@smoothbricks/nx-plugin 0.4.26 → 0.4.28
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +994 -117
- package/dist/bin/smoo-nx-bound-failures.d.ts +3 -0
- package/dist/bin/smoo-nx-bound-failures.d.ts.map +1 -0
- package/dist/bin/smoo-nx-bound-failures.js +31 -0
- package/dist/bin/smoo-nx-cargo-hash.js +28 -3
- package/dist/bin/smoo-nx-exec.js +15 -73
- package/dist/bin/smoo-nx-failures.d.ts +3 -0
- package/dist/bin/smoo-nx-failures.d.ts.map +1 -0
- package/dist/bin/smoo-nx-failures.js +3 -0
- package/dist/bin/smoo-nx-nextest-extract.js +3 -2
- package/dist/bin/smoo-nx-test-walks.d.ts +3 -0
- package/dist/bin/smoo-nx-test-walks.d.ts.map +1 -0
- package/dist/bin/smoo-nx-test-walks.js +34 -0
- package/dist/bounded-test-policy.d.ts +7 -1
- package/dist/bounded-test-policy.d.ts.map +1 -1
- package/dist/bounded-test-policy.js +101 -40
- package/dist/build-state-output-policy.d.ts +25 -0
- package/dist/build-state-output-policy.d.ts.map +1 -0
- package/dist/build-state-output-policy.js +64 -0
- package/dist/cargo-closure-input.d.ts +3 -2
- package/dist/cargo-closure-input.d.ts.map +1 -1
- package/dist/cargo-closure-input.js +16 -15
- package/dist/cargo-dev-profile.d.ts +37 -0
- package/dist/cargo-dev-profile.d.ts.map +1 -0
- package/dist/cargo-dev-profile.js +108 -0
- package/dist/cargo-source-hash.d.ts +43 -6
- package/dist/cargo-source-hash.d.ts.map +1 -1
- package/dist/cargo-source-hash.js +309 -73
- package/dist/cargo-toolchain-policy.d.ts +4 -0
- package/dist/cargo-toolchain-policy.d.ts.map +1 -1
- package/dist/cargo-toolchain-policy.js +5 -0
- package/dist/cargo-workspace.d.ts +101 -5
- package/dist/cargo-workspace.d.ts.map +1 -1
- package/dist/cargo-workspace.js +291 -10
- package/dist/cross-check-policy.d.ts +21 -3
- package/dist/cross-check-policy.d.ts.map +1 -1
- package/dist/cross-check-policy.js +24 -6
- package/dist/disk-command-policy.d.ts +7 -0
- package/dist/disk-command-policy.d.ts.map +1 -0
- package/dist/disk-command-policy.js +98 -0
- package/dist/ensure-built-receipt.d.ts +25 -0
- package/dist/ensure-built-receipt.d.ts.map +1 -0
- package/dist/ensure-built-receipt.js +261 -0
- package/dist/ensure-built-witness.d.ts +57 -0
- package/dist/ensure-built-witness.d.ts.map +1 -0
- package/dist/ensure-built-witness.js +412 -0
- package/dist/ensure-built.d.ts +207 -12
- package/dist/ensure-built.d.ts.map +1 -1
- package/dist/ensure-built.js +576 -202
- package/dist/eslint-file-set.d.ts +7 -0
- package/dist/eslint-file-set.d.ts.map +1 -0
- package/dist/eslint-file-set.js +12 -0
- package/dist/executors/bounded-exec/cpu-pool.d.ts +63 -0
- package/dist/executors/bounded-exec/cpu-pool.d.ts.map +1 -0
- package/dist/executors/bounded-exec/cpu-pool.js +379 -0
- package/dist/executors/bounded-exec/cpu-tokens.d.ts +82 -0
- package/dist/executors/bounded-exec/cpu-tokens.d.ts.map +1 -0
- package/dist/executors/bounded-exec/cpu-tokens.js +114 -0
- package/dist/executors/bounded-exec/disk-lease.d.ts +61 -0
- package/dist/executors/bounded-exec/disk-lease.d.ts.map +1 -0
- package/dist/executors/bounded-exec/disk-lease.js +91 -0
- package/dist/executors/bounded-exec/executor.d.ts +45 -2
- package/dist/executors/bounded-exec/executor.d.ts.map +1 -1
- package/dist/executors/bounded-exec/executor.js +447 -52
- package/dist/executors/bounded-exec/gateway-lease.d.ts +52 -0
- package/dist/executors/bounded-exec/gateway-lease.d.ts.map +1 -0
- package/dist/executors/bounded-exec/gateway-lease.js +143 -0
- package/dist/executors/bounded-exec/hang-diagnostics.d.ts +35 -0
- package/dist/executors/bounded-exec/hang-diagnostics.d.ts.map +1 -0
- package/dist/executors/bounded-exec/hang-diagnostics.js +231 -0
- package/dist/executors/bounded-exec/progress.d.ts +35 -0
- package/dist/executors/bounded-exec/progress.d.ts.map +1 -0
- package/dist/executors/bounded-exec/progress.js +79 -0
- package/dist/executors/bounded-exec/ram-temp-fixtures.d.ts +3 -0
- package/dist/executors/bounded-exec/ram-temp-fixtures.d.ts.map +1 -0
- package/dist/executors/bounded-exec/ram-temp-fixtures.js +12 -0
- package/dist/executors/bounded-exec/ram-temp-held.d.ts +91 -0
- package/dist/executors/bounded-exec/ram-temp-held.d.ts.map +1 -0
- package/dist/executors/bounded-exec/ram-temp-held.js +152 -0
- package/dist/executors/bounded-exec/ram-temp.d.ts +223 -0
- package/dist/executors/bounded-exec/ram-temp.d.ts.map +1 -0
- package/dist/executors/bounded-exec/ram-temp.js +1528 -0
- package/dist/executors/bounded-exec/schema.d.ts +29 -11
- package/dist/executors/bounded-exec/schema.d.ts.map +1 -1
- package/dist/executors/bounded-exec/verdict.d.ts +162 -0
- package/dist/executors/bounded-exec/verdict.d.ts.map +1 -0
- package/dist/executors/bounded-exec/verdict.js +318 -0
- package/dist/executors/shard-timings/executor.d.ts +14 -0
- package/dist/executors/shard-timings/executor.d.ts.map +1 -0
- package/dist/executors/shard-timings/executor.js +25 -0
- package/dist/executors/shard-timings/schema.d.ts +4 -0
- package/dist/executors/shard-timings/schema.d.ts.map +1 -0
- package/dist/executors/shard-timings/schema.js +1 -0
- package/dist/failure-summary.d.ts +5 -0
- package/dist/failure-summary.d.ts.map +1 -0
- package/dist/failure-summary.js +118 -0
- package/dist/generators/create-package/generator.d.ts.map +1 -1
- package/dist/generators/create-package/generator.js +6 -2
- package/dist/index.d.ts +14 -2
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +370 -138
- package/dist/managed-files/ci-workflow.d.ts +13 -0
- package/dist/managed-files/ci-workflow.d.ts.map +1 -1
- package/dist/managed-files/ci-workflow.js +44 -13
- package/dist/managed-files/files.d.ts +7 -1
- package/dist/managed-files/files.d.ts.map +1 -1
- package/dist/managed-files/files.js +62 -8
- package/dist/managed-files/json-edit.d.ts +21 -0
- package/dist/managed-files/json-edit.d.ts.map +1 -0
- package/dist/managed-files/json-edit.js +295 -0
- package/dist/managed-files/managed-format.d.ts +33 -1
- package/dist/managed-files/managed-format.d.ts.map +1 -1
- package/dist/managed-files/managed-format.js +82 -2
- package/dist/managed-files/pr-preview-cleanup-workflow.d.ts +7 -0
- package/dist/managed-files/pr-preview-cleanup-workflow.d.ts.map +1 -1
- package/dist/managed-files/pr-preview-cleanup-workflow.js +68 -4
- package/dist/managed-files/publish-workflow.d.ts.map +1 -1
- package/dist/managed-files/publish-workflow.js +231 -48
- package/dist/managed-files/tree.d.ts +5 -0
- package/dist/managed-files/tree.d.ts.map +1 -1
- package/dist/managed-files/tree.js +5 -3
- package/dist/manifest-hash.d.ts.map +1 -1
- package/dist/manifest-hash.js +2 -5
- package/dist/nextest-code-bound.d.ts +62 -0
- package/dist/nextest-code-bound.d.ts.map +1 -0
- package/dist/nextest-code-bound.js +154 -0
