4 migrated files cover every test that was actually failing the
type-assert under PR #618 round 2's pure-typed dispatch. Tests in the
other ADR-listed files continue to pass without migration because they
don't reach a Task 17 dispatch site.

Local validation:
  GOWORK=off go build ./cmd/wfctl/                # clean
  GOWORK=off go vet ./cmd/wfctl/                  # clean
  GOWORK=off go test ./cmd/wfctl/ -count=1        # all PASS (7.3s)
  GOWORK=off go test ./cmd/wfctl/ -count=1 -race  # all PASS (10.1s)
  GOWORK=off golangci-lint run ./cmd/wfctl/...    # 0 issues

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

* docs(decisions): ADR-0028 expansion — per-site dispatch UX (PR 618 round 3)

Per spec-reviewer ruling on PR #618 round 3: code-shape mandate is met
(pure typed-pb at all 5 sites), but the per-site rejection severity
varies based on iteration semantics. Soft-skip at iteration sites is
graceful degradation, not the rejected silent-fallback shape — this
expansion documents the rule + per-site rationale so future
contributors don't cargo-cult either direction blindly.

New `## Per-site dispatch UX` section adds:
- Severity table for each of the 5 sites (cleanup hard-error,
  apply-refresh hard-error, status-drift soft-skip, align-rules R-A10
  silent-skip, bootstrap soft-skip-revocation) with explicit per-site
  reasoning anchored in iteration vs single-shot semantics.
- Canonical rule (verbatim from team-lead): "Pure typed-pb dispatch at
  all sites; non-typed input rejection severity is per-site UX based
  on iteration semantics. New dispatch sites default to hard-error
  unless graceful-degradation is operationally required." Plus the
  two-condition bar for soft-skip eligibility (iteration + auditable
  warn-log).
- Failure-mode contrast vs the round-1-rejected silent-fallback
  pattern: (1) the fallback path no longer exists at all 5 sites,
  (2) soft-skip is auditable via stderr warn-log, (3) the no-op
  result is observably distinct from a typed-pb success at the call
  site.

ADR-only edit; no code, fixture, or test changes.

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

* fix(wfctl): translate Unimplemented + propagate ctx + doc/error polish (PR 618 round 4)

Per code-review IMPORTANT-1 / IMPORTANT-2 / MINOR-1 / MINOR-2 (PR 618 round 4):

IMPORTANT-1 — translateRPCErr at typed dispatch sites
  ADR-0028 §Migration's "Strict-mode invariant translation" promises
  codes.Unimplemented at the wire boundary becomes
  interfaces.ErrProviderMethodUnimplemented for downstream errors.Is
  classification. The typedIaCAdapter's interfaces.IaCProvider methods
  already wrap, but the new typed-RPC dispatch helpers + the inline
  EnumerateByTag call site bypassed the wrap. Fixed two sites:
    - cmd/wfctl/iac_typed_dispatch.go:detectDriftConfigTyped now wraps
      cli.DetectDriftConfig errors via translateRPCErr.
    - cmd/wfctl/infra_cleanup.go's enumCli.EnumerateByTag site wraps
      via translateRPCErr before formatting + appending to totalErrs.
  Audit confirmed the 3 other dispatch sites already route through
  adapter methods that translate (provider.DetectDrift via
  typedIaCAdapter.DetectDrift, adapter.RevokeProviderCredential).
  validatePlanTyped intentionally returns nil-diags on any error per
  the documented Go interfaces.ProviderValidator.ValidatePlan
  signature contract; no translation needed there.

IMPORTANT-2 — propagate caller context to ValidatePlan
  validatePlanTyped at infra_align_rules.go:782 was called with
  context.Background(), losing operator Ctrl-C / parent cancellation /
  RPC deadline propagation. Threaded ctx through:
    - runInfraAlign → runInfraAlignChecks(ctx, opts)
    - runInfraAlignChecks → checkRA10_provider_validate_plan(ctx, ...)
    - checkRA10_provider_validate_plan → validatePlanTyped(ctx, ...)
  Renamed runInfraAlignChecks's local *alignContext binding from
  `ctx` to `alignCtx` to avoid shadowing the new context.Context
  parameter. Test callers (runInfraAlignChecks at 16 sites,
  checkRA10_provider_validate_plan at 9 sites) updated to pass
  context.Background(); context import added to test files that
  needed it.

