Sync ako/mxcli: test --reuse-build and --watch recovery (#1783, #1787); Go 1.27.4 for GO-2026-6859; page round-trip fixes

.claude/skills/fix-issue/findings/mdl-executor/2026-11-11-page-round-trip-loses-widget-settings.json

@@ -0,0 +1 @@
+{"area": "mdl/executor", "date": "2026-11-11", "refs": ["ako/mxcli#2106"], "ce": [], "symptom": "describe page -> exec of FactoryManagement pages reported Replaced page and changed three settings: a check box's LabelPosition AfterControl became Default (ConversationalUI.TraceFiltering_Edit), a radio button group's RenderHorizontal false became true (TcConnector.TeamcenterConfiguration_NewEdit), and a show page button's NumberOfPagesToClose2 '1' became '' (SnowflakeRESTSQL.ConnectionDetails_NewEdit_Step2). mx check stays green on both sides; the page behaves differently.", "file": "sdk/pages (CheckBox.LabelPosition, PageClientAction.NumberOfPagesToClose); mdl/executor cmd_pages_builder_v3_widgets.go (LabelPosition, RenderHorizontal), cmd_pages_describe_parse.go + cmd_pages_describe_output.go (read and print both), cmd_pages_describe_action_settings.go (NumberOfPagesToClose in with (…)), page_action_settings.go, widget_native_props.go (MDL-WIDGET52 reads), validate_widgets.go (known props); mdl/visitor/visitor_action_settings.go (showPage kind, numeric value); grammar actionSettingV3 takes NUMBER_LITERAL; mdl/backend/modelsdk/widget_write.go (SetLabelPosition, SetNumberOfPagesToClose2). Tests: cmd_pages_round_trip_settings_2106_test.go, visitor_show_page_close_2106_test.go, widget_write_round_trip_settings_2106_test.go", "cause": "None of the three had an MDL spelling. The writer wrote the default each time: NumberOfPagesToClose2 hard-coded to \"\", LabelPosition never set, and RenderHorizontal derived from a RenderDirection nothing ever filled, so every rewritten radio button group became horizontal.", "insight": "**A value-only BSON diff of the first exec, swept over a whole app, separates real losses from unexpressible formatting.** Of FactoryManagement's 14 'Replaced' pages, these three change behaviour; the other eleven differ in values MDL cannot state (TabIndex, trailing newlines in DynamicClasses, XPath whitespace, condition order, an inherited attribute's declaring entity, a null SourceVariable key, an unused placeholder parameter), which a canon carry handles separately. A model field the writer reads but no builder sets (RadioButtons.RenderDirection) is a silent default: grep each writer input for a builder that assigns it. Controls: reverting each of the ten edits fails its test; a new page using all three builds with mx check 0 errors on the Agent Builder Starter App; TraceFiltering_Edit and ConnectionDetails_NewEdit_Step2 now report Unchanged page."}

.claude/skills/fix-issue/findings/modelsdk/2026-11-11-page-round-trip-carry-formatting-and-property-sets.json

@@ -0,0 +1 @@
+{"area": "modelsdk/canon", "date": "2026-11-11", "refs": ["ako/mxcli#2106", "ako/mxcli#2052"], "ce": [], "symptom": "describe page -> exec of 8 FactoryManagement pages reported Replaced page with nothing changed: a DynamicClasses expression lost its trailing newlines (3 pages), an XPath constraint was reformatted (2), an enumeration condition's values were reordered (1), an AssociationSource lost its null SourceVariable key (2), and an inherited createdDate binding was requalified from GenAICommons.Span to GenAICommons.ToolSpan (1).", "file": "modelsdk/canon/unexpressed.go (whitespaceInsensitive for Expression, DynamicClasses, XPathConstraint; sameConditionsReordered; anyIDUsedOutside), mdl/backend/modelsdk/widget_write.go (Forms$AssociationSource NullFields SourceVariable), mdl/executor/cmd_pages_builder_input.go entityAttributeOwners (system members count as the storing entity's own). Tests: modelsdk/canon/unexpressed_2106_test.go, widget_association_source_null_2106_test.go, cmd_pages_builder_system_member_2106_test.go", "cause": "#2052's carry compared only Expression and only by TrimSpace; DynamicClasses and XPathConstraint are expressions too, and Studio Pro keeps the author's interior line breaks. The writer emits enumeration conditions in enumeration order where Studio Pro keeps insertion order. Forms$AssociationSource had no NullFields registration (Studio Pro stores the key on 40 of 40). declaringEntityFor indexed declared attributes only, so a system member a specialization inherits resolved to the context entity.", "insight": "**Census the stored form across a whole app before choosing a carry or a writer fix.** 40 of 40 association sources carry SourceVariable, so it is a property-set gap fixed once in the writer; the condition order differs only by authoring history, so it is a carry. And check a 'formatting' difference before carrying it: the placeholder 'unused parameter' this sweep also showed is the template's only, USED, expression parameter that MDL cannot state - a real loss for new syntax, which the carry rule rightly would not have touched. Controls: each test fails without its change; a full sweep of FactoryManagement's 229 pages turns exactly these 8 Unchanged and regresses none."}

.github/workflows/nightly.yml

@@ -41,7 +41,7 @@ jobs:
 
       - uses: actions/setup-go@v7
         with:
-          go-version: '1.27.3'
+          go-version: '1.27.4'
 
       - name: Cache ANTLR4 JAR
         uses: actions/cache@v6
@@ -90,7 +90,7 @@ jobs:
 
       - uses: actions/setup-go@v7
         with:
-          go-version: '1.27.3'
+          go-version: '1.27.4'
 
       # mxbuild 9 draws through System.Drawing; without libgdiplus every mx call
       # dies with "The type initializer for 'Gdip' threw an exception."
@@ -141,7 +141,7 @@ jobs:
       - uses: actions/checkout@v7
       - uses: actions/setup-go@v7
         with:
-          go-version: '1.27.3'
+          go-version: '1.27.4'
       - name: Test the tunnel seam
         run: bash scripts/test-tunnel-seam.sh
 
@@ -162,7 +162,7 @@ jobs:
       - uses: actions/checkout@v7
       - uses: actions/setup-go@v7
         with:
-          go-version: '1.27.3'
+          go-version: '1.27.4'
       - name: Cache ANTLR4 JAR
         uses: actions/cache@v6
         with:
@@ -193,7 +193,7 @@ jobs:
 
       - uses: actions/setup-go@v7
         with:
-          go-version: '1.27.3'
+          go-version: '1.27.4'
 
       - uses: oven-sh/setup-bun@v2
 

.github/workflows/push-test.yml

@@ -49,7 +49,7 @@ jobs:
         with:
           # The toolchain matters: gofmt's output differs between Go releases
           # (the mismatch fixed in #1759), so this must match the other jobs.
-          go-version: '1.27.3'
+          go-version: '1.27.4'
           cache: false
       - name: Check gofmt
         run: make fmt-check
@@ -100,7 +100,7 @@ jobs:
       - uses: actions/checkout@v7
       - uses: actions/setup-go@v7
         with:
-          go-version: '1.27.3'
+          go-version: '1.27.4'
       - name: Test the tunnel seam
         shell: bash
         run: bash scripts/test-tunnel-seam.sh
@@ -120,7 +120,7 @@ jobs:
       - uses: actions/checkout@v7
       - uses: actions/setup-go@v7
         with:
-          go-version: '1.27.3'
+          go-version: '1.27.4'
       - name: Test the Windows process helpers
         shell: bash
         # Scoped with -run: the rest of the docker package's tests assert POSIX
@@ -165,7 +165,7 @@ jobs:
           submodules: true
       - uses: actions/setup-go@v7
         with:
-          go-version: '1.27.3'
+          go-version: '1.27.4'
       - name: Cache ANTLR4 JAR
         uses: actions/cache@v6
         with:
@@ -289,7 +289,7 @@ jobs:
           submodules: ${{ matrix.suite == 'roundtrip' || matrix.suite == 'parity' }}
       - uses: actions/setup-go@v7
         with:
-          go-version: '1.27.3'
+          go-version: '1.27.4'
       - name: Cache ANTLR4 JAR
         uses: actions/cache@v6
         with:

.github/workflows/release.yml

@@ -17,7 +17,7 @@ jobs:
           fetch-depth: 0
       - uses: actions/setup-go@v7
         with:
-          go-version: '1.27.3'
+          go-version: '1.27.4'
       - uses: oven-sh/setup-bun@v2
       - name: Cache ANTLR4 JAR
         uses: actions/cache@v6

CHANGELOG.md

@@ -8,6 +8,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
 ### Added
 
+- **`mxcli test --local --reuse-build` skips the injection and the build when nothing changed** (mendixlabs/mxcli#1783, closes mendixlabs/mxcli#1782) — a local test run injects its test microflows, builds and boots; when the project, the test files and mxcli are unchanged, the deployment the previous run left is already this run's build, so `--reuse-build` boots it directly (about 12 s against 60 s, measured by the author). The decision is a content hash of the project folder (outputs, caches, `theme-cache/`, and any `.git`/`node_modules` left out), the generated test MDL, the mxcli binary and the mxbuild override, pinned to the identity of `deployment/model/model.mdp`, so a deployment rebuilt by anything else is never taken for the test build. Not under `--watch` or `--skip-build`.
 - **A `genai-runtime` skill for the runtime side of Agents Kit 2** (ako/mxcli#2068) — registering a Mendix Cloud GenAI key or an OpenAI-compatible endpoint, seeding AgentCommons agents from JSON (matched by `Title` only, so a changed JSON never reaches a stored agent), resuming Conversational UI chats, loading PgVector and Mendix Cloud knowledge bases, and traces. It collects the gotchas that `check` and `mx check` cannot see: the encryption key both the OpenAI and PgVector setup need, the `vector` extension only the UI's Test button creates, the 60–120 s indexing delay, the chat that collapses outside its page, and `StoreTraces` defaulting to false. Its examples pass `check --references` and build with 0 errors against the Agent Builder Starter App's modules.
 - **`mxcli test --attach` runs a test without building it, against an app started with `--allow-call`** (mendixlabs/mxcli#1788, mendixlabs/mxcli#1789) — `run --local --test-endpoint --allow-call` switches on an executor in the test endpoint. A `.test.mdl` test whose statements it supports (`declare`, `set`, `create`, `change`, `commit`, `delete`, `call microflow`, database `retrieve` with `where`/`first`; `@expect` with `count`, `@throws`, `@setup`, `@cleanup`) is compiled by mxcli into steps and run in the app with the Mendix runtime API, every expression evaluated by `Core.evaluateExpression`. Nothing is injected or built, so a test edit costs one HTTP round trip: on a small app the suite took 0.26 s against 3.2 s for inject-and-build, with identical verdicts. Retrieves over an association and the list aggregates `count`/`sum`/`average`/`minimum`/`maximum` over an attribute run there too. A test with any other statement is injected and built as before. Only that test is, while the rest of the run still goes straight to the app, and the output names the test and the MDL statement (`the statement if`, `a list operation (find)`). Because the runtime evaluator returns empty for an unknown name, an undeclared `$variable` and a non-existent enumeration value are refused (ERROR) before evaluating, and a model newer than the running build is refused with `mxcli run wait` as the fix under `--watch`. The route takes POST, needs the run's token and a loopback caller, and answers 403 without `--allow-call`, since it calls the microflows a test names in a system context. `.test.mdl` stays the only test language. Design discussed with the authors of #1789; measurements in `mdl-examples/spikes/evaluate-expression.mdl`.
 - **`flows_for()` and `CATALOG.FLOWS`: a flow's sequence flows in lint rules, and each activity's place on the canvas** (ako/mxcli#1299, mendixlabs/mxcli#1308) — `activities_for()` listed the objects but not which leads to which, so a rule could not tell which log belongs to which custom error handler, which activity an annotation explains, or which branch of a split is `true`. Each sequence and annotation flow is now a row with its origin and destination activity, `is_error_handler`, the branch's `case_value` and the sides it connects; activities gain `x`, `y`, `width` and `height`. Full catalog; the schema moves to version 44, so a cached catalog is rebuilt once.
@@ -26,6 +27,9 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
 ### Fixed
 