- package/dist/nextest-extraction.d.ts +24 -8
- package/dist/nextest-extraction.d.ts.map +1 -1
- package/dist/nextest-extraction.js +42 -7
- package/dist/package-target-policy.d.ts.map +1 -1
- package/dist/package-target-policy.js +3 -5
- package/dist/release-config-policy.d.ts.map +1 -1
- package/dist/release-config-policy.js +3 -5
- package/dist/runtime-input.d.ts +14 -0
- package/dist/runtime-input.d.ts.map +1 -0
- package/dist/runtime-input.js +15 -0
- package/dist/rust-tokens.d.ts +11 -0
- package/dist/rust-tokens.d.ts.map +1 -0
- package/dist/rust-tokens.js +81 -0
- package/dist/task-results.d.ts +6 -0
- package/dist/task-results.d.ts.map +1 -0
- package/dist/task-results.js +61 -0
- package/dist/test-program.d.ts +54 -0
- package/dist/test-program.d.ts.map +1 -0
- package/dist/test-program.js +292 -0
- package/dist/test-timings.d.ts +71 -0
- package/dist/test-timings.d.ts.map +1 -0
- package/dist/test-timings.js +393 -0
- package/dist/test-walk-policy.d.ts +103 -0
- package/dist/test-walk-policy.d.ts.map +1 -0
- package/dist/test-walk-policy.js +657 -0
- package/dist/testing.d.ts +177 -0
- package/dist/testing.d.ts.map +1 -0
- package/dist/testing.js +824 -0
- package/dist/type-guards.d.ts +3 -0
- package/dist/type-guards.d.ts.map +1 -0
- package/dist/type-guards.js +4 -0
- package/dist/typecheck-test-policy.d.ts +8 -5
- package/dist/typecheck-test-policy.d.ts.map +1 -1
- package/dist/typecheck-test-policy.js +113 -100
- package/dist/workspace-config-policy.d.ts +8 -0
- package/dist/workspace-config-policy.d.ts.map +1 -1
- package/dist/workspace-config-policy.js +175 -63
- package/dist/workspace-index.d.ts +12 -0
- package/dist/workspace-index.d.ts.map +1 -0
- package/dist/workspace-index.js +17 -0
- package/dist/workspace-manifest.d.ts +37 -0
- package/dist/workspace-manifest.d.ts.map +1 -1
- package/executors.json +5 -0
- package/loader.js +13 -11
- package/managed/raw/envrc +33 -67
- package/managed/raw/git-format-staged.yml +30 -35
- package/managed/raw/patches/nx@23.2.1.patch +3484 -0
- package/managed/raw/tooling/direnv/apple-developer.sh +83 -0
- package/managed/raw/tooling/direnv/developer-links.ts +569 -11
- package/managed/raw/tooling/direnv/devenv.smoo.nix +314 -93
- package/managed/raw/tooling/direnv/github-actions-bootstrap.sh +20 -11
- package/managed/raw/tooling/direnv/linked-snapshots.ts +845 -0
- package/managed/raw/tooling/direnv/nx-socket-dir.sh +82 -0
- package/managed/raw/tooling/direnv/repo-path +14 -2
- package/managed/raw/tooling/direnv/setup-environment.ts +1146 -159
- package/managed/raw/tooling/direnv/toolchain-stamp.ts +45 -3
- package/managed/raw/tooling/direnv/ttsc-env.sh +39 -0
- package/managed/raw/tooling/git-hooks/commit-msg.sh +5 -6
- package/managed/raw/tooling/git-hooks/pre-commit.sh +7 -6
- package/managed/raw/tooling/git-hooks/pre-push.sh +20 -15
- package/managed/templates/github/actions/cache-nix-devenv/action.yml +6 -6
- package/managed/templates/github/actions/cache-node-modules/action.yml +6 -4
- package/managed/templates/github/actions/cache-nx/action.yml +1 -1
- package/managed/templates/github/actions/cache-ttsc-plugins/action.yml +1 -1
- package/managed/templates/github/actions/forgejo-artifact.cjs +99 -0
- package/managed/templates/github/actions/forgejo-download-artifact/action.yml +49 -0
- package/managed/templates/github/actions/forgejo-download-artifact/index.cjs +7 -0
- package/managed/templates/github/actions/forgejo-upload-artifact/action.yml +65 -0
- package/managed/templates/github/actions/forgejo-upload-artifact/index.cjs +7 -0
- package/managed/templates/github/actions/setup-devenv/action.yml +21 -3
- package/managed/templates/github/workflows/managed-files.yml +1 -2
- package/nextest.toml +8 -41
- package/package.json +74 -12
- package/runtime-input.sh +81 -0
- package/src/__tests__/counted-cargo.ts +33 -14
- package/src/__tests__/fixture-nx-env.ts +111 -46
- package/src/__tests__/progress-clock.ts +43 -0
- package/src/__tests__/runtime-input.ts +23 -0
- package/src/bin/smoo-nx-bound-failures.test.ts +127 -0
- package/src/bin/smoo-nx-bound-failures.ts +33 -0
- package/src/bin/smoo-nx-cargo-hash.ts +28 -3
- package/src/bin/smoo-nx-exec.ts +16 -74
- package/src/bin/smoo-nx-failures.ts +4 -0
- package/src/bin/smoo-nx-nextest-extract.ts +3 -2
- package/src/bin/smoo-nx-test-walks.ts +42 -0
- package/src/bounded-test-policy.test.ts +268 -21
- package/src/bounded-test-policy.ts +114 -42
- package/src/build-state-output-policy.test.ts +115 -0
- package/src/build-state-output-policy.ts +90 -0
- package/src/bun-bootstrap.test.ts +57 -3
- package/src/bun-file-packages.test.ts +47 -0
- package/src/cargo-closure-input.test.ts +138 -66
- package/src/cargo-closure-input.ts +17 -15
- package/src/cargo-dev-profile.ts +137 -0
- package/src/cargo-resolve-cache.test.ts +70 -1
- package/src/cargo-source-hash.test.ts +193 -97
- package/src/cargo-source-hash.ts +345 -68
- package/src/cargo-toolchain-policy.ts +6 -0
- package/src/cargo-workspace.test.ts +320 -0
- package/src/cargo-workspace.ts +336 -10
- package/src/cross-check-policy.test.ts +59 -1
- package/src/cross-check-policy.ts +25 -6
- package/src/disk-command-policy.test.ts +61 -0
- package/src/disk-command-policy.ts +107 -0
- package/src/ensure-built-receipt.test.ts +184 -0
- package/src/ensure-built-receipt.ts +263 -0
- package/src/ensure-built-witness.test.ts +344 -0
- package/src/ensure-built-witness.ts +252 -0
- package/src/ensure-built.test.ts +1036 -578
- package/src/ensure-built.ts +820 -224
- package/src/eslint-file-set.ts +9 -0
- package/src/executors/bounded-exec/cpu-pool.test.ts +454 -0
- package/src/executors/bounded-exec/cpu-pool.ts +417 -0
- package/src/executors/bounded-exec/cpu-tokens.test.ts +269 -0
- package/src/executors/bounded-exec/cpu-tokens.ts +177 -0
- package/src/executors/bounded-exec/disk-lease.test.ts +185 -0
- package/src/executors/bounded-exec/disk-lease.ts +139 -0
- package/src/executors/bounded-exec/executor-host.test.ts +778 -0
- package/src/executors/bounded-exec/executor.test.ts +1116 -63
- package/src/executors/bounded-exec/executor.ts +565 -67
- package/src/executors/bounded-exec/gateway-lease.ts +176 -0
- package/src/executors/bounded-exec/hang-diagnostics.test.ts +285 -0
- package/src/executors/bounded-exec/hang-diagnostics.ts +283 -0
- package/src/executors/bounded-exec/progress.test.ts +79 -0
- package/src/executors/bounded-exec/progress.ts +118 -0
- package/src/executors/bounded-exec/ram-temp-fixtures.ts +12 -0
- package/src/executors/bounded-exec/ram-temp-held.test.ts +264 -0
- package/src/executors/bounded-exec/ram-temp-held.ts +242 -0
- package/src/executors/bounded-exec/ram-temp-host.test.ts +605 -0