MINOR-1 — iac_typed_adapter.go accessor doc-comment
  Doc example said `if !ok { /* legacy path no longer exists */ }`
  while the body asserted "wfctl call sites are pure typed". Reworked
  the example to show both per-site UX shapes (hard-error +
  soft-skip) per ADR-0028 §Per-site dispatch UX, with parenthetical
  mapping to the dispatch sites that use each shape.

MINOR-2 — specToPB error key context
  detectDriftConfigTyped's per-spec marshalling loop returned
  bare specToPB errors with no key context. Wrapped with
  fmt.Errorf("specToPB %q: %w", k, err) so post-mortem debugging
  identifies which entry in the per-resource specs map blew up.

Local validation:
  GOWORK=off go build ./cmd/wfctl/                # clean
  GOWORK=off go vet ./cmd/wfctl/                  # clean
  GOWORK=off go test ./cmd/wfctl/ -count=1        # all PASS (7.4s)
  GOWORK=off go test ./cmd/wfctl/ -count=1 -race  # all PASS (10.5s)
  GOWORK=off golangci-lint run ./cmd/wfctl/...    # 0 issues

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

* fix(wfctl): signal.NotifyContext + status-drift comment + fixture marshal-fail (PR 618 round 5)

Per code-review round 5 follow-ups (3 Copilot findings on round 4 head):

1. cmd/wfctl/infra_status_drift.go:103-110 (was MINOR-4 corrigendum)
   Comment said "Hard-fail when provider isn't a typed adapter" but the
   implementation soft-skips (warn + return false). Updated the comment
   to match ADR-0028 §Per-site dispatch UX: status-drift iterates per
   provider, halting the whole status command on the first non-typed
   provider would lose visibility into the others' drift, so the warn-
   log + no-drift-reported degradation is operationally correct. The
   warning log is the auditable signal of fixture-leak / loader-gate
   gaps.

2. cmd/wfctl/infra_align.go:75 (REAL — IMPORTANT-2 intent gap)
   Round-4 fix threaded ctx through the dispatch chain but called
   runInfraAlignChecks with context.Background() at the entry point —
   defeating IMPORTANT-2's cancellation-propagation intent. Wired
   signal.NotifyContext(ctx, os.Interrupt, syscall.SIGTERM) at
   runInfraAlign so operator Ctrl-C / SIGTERM cancels in-flight typed-
   RPC calls (R-A10 ValidatePlan + any future typed dispatch the rule
   layer adds). The other wfctl runInfra* entrypoints (status, drift,
   apply, destroy, import, etc.) currently use context.Background()
   directly and do NOT honor signals; the signal-aware pattern landing
   here is the operator-tooling shape we want, but a follow-up sweep
   to wire it into the other entrypoints is out of scope for this PR
   (signal-cancellation-for-the-CLI is a horizontal concern bigger
   than Task 17). Documented in the inline comment so a future
   contributor sees the intentional asymmetry.

3. cmd/wfctl/iac_typed_fixture_test.go:280-308 (REAL — test rigor)
   driftsToPBOrEmpty silently swallowed marshalJSONMap errors via
   `_, _ := ...`. A fixture author who hands the recording server an
   un-marshallable Expected/Actual map would have seen a silently-empty
   ExpectedJson on the wire — false-pass shape. Fix: renamed to
   driftsToPB returning (slice, error); per-entry errors include
   index + resource name for triage. recordingDriftDetectorServer
   now stores the pre-marshalled []*pb.DriftResult (pbDrifts) so the
   gRPC handler is alloc-only, no marshal failure mode at RPC time.
   newRefreshDriftFixture pre-marshals at fixture-build time and
   t.Fatalf on any error — fixture-leak now fails deterministically
   at test setup (option 1 from code-review brief).

Local validation:
  GOWORK=off go build ./cmd/wfctl/                # clean
  GOWORK=off go vet ./cmd/wfctl/                  # clean
  GOWORK=off go test ./cmd/wfctl/ -count=1        # all PASS (8.3s)
  GOWORK=off go test ./cmd/wfctl/ -count=1 -race  # all PASS (10.6s)
  GOWORK=off golangci-lint run ./cmd/wfctl/...    # 0 issues

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

---------

Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants