Skip to content

feat(wfctl): typed-IaC cutover — replace remoteIaCProvider with pb.IaCProviderRequiredClient (PR 4 / Task 16)#609

Merged
intel352 merged 3 commits into
mainfrom
feat/iac-wfctl-typed-cutover-task16
May 10, 2026
Merged

feat(wfctl): typed-IaC cutover — replace remoteIaCProvider with pb.IaCProviderRequiredClient (PR 4 / Task 16)#609
intel352 merged 3 commits into
mainfrom
feat/iac-wfctl-typed-cutover-task16

Conversation

@intel352
Copy link
Copy Markdown
Contributor

@intel352 intel352 commented May 10, 2026

Summary

  • discoverAndLoadIaCProvider rewritten to construct typedIaCAdapter (Task 30, PR feat(wfctl): typed-IaC adapter satisfying interfaces.IaCProvider (PR 4 / Task 30) #605) directly from the plugin's gRPC ClientConn + ContractRegistry registered-service map. The legacy InvokeService string-dispatch surface is REMOVED entirely.
  • plugin/external/grpc_plugin.go + plugin/external/adapter.go retain *grpc.ClientConn on PluginClient and add Conn() accessors on both — architectural prerequisite (commit ad7d946) bundled in the same PR per Path A direction.
  • DELETED remoteIaCProvider struct + every interface method (~620 lines), remoteResourceDriver struct + every interface method (~300 lines), remoteServiceInvoker interfaces, loadIaCPlugin var + defaultLoadIaCPlugin, readIaCPluginComputePlanVersion, plus orphaned helpers (jsonToAny, anyToStruct, sensitiveToAny, decodeResourceOutput, isPluginMethodUnimplemented, stringVal, stringFromMap).
  • DELETED 5 dependent test files (~2767 lines) per plan §Task 16 enumeration:
    • deploy_providers_remote_iac_test.go (1090)
    • deploy_providers_remote_driver_test.go (820)
    • deploy_providers_dispatch_matrix_test.go (477)
    • deploy_providers_strict_bridge_coverage_test.go (302)
    • deploy_providers_remote_iac_compat_test.go (65)
    • deploy_remote_provider_test.go (173 — additional remote* coverage subsumed by typed paths)
  • KEPT wrapIaCError + containsAny + retryOnTransient + deployRetryDelays + deployOpError + closerFunc + findIaCPluginDir + iacPluginManifest — provider-agnostic helpers + plugin discovery still used by the typed cutover.
  • ADDED registeredIaCServices(reg *pb.ContractRegistry) map[string]bool — walks the plugin's ContractRegistry response, collects SERVICE-kind contract names so typedIaCAdapter can gate optional-client construction.
  • UPDATED cmd/wfctl/deploy_providers_test.go — removed TestResolveIaCProviderSurfacesPluginError, pruned now-unused imports.

Net diff: 1 source file -957 / +69, 5 test files deleted -2927, 1 test file updated -31 / +10, plus 2 files in plugin/external/ +44 / 0 from the prerequisite. ≈3.9k line reduction in cmd/wfctl/ + ~44 line addition in plugin/external/.

Why

PR 4 of docs/plans/2026-05-10-strict-contracts-force-cutover.md (Task 16). Closes the bug class the legacy remoteIaCProvider introduced — every IaCProvider call now travels through a typed gRPC RPC instead of InvokeService("IaCProvider.Method", map[string]any{...}) string-dispatch + structpb round-trip. ADR-0026 honored: NOT a hand-written marshalling proxy.

Plugins that don't register the typed workflow.plugin.external.iac.IaCProviderRequired service are rejected at load time with a wfctl plugin update hint — the surface is deleted, not deprecated.

Required follow-ups before merge (CHANGES REQUESTED — spec-reviewer)

  1. Wire PR feat(wfctl): IaC plugin pre-flight gate building block + iac-typed-cutover runbook (Task 18) #610's AssertIaCPluginAdvertisesRequiredService in place of the inline if !registered[iacServiceRequired] check. Returned error already wraps errLegacyIaCPlugin; callers can errors.Is(err, errLegacyIaCPlugin) for typed dispatch. Drop the duplicate iacServiceRequired const (coordinate naming with implementer-3 — DM in flight).
  2. Add TestDiscoverAndLoadIaCProvider_ReturnsTypedClient asserting the boundary returns *typedIaCAdapter. In-process gRPC fake plugin + concrete-type assertion.
  3. Documented belowComputePlanVersionDeclarer regression Plan-correction note added (option (d) batched with canonical_keys per team-lead ruling).

Both code fixes require PR #610 to merge first (predicate lives in iac_loader_gate.go). Sequence: PR 598 → 599 → 600 → 602 → 603 → 605 → 606 → 610 → THIS PR (rebased). Plan to apply all three in one fix-up commit post-rebase.

Plan-correction notes

Per team-lead Path A ruling: the plugin/external/ Conn() prerequisite (commit ad7d946) is included in the SAME PR as the cutover. Spec gap; same fix-forward precedent as Tasks 1 (ADR-0027) and 2 (PR #597 plan-correction notes block).

Two engine-side capability behaviors don't have a corresponding RPC in PR #598's typed proto surface. Per team-lead + spec-reviewer ruling, both close via the SAME follow-up additive PR (option (d) — capability discovery field on CapabilitiesResponse.IaCCapabilityDeclaration, NOT new RPCs, NOT touching PR #598). Sequencing: between Task 17 (#19) and Task 20 (#23) per the manifest. ONE PR, ONE set of plugin updates, ONE cascade-amendment.

1. SupportedCanonicalKeys() regression.

  • typedIaCAdapter (PR feat(wfctl): typed-IaC adapter satisfying interfaces.IaCProvider (PR 4 / Task 30) #605) returns interfaces.CanonicalKeys() (full set).
  • Verification: workflow-plugin-digitalocean/internal/provider.go:DOProvider.SupportedCanonicalKeys filters via doUnsupportedCanonicalKeys to a strict subset. Azure suspected; AWS/GCP not in our checkout.
  • Follow-up adds repeated string canonical_keys to CapabilitiesResponse.IaCCapabilityDeclaration. typedIaCAdapter then reads from there. Current full-set default behavior preserved as the no-capability-data fallback after the follow-up ships.

2. ComputePlanVersionDeclarer.ComputePlanVersion() regression.

  • typedIaCAdapter does NOT implement ComputePlanVersion(); apply-path falls back to v1 dispatch silently.
  • Verification (mirroring the canonical-keys methodology): only workflow-plugin-digitalocean/plugin.json declares "computePlanVersion": "v2" — sole non-default override across workflow-plugin-{aws,gcp,azure,digitalocean}/plugin.json (verified grep "computePlanVersion" {plugin}/plugin.json empty for the first three).
  • Follow-up adds string compute_plan_version to CapabilitiesResponse.IaCCapabilityDeclaration (or per-plugin Capabilities; final shape TBD). typedIaCAdapter then implements ComputePlanVersion() reading from there.
  • Net regression window: gap between this PR landing and the follow-up + DO plugin v1.0.0 (Task 9 → Task 13). DO falls back to v1 (legacy provider.Apply) instead of v2 (wfctlhelpers.ApplyPlan + drift postcondition). Acceptable transient per team-lead.

Spec line-count estimate ~2x underestimate (plan §Task 16 estimated "~1500-2000 line aggregate diff"; actual ~3900). Documented for retro batch.

Dependency note ⚠️

This branch's CI will FAIL until PR #598 (Task 3 proto), PR #605 (Task 30 typed adapter), AND PR #610 (Task 18 loader gate) merge. The new discoverAndLoadIaCProvider references pb.ContractRegistry, pb.ContractKind_CONTRACT_KIND_SERVICE, newTypedIaCAdapter, iacServiceRequired, AND will reference AssertIaCPluginAdvertisesRequiredService post-fix. Local validation passes against an overlay of all three branches in the working tree (proto + sdk + adapter + gate overlaid for compile-check only — NOT committed in this PR per Path A direction):

$ GOWORK=off go build ./...                               # clean
$ GOWORK=off go vet ./cmd/wfctl/ ./plugin/external/...    # clean
$ GOWORK=off go test ./cmd/wfctl/ -count=1 -short         # all PASS (7.05s)

Test plan

Rollback

Revert both commits. The legacy remoteIaCProvider proxy reappears in cmd/wfctl/deploy_providers.go; the deleted test files restore. plugin/external/Conn() accessors remain (additive, useful, non-load-bearing for the legacy path); they can be left in place even on rollback to preserve future cutover capability.

🤖 Generated with Claude Code

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 small but necessary API surface to retain and expose the underlying *grpc.ClientConn for external plugins, enabling future typed gRPC client construction (e.g., for the upcoming wfctl typed-IaC cutover) without losing access to the original connection.

Changes:

  • Store the *grpc.ClientConn on external.PluginClient and expose it via Conn().
  • Expose the same connection from external.ExternalPluginAdapter via Conn() (delegating to the wrapped client).

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File Description
plugin/external/grpc_plugin.go Retains the gRPC connection on PluginClient and adds a Conn() accessor.
plugin/external/adapter.go Adds ExternalPluginAdapter.Conn() to surface the plugin connection to downstream callers.

Comment thread plugin/external/adapter.go Outdated
@github-actions
Copy link
Copy Markdown

github-actions Bot commented May 10, 2026

⏱ 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:352157: parsing iteration count: invalid syntax
baseline-bench.txt:630762: parsing iteration count: invalid syntax
baseline-bench.txt:896224: parsing iteration count: invalid syntax
baseline-bench.txt:1229858: parsing iteration count: invalid syntax
baseline-bench.txt:1546437: parsing iteration count: invalid syntax
benchmark-results.txt:262: parsing iteration count: invalid syntax
benchmark-results.txt:265382: parsing iteration count: invalid syntax
benchmark-results.txt:581928: parsing iteration count: invalid syntax
benchmark-results.txt:901189: parsing iteration count: invalid syntax
benchmark-results.txt:1190057: parsing iteration count: invalid syntax
benchmark-results.txt:1486551: 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 │        benchmark-results.txt        │
                            │       sec/op       │    sec/op      vs base              │
InterpreterCreation-4              3.959m ± 160%   4.028m ± 150%       ~ (p=0.589 n=6)
ComponentLoad-4                    3.577m ±   2%   3.578m ±  11%       ~ (p=0.699 n=6)
ComponentExecute-4                 1.934µ ±   2%   1.968µ ±   0%  +1.78% (p=0.017 n=6)
PoolContention/workers-1-4         1.098µ ±   3%   1.092µ ±   1%       ~ (p=0.260 n=6)
PoolContention/workers-2-4         1.095µ ±   2%   1.088µ ±   3%       ~ (p=0.853 n=6)
PoolContention/workers-4-4         1.097µ ±   1%   1.090µ ±   1%       ~ (p=0.074 n=6)
PoolContention/workers-8-4         1.107µ ±   1%   1.101µ ±   2%       ~ (p=0.840 n=6)
PoolContention/workers-16-4        1.108µ ±   2%   1.107µ ±   0%       ~ (p=1.000 n=6)
ComponentLifecycle-4               3.706m ±   1%   3.589m ±   0%  -3.15% (p=0.002 n=6)
SourceValidation-4                 2.425µ ±   2%   2.333µ ±   1%  -3.77% (p=0.002 n=6)
RegistryConcurrent-4               875.5n ±   4%   804.9n ±   2%  -8.06% (p=0.002 n=6)
LoaderLoadFromString-4             3.719m ±   0%   3.613m ±   1%  -2.84% (p=0.002 n=6)
geomean                            18.17µ          17.91µ         -1.44%