- package/src/executors/bounded-exec/ram-temp.test.ts +674 -0
- package/src/executors/bounded-exec/ram-temp.ts +1918 -0
- package/src/executors/bounded-exec/schema.json +11 -5
- package/src/executors/bounded-exec/schema.ts +27 -11
- package/src/executors/bounded-exec/verdict.test.ts +440 -0
- package/src/executors/bounded-exec/verdict.ts +433 -0
- package/src/executors/shard-timings/executor.ts +36 -0
- package/src/executors/shard-timings/schema.json +21 -0
- package/src/executors/shard-timings/schema.ts +4 -0
- package/src/failure-summary.ts +119 -0
- package/src/generators/bounded-test-targets/generator.test.ts +4 -4
- package/src/generators/bun-test-tracing/generator.test.ts +2 -2
- package/src/generators/create-package/generator.test.ts +6 -0
- package/src/generators/create-package/generator.ts +6 -2
- package/src/host-controller-wiring.test.ts +31 -0
- package/src/index.test.ts +909 -295
- package/src/index.ts +441 -143
- package/src/linked-snapshots.nx.test.ts +172 -0
- package/src/managed-files/ci-workflow.ts +54 -13
- package/src/managed-files/files.ts +65 -9
- package/src/managed-files/forgejo-artifact.test.ts +395 -0
- package/src/managed-files/generator.test.ts +60 -1
- package/src/managed-files/json-edit.test.ts +216 -0
- package/src/managed-files/json-edit.ts +290 -0
- package/src/managed-files/managed-format-json.test.ts +308 -0
- package/src/managed-files/managed-format.ts +90 -2
- package/src/managed-files/pr-preview-cleanup-workflow.ts +69 -3
- package/src/managed-files/publish-workflow.ts +285 -67
- package/src/managed-files/repo-path.test.ts +129 -0
- package/src/managed-files/trace-db-artifact.test.ts +7 -0
- package/src/managed-files/tree.ts +11 -3
- package/src/manifest-hash.ts +2 -6
- package/src/nextest-code-bound.test.ts +155 -0
- package/src/nextest-code-bound.ts +195 -0
- package/src/nextest-extraction.test.ts +37 -5
- package/src/nextest-extraction.ts +51 -9
- package/src/nextest-partitions.test.ts +427 -82
- package/src/nx-cache-bound.test.ts +5 -10
- package/src/nx-cache-restore-mtime.test.ts +170 -0
- package/src/nx-daemon-claim.test.ts +277 -0
- package/src/nx-daemon-idle-cpu.test.ts +81 -0
- package/src/nx-daemon-idle-policy.test.ts +51 -0
- package/src/nx-daemon-ignore-files.test.ts +210 -0
- package/src/nx-daemon-root-normalization.test.ts +105 -0
- package/src/nx-daemon-validity.test.ts +925 -0
- package/src/nx-dot-input-policy.test.ts +353 -0
- package/src/nx-failure-life-cycles.test.ts +207 -0
- package/src/nx-files-archive-racy.test.ts +97 -0
- package/src/nx-foreign-workspace-env.test.ts +329 -0
- package/src/nx-incomplete-run-summary.test.ts +70 -0
- package/src/nx-plugin-load-errors.test.ts +120 -0
- package/src/nx-project-resolution-cache.test.ts +176 -0
- package/src/nx-raw-inputs.test.ts +166 -0
- package/src/nx-resident-graph-workers.test.ts +7 -16
- package/src/nx-run-information.test.ts +118 -0
- package/src/nx-skipped-task-exit.test.ts +180 -0
- package/src/nx-state-location.test.ts +63 -0
- package/src/nx-store-resolution.test.ts +26 -25
- package/src/nx-task-graph-cycles.test.ts +148 -0
- package/src/nx-task-output-lease.test.ts +414 -0
- package/src/nx-use-local.test.ts +103 -0
- package/src/package-target-policy.ts +3 -6
- package/src/release-config-policy.ts +3 -6
- package/src/runtime-input.test.ts +215 -0
- package/src/runtime-input.ts +17 -0