+- **`describe page` → `exec` no longer rewrites a page over formatting and stored keys MDL cannot state** (ako/mxcli#2106) — on FactoryManagement, 8 pages reported `Replaced page` with nothing changed: a dynamic-classes expression's trailing newlines, an XPath constraint's line breaks and spacing, the order of an enumeration condition's values, an association source's null `SourceVariable` key, and an inherited `createdDate` binding requalified to the specialization. The stored value is now kept where only its form differs, the writer stores the key Studio Pro stores, and a system member resolves to the entity that stores it. A full sweep of the app's 229 pages turns those 8 `Unchanged` and changes no other.
+- **`describe page` → `exec` keeps a check box's label position, a radio button group's orientation and a show page action's close count** (ako/mxcli#2106) — none had an MDL spelling, so a round trip of a Studio Pro page reset a check box with its label after the control to the default, turned every vertical radio button group horizontal, and stopped a show page button from closing the page it came from, with `mx check` green. `checkbox` takes `LabelPosition: Default|BeforeControl|AfterControl`, `radiobuttons` takes `RenderHorizontal: false`, and `show page` takes `with (NumberOfPagesToClose: n)`; `describe` prints each when it differs from the default.
+- **`mxcli test --local --watch` keeps applying changes on macOS and Linux mxbuild** (mendixlabs/mxcli#1787, refs mendixlabs/mxcli#1786) — the test runner's rebuild called mxbuild's serve process directly, without the recovery `run --local --watch` has, so the second build of a serve process failed on regenerated pages and every later change reported "could not be applied". The rebuild now goes through the same recovery, which also recognises mxbuild's resource-key form of that failure (`IO_PathNotFound_Path`, as macOS and Linux report it) and skips the web-bundle check, which only matters when a browser is served.
 - **`exec` and `check` no longer rescan an MPR v1 project for every lookup** (ako/mxcli#2094, mendixlabs/mxcli#1795) — each statement cost 0.3–0.9 s on a large single-file project, whatever it changed: every typed listing re-read the metadata of the whole `Unit` table, and every lookup by name read every microflow's contents. The reader now keeps the scan, and each unit's name, until a write or another process's commit says the table may have changed. On a 3,400-unit project a `create microflow` went from 0.30 s to 0.11 s, a one-activity `create or modify` from 0.86 s to 0.35 s, and `check` of 150 such statements from 76 s to 15 s. MPR v2 projects already kept this index.
 - **`mxcli test --local` beside a `run --local --test-endpoint` dev loop runs, and leaves the project buildable** (ako/mxcli#2119) — the test run read the dev loop's `MxTest.RegisterEndpoint` as the project's after-startup microflow and chained its own endpoint to it, to itself: the scratch runtime failed to boot with a `StackOverflowError` and no test ran. Its cleanup then dropped the dev loop's endpoint while the setting still named it, so the dev loop's next build failed with CE1613. A live host's restore record now says what the endpoint replaced, and the test run chains to that; its cleanup removes only its test microflows and leaves the dev loop's endpoint, Java source and setting as it found them. An endpoint whose run died is removed first, as `run` does, and one with no record of what it replaced is refused rather than chained to. When the test run rebuilt the `deployment/` folder both share, the dev loop builds its own back once the run ends (a reload or a restart), so a following `--attach` is not refused for a restart.
 - **A test run no longer rebuilds or restarts the dev loop it shares a project with** (ako/mxcli#2063, field test on ako/mxcli#2104) — `mxcli test --local`, and `--attach` with a test that has to be built, write their test microflows into the project a `run --local --watch` dev loop is watching, so the dev loop rebuilt for the injection and its cleanup, and applied it by restarting the runtime, which drops every browser session, `--hub` preview included; on a small app the restart itself failed (`EOFException`). The test run now holds the dev loop's watcher (`.mxcli/watch-hold.json`) while it owns the project, `--attach` now puts the project file back byte for byte as `--local` already did, and on release the watcher compares the model with the one the hold began from: the same model is not built at all, and `run status` / `run wait` read it as covered. An edit made before or during the run is still built once it ends, and a hold whose run died is ignored.