                            │ baseline-bench.txt │        benchmark-results.txt         │
                            │        B/op        │     B/op      vs base                │
InterpreterCreation-4               2.027Mi ± 0%   2.027Mi ± 0%       ~ (p=0.509 n=6)
ComponentLoad-4                     2.180Mi ± 0%   2.180Mi ± 0%       ~ (p=0.937 n=6)
ComponentExecute-4                  1.203Ki ± 0%   1.203Ki ± 0%       ~ (p=1.000 n=6) ¹
PoolContention/workers-1-4          1.203Ki ± 0%   1.203Ki ± 0%       ~ (p=1.000 n=6) ¹
PoolContention/workers-2-4          1.203Ki ± 0%   1.203Ki ± 0%       ~ (p=1.000 n=6) ¹
PoolContention/workers-4-4          1.203Ki ± 0%   1.203Ki ± 0%       ~ (p=1.000 n=6) ¹
PoolContention/workers-8-4          1.203Ki ± 0%   1.203Ki ± 0%       ~ (p=1.000 n=6) ¹
PoolContention/workers-16-4         1.203Ki ± 0%   1.203Ki ± 0%       ~ (p=1.000 n=6) ¹
ComponentLifecycle-4                2.183Mi ± 0%   2.183Mi ± 0%       ~ (p=0.697 n=6)
SourceValidation-4                  1.984Ki ± 0%   1.984Ki ± 0%       ~ (p=1.000 n=6) ¹
RegistryConcurrent-4                1.133Ki ± 0%   1.133Ki ± 0%       ~ (p=1.000 n=6) ¹
LoaderLoadFromString-4              2.182Mi ± 0%   2.182Mi ± 0%       ~ (p=0.974 n=6)
geomean                             15.25Ki        15.25Ki       -0.00%
¹ all samples are equal

                            │ baseline-bench.txt │        benchmark-results.txt        │
                            │     allocs/op      │  allocs/op   vs base                │
InterpreterCreation-4                15.68k ± 0%   15.68k ± 0%       ~ (p=1.000 n=6)
ComponentLoad-4                      18.02k ± 0%   18.02k ± 0%       ~ (p=1.000 n=6)
ComponentExecute-4                    25.00 ± 0%    25.00 ± 0%       ~ (p=1.000 n=6) ¹
PoolContention/workers-1-4            25.00 ± 0%    25.00 ± 0%       ~ (p=1.000 n=6) ¹
PoolContention/workers-2-4            25.00 ± 0%    25.00 ± 0%       ~ (p=1.000 n=6) ¹
PoolContention/workers-4-4            25.00 ± 0%    25.00 ± 0%       ~ (p=1.000 n=6) ¹
PoolContention/workers-8-4            25.00 ± 0%    25.00 ± 0%       ~ (p=1.000 n=6) ¹
PoolContention/workers-16-4           25.00 ± 0%    25.00 ± 0%       ~ (p=1.000 n=6) ¹
ComponentLifecycle-4                 18.07k ± 0%   18.07k ± 0%       ~ (p=1.000 n=6) ¹
SourceValidation-4                    32.00 ± 0%    32.00 ± 0%       ~ (p=1.000 n=6) ¹
RegistryConcurrent-4                  2.000 ± 0%    2.000 ± 0%       ~ (p=1.000 n=6) ¹
LoaderLoadFromString-4               18.06k ± 0%   18.06k ± 0%       ~ (p=1.000 n=6) ¹
geomean                               183.3         183.3       +0.00%
¹ all samples are equal

pkg: github.com/GoCodeAlone/workflow/middleware
                                  │ baseline-bench.txt │       benchmark-results.txt       │
                                  │       sec/op       │   sec/op     vs base              │
CircuitBreakerDetection-4                  284.0n ± 2%   286.5n ± 5%  +0.88% (p=0.039 n=6)
CircuitBreakerExecution_Success-4          21.50n ± 0%   21.54n ± 0%       ~ (p=0.260 n=6)
CircuitBreakerExecution_Failure-4          66.19n ± 0%   66.09n ± 0%       ~ (p=0.119 n=6)
geomean                                    73.93n        74.16n       +0.30%

                                  │ baseline-bench.txt │       benchmark-results.txt        │
                                  │        B/op        │    B/op     vs base                │
CircuitBreakerDetection-4                 144.0 ± 0%     144.0 ± 0%       ~ (p=1.000 n=6) ¹
CircuitBreakerExecution_Success-4         0.000 ± 0%     0.000 ± 0%       ~ (p=1.000 n=6) ¹
CircuitBreakerExecution_Failure-4         0.000 ± 0%     0.000 ± 0%       ~ (p=1.000 n=6) ¹
geomean                                              ²               +0.00%               ²
¹ all samples are equal
² summaries must be >0 to compute geomean

                                  │ baseline-bench.txt │       benchmark-results.txt        │
                                  │     allocs/op      │ allocs/op   vs base                │
CircuitBreakerDetection-4                 1.000 ± 0%     1.000 ± 0%       ~ (p=1.000 n=6) ¹
CircuitBreakerExecution_Success-4         0.000 ± 0%     0.000 ± 0%       ~ (p=1.000 n=6) ¹
CircuitBreakerExecution_Failure-4         0.000 ± 0%     0.000 ± 0%       ~ (p=1.000 n=6) ¹
geomean                                              ²               +0.00%               ²
¹ all samples are equal
² summaries must be >0 to compute geomean

pkg: github.com/GoCodeAlone/workflow/module
                                 │ baseline-bench.txt │       benchmark-results.txt        │
                                 │       sec/op       │    sec/op     vs base              │
JQTransform_Simple-4                     880.3n ± 27%   884.5n ± 28%       ~ (p=0.699 n=6)
JQTransform_ObjectConstruction-4         1.453µ ±  1%   1.474µ ± 29%  +1.41% (p=0.024 n=6)
JQTransform_ArraySelect-4                3.332µ ±  1%   3.385µ ±  1%  +1.58% (p=0.009 n=6)
JQTransform_Complex-4                    38.24µ ±  1%   38.54µ ±  1%       ~ (p=0.065 n=6)
JQTransform_Throughput-4                 1.777µ ±  0%   1.797µ ±  1%  +1.13% (p=0.002 n=6)
SSEPublishDelivery-4                     64.40n ±  1%   63.69n ±  1%  -1.11% (p=0.009 n=6)
geomean                                  1.628µ         1.640µ        +0.71%

                                 │ baseline-bench.txt │        benchmark-results.txt         │
                                 │        B/op        │     B/op      vs base                │
JQTransform_Simple-4                   1.273Ki ± 0%     1.273Ki ± 0%       ~ (p=1.000 n=6) ¹
JQTransform_ObjectConstruction-4       1.773Ki ± 0%     1.773Ki ± 0%       ~ (p=1.000 n=6) ¹
JQTransform_ArraySelect-4              2.625Ki ± 0%     2.625Ki ± 0%       ~ (p=1.000 n=6) ¹
JQTransform_Complex-4                  16.22Ki ± 0%     16.22Ki ± 0%       ~ (p=1.000 n=6) ¹
JQTransform_Throughput-4               1.984Ki ± 0%     1.984Ki ± 0%       ~ (p=1.000 n=6) ¹
SSEPublishDelivery-4                     0.000 ± 0%       0.000 ± 0%       ~ (p=1.000 n=6) ¹
geomean                                             ²                 +0.00%               ²
¹ all samples are equal
² summaries must be >0 to compute geomean

                                 │ baseline-bench.txt │       benchmark-results.txt        │
                                 │     allocs/op      │ allocs/op   vs base                │
JQTransform_Simple-4                     10.00 ± 0%     10.00 ± 0%       ~ (p=1.000 n=6) ¹
JQTransform_ObjectConstruction-4         15.00 ± 0%     15.00 ± 0%       ~ (p=1.000 n=6) ¹
JQTransform_ArraySelect-4                30.00 ± 0%     30.00 ± 0%       ~ (p=1.000 n=6) ¹
JQTransform_Complex-4                    324.0 ± 0%     324.0 ± 0%       ~ (p=1.000 n=6) ¹
JQTransform_Throughput-4                 17.00 ± 0%     17.00 ± 0%       ~ (p=1.000 n=6) ¹
SSEPublishDelivery-4                     0.000 ± 0%     0.000 ± 0%       ~ (p=1.000 n=6) ¹
geomean                                             ²               +0.00%               ²
¹ all samples are equal
² summaries must be >0 to compute geomean

pkg: github.com/GoCodeAlone/workflow/schema
                                    │ baseline-bench.txt │       benchmark-results.txt        │
                                    │       sec/op       │    sec/op     vs base              │
SchemaValidation_Simple-4                   1.119µ ± 13%   1.118µ ± 23%       ~ (p=0.818 n=6)
SchemaValidation_AllFields-4                1.666µ ±  2%   1.646µ ±  4%       ~ (p=0.818 n=6)
SchemaValidation_FormatValidation-4         1.590µ ±  1%   1.580µ ±  2%       ~ (p=0.327 n=6)
SchemaValidation_ManySchemas-4              1.774µ ±  4%   1.812µ ±  3%       ~ (p=0.255 n=6)
geomean                                     1.514µ         1.515µ        +0.06%

                                    │ baseline-bench.txt │       benchmark-results.txt        │
                                    │        B/op        │    B/op     vs base                │
SchemaValidation_Simple-4                   0.000 ± 0%     0.000 ± 0%       ~ (p=1.000 n=6) ¹
SchemaValidation_AllFields-4                0.000 ± 0%     0.000 ± 0%       ~ (p=1.000 n=6) ¹
SchemaValidation_FormatValidation-4         0.000 ± 0%     0.000 ± 0%       ~ (p=1.000 n=6) ¹
SchemaValidation_ManySchemas-4              0.000 ± 0%     0.000 ± 0%       ~ (p=1.000 n=6) ¹
geomean                                                ²               +0.00%               ²
¹ all samples are equal
² summaries must be >0 to compute geomean

                                    │ baseline-bench.txt │       benchmark-results.txt        │
                                    │     allocs/op      │ allocs/op   vs base                │
SchemaValidation_Simple-4                   0.000 ± 0%     0.000 ± 0%       ~ (p=1.000 n=6) ¹
SchemaValidation_AllFields-4                0.000 ± 0%     0.000 ± 0%       ~ (p=1.000 n=6) ¹
SchemaValidation_FormatValidation-4         0.000 ± 0%     0.000 ± 0%       ~ (p=1.000 n=6) ¹
SchemaValidation_ManySchemas-4              0.000 ± 0%     0.000 ± 0%       ~ (p=1.000 n=6) ¹
geomean                                                ²               +0.00%               ²
¹ all samples are equal
² summaries must be >0 to compute geomean

pkg: github.com/GoCodeAlone/workflow/store
                                   │ baseline-bench.txt │       benchmark-results.txt        │
                                   │       sec/op       │    sec/op     vs base              │
EventStoreAppend_InMemory-4                1.215µ ± 10%   1.205µ ± 18%       ~ (p=0.818 n=6)
EventStoreAppend_SQLite-4                  1.280m ±  4%   1.376m ±  8%  +7.47% (p=0.002 n=6)
GetTimeline_InMemory/events-10-4           13.79µ ±  2%   14.00µ ±  2%  +1.53% (p=0.041 n=6)
GetTimeline_InMemory/events-50-4           76.19µ ± 18%   75.99µ ± 20%       ~ (p=0.818 n=6)
GetTimeline_InMemory/events-100-4          125.1µ ±  1%   122.8µ ±  1%  -1.85% (p=0.002 n=6)
GetTimeline_InMemory/events-500-4          643.2µ ±  1%   630.6µ ±  0%  -1.96% (p=0.002 n=6)
GetTimeline_InMemory/events-1000-4         1.316m ±  1%   1.286m ±  1%  -2.25% (p=0.002 n=6)
GetTimeline_SQLite/events-10-4             106.3µ ±  1%   107.7µ ±  0%  +1.31% (p=0.004 n=6)
GetTimeline_SQLite/events-50-4             251.1µ ±  0%   248.6µ ±  0%  -0.99% (p=0.002 n=6)
GetTimeline_SQLite/events-100-4            427.1µ ±  1%   423.7µ ±  2%       ~ (p=0.589 n=6)
GetTimeline_SQLite/events-500-4            1.830m ±  0%   1.794m ±  1%  -1.96% (p=0.002 n=6)
GetTimeline_SQLite/events-1000-4           3.565m ±  0%   3.494m ±  0%  -1.99% (p=0.002 n=6)
geomean                                    220.0µ         219.5µ        -0.25%

                                   │ baseline-bench.txt │        benchmark-results.txt         │
                                   │        B/op        │     B/op      vs base                │
EventStoreAppend_InMemory-4                  802.0 ± 6%     777.5 ± 6%       ~ (p=0.310 n=6)
EventStoreAppend_SQLite-4                  1.987Ki ± 2%   1.987Ki ± 1%       ~ (p=1.000 n=6)
GetTimeline_InMemory/events-10-4           7.953Ki ± 0%   7.953Ki ± 0%       ~ (p=1.000 n=6) ¹
GetTimeline_InMemory/events-50-4           46.62Ki ± 0%   46.62Ki ± 0%       ~ (p=1.000 n=6) ¹
GetTimeline_InMemory/events-100-4          94.48Ki ± 0%   94.48Ki ± 0%       ~ (p=1.000 n=6) ¹
GetTimeline_InMemory/events-500-4          472.8Ki ± 0%   472.8Ki ± 0%       ~ (p=0.061 n=6)
GetTimeline_InMemory/events-1000-4         944.3Ki ± 0%   944.3Ki ± 0%  +0.00% (p=0.006 n=6)
GetTimeline_SQLite/events-10-4             16.74Ki ± 0%   16.74Ki ± 0%       ~ (p=1.000 n=6) ¹
GetTimeline_SQLite/events-50-4             87.14Ki ± 0%   87.14Ki ± 0%       ~ (p=1.000 n=6) ¹
GetTimeline_SQLite/events-100-4            175.4Ki ± 0%   175.4Ki ± 0%       ~ (p=1.000 n=6) ¹
GetTimeline_SQLite/events-500-4            846.1Ki ± 0%   846.1Ki ± 0%  -0.00% (p=0.013 n=6)
GetTimeline_SQLite/events-1000-4           1.639Mi ± 0%   1.639Mi ± 0%       ~ (p=0.132 n=6)
geomean                                    67.44Ki        67.26Ki       -0.26%
¹ all samples are equal

                                   │ baseline-bench.txt │        benchmark-results.txt        │
                                   │     allocs/op      │  allocs/op   vs base                │
EventStoreAppend_InMemory-4                  7.000 ± 0%    7.000 ± 0%       ~ (p=1.000 n=6) ¹
EventStoreAppend_SQLite-4                    53.00 ± 0%    53.00 ± 0%       ~ (p=1.000 n=6) ¹
GetTimeline_InMemory/events-10-4             125.0 ± 0%    125.0 ± 0%       ~ (p=1.000 n=6) ¹
GetTimeline_InMemory/events-50-4             653.0 ± 0%    653.0 ± 0%       ~ (p=1.000 n=6) ¹
GetTimeline_InMemory/events-100-4           1.306k ± 0%   1.306k ± 0%       ~ (p=1.000 n=6) ¹
GetTimeline_InMemory/events-500-4           6.514k ± 0%   6.514k ± 0%       ~ (p=1.000 n=6) ¹
GetTimeline_InMemory/events-1000-4          13.02k ± 0%   13.02k ± 0%       ~ (p=1.000 n=6) ¹
GetTimeline_SQLite/events-10-4               382.0 ± 0%    382.0 ± 0%       ~ (p=1.000 n=6) ¹
GetTimeline_SQLite/events-50-4              1.852k ± 0%   1.852k ± 0%       ~ (p=1.000 n=6) ¹
GetTimeline_SQLite/events-100-4             3.681k ± 0%   3.681k ± 0%       ~ (p=1.000 n=6) ¹
GetTimeline_SQLite/events-500-4             18.54k ± 0%   18.54k ± 0%       ~ (p=1.000 n=6) ¹
GetTimeline_SQLite/events-1000-4            37.29k ± 0%   37.29k ± 0%       ~ (p=1.000 n=6) ¹
geomean                                     1.162k        1.162k       +0.00%
¹ all samples are equal

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

@intel352 intel352 changed the title [WIP] feat: wfctl typed-IaC cutover (PR 4 / Task 16) — Phase 1: Conn() prerequisite feat(wfctl): typed-IaC cutover — replace remoteIaCProvider with pb.IaCProviderRequiredClient (PR 4 / Task 16) May 10, 2026
@intel352 intel352 marked this pull request as ready for review May 10, 2026 07:08
Copilot AI review requested due to automatic review settings May 10, 2026 07:08
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 10 out of 10 changed files in this pull request and generated 5 comments.

Comment thread cmd/wfctl/deploy_providers.go Outdated
Comment on lines +192 to 196
registered := registeredIaCServices(adapter.ContractRegistry())
if !registered[iacServiceRequired] {
_ = closer.Close()
return nil, nil, fmt.Errorf("plugin %q iac.provider factory returned nil (unexpected — file an issue)", pluginName)
return nil, nil, fmt.Errorf("plugin %q does not register the required %q gRPC service — upgrade with: wfctl plugin update %s", pName, iacServiceRequired, pName)
}
Comment on lines +147 to +161
// discoverAndLoadIaCProvider implements the default resolveIaCProvider: it
// scans the plugin directory for a plugin that declares
// iacProvider.name == providerName, loads it via ExternalPluginManager, and
// returns a typedIaCAdapter (satisfying interfaces.IaCProvider) plus a
// Closer that shuts down the plugin subprocess. The caller must call
// Close() on the returned Closer when done.
//
// Per plan §Task 16 (strict-contracts force-cutover, rev5): the loader
// constructs the typed pb.IaCProviderRequiredClient + per-optional-service
// clients directly from the plugin's gRPC connection and wraps them in
// typedIaCAdapter (Task 30, PR #605). The legacy remoteIaCProvider
// InvokeService string-dispatch surface is removed entirely — plugins
// that do not register the typed IaCProviderRequired service are
// rejected at load time with an actionable upgrade message.
func discoverAndLoadIaCProvider(ctx context.Context, providerName string, cfg map[string]any) (interfaces.IaCProvider, io.Closer, error) {
Comment thread cmd/wfctl/deploy_providers_test.go Outdated
Comment on lines 167 to 168
pName, _, hasBinary, findErr := findIaCPluginDir(pluginDir, providerName)
if findErr != nil {
intel352 added a commit that referenced this pull request May 10, 2026
Cross-task naming coordination with PR #605/#609 (implementer-2,
Tasks 30/16). Their iac_typed_adapter.go declares 8 sibling consts
naming every typed IaC service:

  iacServiceRequired
  iacServiceEnumerator
  iacServiceDriftDetector
  iacServiceCredentialRevoker
  iacServiceMigrationRepairer
  iacServiceValidator
  iacServiceDriftConfigDetect
  iacServiceResourceDriver

PR #610 originally introduced its own const iacRequiredServiceName
for the same Required service FQN. Spec-reviewer flagged the
inconsistency and asked us to coordinate. Implementer-2 + I agreed
their convention wins (8 siblings establish the pattern; my single
const matches).