cmd/mxcli/cmd_test_run.go

@@ -166,6 +166,7 @@ Examples:
 		junitOutput, _ := cmd.Flags().GetString("junit")
 		requireAssertions, _ := cmd.Flags().GetBool("require-assertions")
 		skipBuild, _ := cmd.Flags().GetBool("skip-build")
+		reuseBuild, _ := cmd.Flags().GetBool("reuse-build")
 		local, _ := cmd.Flags().GetBool("local")
 		legacyRunner, _ := cmd.Flags().GetBool("legacy-runner")
 		watch, _ := cmd.Flags().GetBool("watch")
@@ -214,6 +215,7 @@ Examples:
 			ProjectPath:       projectPath,
 			TestFiles:         resolveTestPaths(args, projectPath),
 			SkipBuild:         skipBuild,
+			ReuseBuild:        reuseBuild,
 			Local:             local,
 			LegacyRunner:      legacyRunner,
 			Watch:             watch,

cmd/mxcli/docker/localapp.go

@@ -108,6 +108,9 @@ type LocalApp struct {
 	RuntimeLogPath string
 
 	serve *ServeServer
+	// rec runs Rebuild's builds with the recovery `run --local --watch` has: a
+	// build whose outcome only a fresh mxbuild --serve fixes is repeated on one.
+	rec *serveRecovery
 }
 
 func (o *LocalAppOptions) applyDefaults() {
@@ -323,6 +326,7 @@ func StartLocalApp(opts LocalAppOptions) (*LocalApp, error) {
 			return nil, fmt.Errorf("starting mxbuild serve: %w", err)
 		}
 		app.serve = serve
+		app.rec = &serveRecovery{serve: serve, deployDir: opts.DeployDir, out: w, headless: true}
 
 		build, err := buildRetryingServeTimeout(serve, BuildRequest{Target: TargetDeploy, ProjectFilePath: opts.ProjectPath}, w)
 		if err != nil {
@@ -365,7 +369,14 @@ func (a *LocalApp) Rebuild(projectPath string) (ApplyAction, *BuildResult, error
 	if a.Runtime == nil {
 		return ActionReload, nil, fmt.Errorf("the runtime is not running")
 	}
-	build, err := a.serve.Build(BuildRequest{Target: TargetDeploy, ProjectFilePath: projectPath})
+	req := BuildRequest{Target: TargetDeploy, ProjectFilePath: projectPath}
+	var build *BuildResult
+	var err error
+	if a.rec != nil {
+		build, err = a.rec.Build(req)
+	} else {
+		build, err = a.serve.Build(req)
+	}
 	if err != nil {
 		return ActionReload, nil, err
 	}

cmd/mxcli/docker/serve_recovery.go

@@ -74,6 +74,10 @@ type serveRecovery struct {
 	// failed. A theme failure then is not trustworthy: the process may be
 	// reporting the earlier error rather than the source on disk.
 	themeFailed bool
+
+	// headless: nothing serves the client (the test runner), so a build that
+	// succeeds with an incomplete web/dist is a good build and is not repeated.
+	headless bool
 }
 
 func (r *serveRecovery) Build(req BuildRequest) (*BuildResult, error) {
@@ -105,7 +109,7 @@ func (r *serveRecovery) Build(req BuildRequest) (*BuildResult, error) {
 	}
 
 	if res.OK() {
-		if !prebuilt {
+		if !prebuilt || r.headless {
 			return res, nil
 		}
 		after, _ := readPrebuiltBundle(r.deployDir)