Pure rename; no behavior change. Tests still pass.

Verification:
  GOWORK=off go test ./cmd/wfctl/ -run \
    "TestAssertIaCPluginAdvertises|TestIsLegacyIaCPluginErr" \
    -count=1 → PASS (7/7); gofmt clean.

Once PR #610 merges, implementer-2's PR #609 rebase can drop their
duplicate const and import iacServiceRequired from here.
intel352 and others added 3 commits May 10, 2026 06:59
… / ExternalPluginAdapter.Conn()

Architectural prerequisite for plan §Task 16 (wfctl typed-IaC cutover).
The current PluginClient drops the *grpc.ClientConn after constructing
its embedded pb.PluginServiceClient, so callers had no way to build
additional typed gRPC service clients (e.g. pb.IaCProviderRequiredClient,
pb.ResourceDriverClient) against the same plugin process. The legacy
remoteIaCProvider sidestepped this by routing every call through
PluginServiceClient.InvokeService string-dispatch — which is the bug
class force-cutover Task 16 is closing.

Surface (additive — zero call-site impact):
  - `PluginClient.conn *grpc.ClientConn` — retained from GRPCPlugin
    GRPCClient construction.
  - `(p *PluginClient) Conn() *grpc.ClientConn` — opaque accessor;
    exposed via method (not public field) so the rest of PluginClient
    stays internal.
  - `(a *ExternalPluginAdapter) Conn() *grpc.ClientConn` — delegates
    to client.Conn(); nil-safe for adapters constructed via
    `newExternalPluginAdapterWithContractRegistry` (test fixtures).

The connection lifecycle is owned by the host's plugin manager —
callers MUST NOT Close() the returned conn. The plugin shutdown
path tears it down via the registered Closer; closing it externally
would break every other typed-client constructed against the same
process.

## Plan-correction notes

This commit is NOT in plan §Task 16 Files: section (spec gap):
  - Spec assumes the typed pb.IaCProviderRequiredClient can be
    constructed from the existing plugin loader output, but the
    plugin/external surface as it stands strips the underlying conn.
    Task 16 is physically impossible without first exposing it.

Per scope-lock skill the prerequisite lands in the SAME PR as the
Task 16 cutover (this branch) rather than as a separate PR — same
precedent as Task 2's plan-correction notes block. Documented in
the PR description.

Local validation:
  GOWORK=off go build ./plugin/external/...                    → clean
  GOWORK=off go vet ./plugin/external/...                      → clean
  GOWORK=off go test ./plugin/external/... -count=1 -short     → all PASS

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…CProviderRequiredClient (Task 16)

See above commit body in this branch's PR description for full detail.
Summary:
- discoverAndLoadIaCProvider rewritten to construct typedIaCAdapter
  from adapter.Conn() + ContractRegistry-derived registered service map.
- DELETED ~3856 lines: remoteIaCProvider, remoteResourceDriver,
  remoteServiceInvoker/ContextInvoker, jsonToAny, anyToStruct,
  sensitiveToAny, decodeResourceOutput, isPluginMethodUnimplemented,
  stringVal, stringFromMap, loadIaCPlugin var, defaultLoadIaCPlugin,
  readIaCPluginComputePlanVersion, plus 5 dependent test files
  (~2767 lines) and the loadIaCPlugin-using TestResolveIaCProviderSurfacesPluginError.
- KEPT wrapIaCError + retryOnTransient + deployOpError as
  provider-agnostic helpers used by pluginDeployProvider against
  typed RPC errors.
- ADDED registeredIaCServices(reg) helper.

DEPENDENCY: branch CI red until PRs #598 (Task 3 proto) + #605
(Task 30 adapter) merge. Local validation passes against working-tree
overlay of both: build clean, vet clean, ./cmd/wfctl/... -short PASS.

Plan-correction notes (per Path A ruling): Conn() prerequisite shipped
as commit ad7d946 in this same PR. SupportedCanonicalKeys regression
acceptable transient (closes via follow-up additive PR per team-lead
ruling). ComputePlanVersionDeclarer regression tracked for follow-up.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Code-review feedback on PR #609 (Task 16) and post-rebase consolidation
after PR #605 merged. Addresses 4 of 6 Copilot findings + the IMPORTANT
spec-reviewer Step 1 test gap. Fix #1 (Task 18 wiring) deferred until
PR #610 merges (predicate not yet on main); Fix #5 (test file comment)
already cleaned up post-rebase. Fix #6 (computePlanVersion stale return)
documented in-place per team-lead recommendation (avoids 2-cycle churn
when the follow-up capability-extension PR wires it back in).

**Copilot fix #1 — `plugin/external/adapter.go:Conn()` doc broadened.**
The previous doc claimed `nil` only for adapters constructed without a
backing PluginClient (test fixtures). It can ALSO return nil when the
PluginClient is non-nil but its underlying *grpc.ClientConn is nil
(in-process test plumbing wiring only the PluginServiceClient interface
without a real conn). Doc now enumerates both cases.

**Copilot fix #3 — `cmd/wfctl/deploy_providers.go` surfaces ContractRegistryError.**
The previous loader path called `registeredIaCServices(adapter.ContractRegistry())`
without first checking `adapter.ContractRegistryError()`. A transport-
level RPC failure (codes.Unimplemented from a legacy plugin, transient
network reset, etc.) silently degraded to an empty registry, then the
next `if !registered[iacServiceRequired]` branch fired the misleading
"does not register the required service" error — masking the real cause.
Now: surface ContractRegistryError() FIRST with `wfctl plugin update`
hint; fall through to the registration-check only when the RPC succeeded.
Test coverage: `TestDiscoverAndLoadIaCProvider_SurfacesContractRegistryError`.

**Copilot fix #6 — `findIaCPluginDir.computePlanVersion` documented as
reserved for follow-up.** discoverAndLoadIaCProvider no longer reads
the value (the legacy reader `readIaCPluginComputePlanVersion` was
deleted with remoteIaCProvider). Per team-lead: leave the return in
place rather than churning the signature now; a follow-up PR adds
`compute_plan_version` to `CapabilitiesResponse.IaCCapabilityDeclaration`
(option (d), batched with canonical_keys between Task 17 and Task 20)
and wires it back in via the typed Capabilities RPC. In-line comment
documents the reservation + the follow-up plan.

**IMPORTANT spec-reviewer Fix 2 — Step 1 boundary test added.**
New file `cmd/wfctl/discover_typed_loader_test.go` extracts a unit-
testable seam `buildTypedIaCAdapterFrom(adapter)` from the loader's
post-LoadPlugin half (factored out of discoverAndLoadIaCProvider with
the `iacAdapterAccessor` interface so tests don't pay the subprocess
cost). Three boundary tests:

- TestDiscoverAndLoadIaCProvider_ReturnsTypedClient — asserts the
  cutover invariant: loader returns `*typedIaCAdapter`, NOT the
  legacy `*remoteIaCProvider` (which no longer compiles post-cutover).
  In-process gRPC server with a stub IaCProviderRequiredServer +
  Initialize-only response.
- TestDiscoverAndLoadIaCProvider_RejectsMissingRequiredService —
  asserts the strict-contracts hard-cutover invariant: plugins whose
  ContractRegistry omits `IaCProviderRequired` are rejected at load
  time with an actionable `wfctl plugin update` hint. Verifies
  message contract for operator UX.
- TestDiscoverAndLoadIaCProvider_SurfacesContractRegistryError —
  asserts Copilot fix #3 above; transport-level ContractRegistry
  failure is surfaced via errors.Is + RPC-failure framing.

**Test-file comment refresh** in deploy_providers_test.go cite of
iac_typed_adapter_test.go updated post-rebase: file now lives on main
via PR #605 (no longer "not present in this PR"). Also dropped the
unused `noopCloser` helper that lint flagged after the legacy
TestResolveIaCProviderSurfacesPluginError removal.

**Cutover dependency status:** PR #605 (typed adapter) + PR #611 (sdk
auto-register) MERGED to main; rebase clean. PR #610 (Task 18 loader
gate) still pending — Fix 1 (replace inline gate with
AssertIaCPluginAdvertisesRequiredService predicate) lands in a
follow-up commit on this branch once PR #610 merges and I rebase
again. Const naming aligned via impl-3's PR #610 rename
(`iacServiceRequired` is canonical across both files).

Local validation:
  GOWORK=off go build ./...                                    → clean
  GOWORK=off go vet ./cmd/wfctl/ ./plugin/external/...         → clean
  GOWORK=off go test ./cmd/wfctl/ -count=1 -short              → all PASS (7.0s)
  GOWORK=off go test ./cmd/wfctl/ -run TestDiscoverAndLoadIaCProvider -count=1 → 3 PASS (1.5s)
  GOWORK=off golangci-lint run --enable=gocritic,gosec ./cmd/wfctl/...  → 0 issues

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
@intel352 intel352 force-pushed the feat/iac-wfctl-typed-cutover-task16 branch from b5d3fae to b176057 Compare May 10, 2026 11:13
@intel352 intel352 merged commit e32e02b into main May 10, 2026
18 checks passed
@intel352 intel352 deleted the feat/iac-wfctl-typed-cutover-task16 branch May 10, 2026 11:26
intel352 added a commit that referenced this pull request May 10, 2026
…tover runbook (Task 18) (#610)

* feat(proto): add iac.proto with IaCProviderRequired + 6 optional services + 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.

* feat(sdk): RegisterAllIaCProviderServices auto-registration helper

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.

* feat(sdk): ServeIaCPlugin high-level entrypoint with go-plugin GRPCServer 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.

* feat(sdk): BuildContractRegistry advertises registered IaC services

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).

* feat(wfctl): IaC plugin pre-flight gate building block + iac-typed-cutover runbook

Task 18 of the strict-contracts force-cutover plan
(docs/plans/2026-05-10-strict-contracts-force-cutover.md, rev5).

Adds two pieces:

1. cmd/wfctl/iac_loader_gate.go — the pre-flight gate building
   block. AssertIaCPluginAdvertisesRequiredService(name, version,
   registry) inspects a *pb.ContractRegistry response from
   GetContractRegistry and returns nil iff the plugin advertises
   workflow.plugin.external.iac.IaCProviderRequired as a
   CONTRACT_KIND_SERVICE descriptor. On failure: an actionable
   error naming the offending plugin + version, citing
   .wfctl-lock.yaml as the migration target, and pointing to
   docs/runbooks/iac-typed-cutover.md, wrapping
   errLegacyIaCPlugin for IsLegacyIaCPluginErr dispatch.

   The function is the load-bearing predicate the deploy /
   infra-plan / infra-apply call sites will invoke after their
   GetContractRegistry RPC succeeds (wiring left for the typed-
   client cutover already in progress in Task 16; this commit
   ships the testable predicate so the wiring is one-line drop-in).

2. docs/runbooks/iac-typed-cutover.md — operator-facing runbook
   for the cutover. Covers:

   - Compatibility matrix (wfctl × DO plugin version combinations,
     pre-flight expected outcomes).
   - Upgrade order: plugins first, then wfctl. Documents WHY this
     ordering matters (legacy wfctl + v1.0.0 plugin works; v1.0.0
     wfctl + legacy plugin refuses).
   - .wfctl-lock.yaml migration (lockfile shape unchanged; only
     the version pin needs editing).
   - Troubleshooting: legacy-plugin pre-flight error,
     EnumeratorAll-style messages from pre-cutover wfctl,
     state-file decode regressions, advertised-but-buggy plugins.
   - Backout plan: re-pin legacy plugin + wfctl while in the rc1
     window.

Tests (cmd/wfctl/iac_loader_gate_test.go) — 7 cases, all PASS:

- TypedRegistryAccepts: SERVICE-kind descriptor for IaCProviderRequired.
- LegacyRegistryRejects: ContractRegistry without the required
  service → wrapped error names plugin + version + runbook.
- NilRegistryRejects: defensive contract for nil registry input.
- EmptyContractsRejects: no descriptors at all → still legacy.
- WrongKindRejects: ServiceName matches but Kind != SERVICE.
- EmptyMetadataDefaults: name/version absent → graceful "<unknown>".
- IsLegacyIaCPluginErr_NoFalsePositives: sentinel match is typed
  (not string-based) so dispatch sites can errors.Is cleanly.

Stacked on feat/iac-sdk-contracts-task5 (Task 5 PR #602 provides
BuildContractRegistry — the wfctl-side gate inspects what plugins
emit via that helper, so this PR builds on it).

Verification: GOWORK=off go test -race ./cmd/wfctl/... PASS;
GOWORK=off go vet ./cmd/wfctl/ clean; gofmt clean.

Rollback: revert this commit. The pre-flight gate is a standalone
predicate function — no production code currently depends on it
(dispatch wiring lands in the typed-client cutover Task 16). The
runbook is documentation; no operational impact from removal.

* docs(runbook): correct cutover model — wfctl-side rc1 adapter, not plugin compat shim

Per spec-reviewer's PR 610 IMPORTANT-1 + team-lead's ruling on the
correct cutover model (which I as ADR 0024 author should have known
already):

The runbook previously claimed "v1.0.0 plugins also expose the legacy
InvokeService surface during the rc1 window for backward compatibility"
and "backout fully supported through rc1 window (workflow v1.0.0-rc1
+ DO plugin v1.0.0)". Both wrong. Contradicts ADR 0024
(force-cutover, no compat shim, no build-tag dual-path), Task 9 spec
literal ("DELETE module_instance.go"), and workspace memory
feedback_force_strict_contracts_no_compat.

Corrected cutover model:

- workflow v1.0.0-rc1 ships the wfctl-SIDE typed-client adapter
  alongside the existing wfctl remoteIaCProvider. Plugins are
  unchanged at v0.14.x; the typed adapter is exercised only when a
  typed-aware plugin is loaded. This is rc1's role: wfctl-side
  additive, not plugin-side compat shim.
- DO plugin v1.0.0 ships typed-only (Task 9 deletes the legacy
  internal/module_instance.go switch dispatcher entirely; ResourceDriver
  Task 11 follows the same pattern).
- workflow v1.0.0 final ships typed-only on the wfctl side too (Task
  20 removes wfctl-side remoteIaCProvider). Both legacy paths are gone.

Specific revisions applied:

1. Compatibility matrix: added the missing v0.27.x × v1.0.0 row
   marked "BROKEN by design" — operators MUST run wfctl rc1+ to
   consume DO v1.0.0 because the legacy wfctl binary cannot dispatch
   through a typed-only plugin. Added explicit Step 4 troubleshooting
   block for operators who skipped wfctl rc1.

2. Upgrade order rewrite: 5 steps (was 3). Order is wfctl rc1 first
   (test against v0.14.x plugin set), then DO plugin v1.0.0, then
   wfctl v1.0.0 final. Each step has its smoke-test command and
   rollback escape.

3. Backout plan rewrite: explicit "rollback BOTH wfctl AND plugin
   together" for the post-v1.0.0-final case. Single-side rollback is
   broken — wfctl v1.0.0 final binary cannot dispatch v0.14.x plugin
   (legacy adapter deleted in Task 20); legacy v0.27.x wfctl cannot
   dispatch typed-only DO v1.0.0 (no typed adapter on the legacy
   binary). Documents the partial-rollback option that exists during
   the rc1 window only.

4. Top-of-runbook callout: explicit statement that this is a hard
   cutover with no plugin-side compat shim, citing ADR 0024 and
   feedback_force_strict_contracts_no_compat.

No code changes; documentation only.

Verification: markdown rendered locally; cross-references to ADRs +
plan + design + state_compat_test.go all resolve.

Followups (per spec-reviewer + team-lead Finding 2):
- PR description gets a "Wiring deferred to Task 16 (PR 609)" callout
  with predicate symbol + sentinel + call-site location
- Coordination DM to implementer-2 with the wiring contract

* refactor(wfctl): rename iacRequiredServiceName → iacServiceRequired

Cross-task naming coordination with PR #605/#609 (implementer-2,
Tasks 30/16). Their iac_typed_adapter.go declares 8 sibling consts
naming every typed IaC service:

  iacServiceRequired
  iacServiceEnumerator
  iacServiceDriftDetector
  iacServiceCredentialRevoker
  iacServiceMigrationRepairer
  iacServiceValidator
  iacServiceDriftConfigDetect
  iacServiceResourceDriver

PR #610 originally introduced its own const iacRequiredServiceName
for the same Required service FQN. Spec-reviewer flagged the
inconsistency and asked us to coordinate. Implementer-2 + I agreed
their convention wins (8 siblings establish the pattern; my single
const matches).

Pure rename; no behavior change. Tests still pass.

Verification:
  GOWORK=off go test ./cmd/wfctl/ -run \
    "TestAssertIaCPluginAdvertises|TestIsLegacyIaCPluginErr" \
    -count=1 → PASS (7/7); gofmt clean.

Once PR #610 merges, implementer-2's PR #609 rebase can drop their
duplicate const and import iacServiceRequired from here.

* fix(wfctl): drop duplicate iacServiceRequired const (now lives in iac_typed_adapter.go on main)

PR #605 merged iac_typed_adapter.go to main with iacServiceRequired const.
PR #610's iac_loader_gate.go declared it independently; collision after
cascade-merge. Removed the duplicate; use canonical const from
iac_typed_adapter.go directly. Build clean.

🤖 Generated with [Claude Code](https://claude.com/claude-code)
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