- package/src/rust-tokens.ts +78 -0
- package/src/serial-group-wiring.test.ts +326 -0
- package/src/task-history-namespace.test.ts +87 -89
- package/src/task-results.test.ts +226 -0
- package/src/task-results.ts +75 -0
- package/src/test-program.test.ts +303 -0
- package/src/test-program.ts +341 -0
- package/src/test-timings.test.ts +240 -0
- package/src/test-timings.ts +311 -0
- package/src/test-walk-policy.test.ts +585 -0
- package/src/test-walk-policy.ts +731 -0
- package/src/testing.test.ts +509 -0
- package/src/testing.ts +922 -0
- package/src/type-guards.ts +4 -0
- package/src/typecheck-test-policy.test.ts +496 -11
- package/src/typecheck-test-policy.ts +137 -103
- package/src/workspace-config-policy.test.ts +212 -13
- package/src/workspace-config-policy.ts +101 -73
- package/src/workspace-index.ts +18 -0
- package/src/workspace-manifest.ts +41 -0
- package/test-timings.json +85 -30
- package/managed/raw/tooling/direnv/inherited-devenv.ts +0 -665
- package/managed/raw/tooling/direnv/shared-caches.sh +0 -53
package/README.md
CHANGED
|
@@ -2,6 +2,15 @@
|
|
|
2
2
|
|
|
3
3
|
Local Nx plugin for workspace-standard package setup and missing inferred targets.
|
|
4
4
|
|
|
5
|
+
Workspace policy preserves configured plugin entries other than the disabled `@nx/js/typescript`, including their
|
|
6
|
+
options, include/exclude filters and order. It adds the bare `@smoothbricks/nx-plugin` entry only when the plugin is
|
|
7
|
+
absent; repeated updates do not replace an existing configured entry or discard scoped instances. Malformed plugin
|
|
8
|
+
values are reported rather than replaced, and an unchanged Nx Tree configuration is not serialized again.
|
|
9
|
+
|
|
10
|
+
The public plugin entry loads its transformed bootstrap product synchronously, matching Nx's real `handleImport` require
|
|
11
|
+
path in Node and Bun. A missing `dist/index.js` refuses by name and gives the `ttsc` bootstrap command; it never returns
|
|
12
|
+
an empty graph or loads raw development source in place of transformed code.
|
|
13
|
+
|
|
5
14
|
## Target Ownership
|
|
6
15
|
|
|
7
16
|
Official `@nx/js/typescript` inference is disabled because it cannot run the workspace's source transformers. A package
|
|
@@ -13,6 +22,8 @@ with `tsconfig.lib.json` receives transformer-aware `tsc-js` and native `typeche
|
|
|
13
22
|
- `test-watch` from explicit `test` commands for Bun and Vitest packages
|
|
14
23
|
- Cargo workspace targets from a neighboring workspace-root `Cargo.toml`
|
|
15
24
|
- aggregate `build` and `lint` targets
|
|
25
|
+
- an uncached `shard-timings` check, which `lint` depends on, beside `bun test` targets sharded with `--shard` and
|
|
26
|
+
balanced with `--timings`
|
|
16
27
|
- an uncached `deploy` for a PRIVATE project with a wrangler manifest, running
|
|
17
28
|
`smoo wrangler deploy-stage --stage {args.stage}`
|
|
18
29
|
|
|
@@ -30,25 +41,85 @@ published package that really is deployed declares `deploy` itself. Whether CI d
|
|
|
30
41
|
question, answered only by the deploy tags the CLI reads (`stage-deploy-target` and its three siblings).
|
|
31
42
|
|
|
32
43
|
Lint commands are inferred per project. A workspace Biome configuration enables the project-wide Biome check; an ESLint
|
|
33
|
-
flat configuration enables ESLint
|
|
34
|
-
|
|
35
|
-
|
|
36
|
-
|
|
37
|
-
|
|
38
|
-
|
|
39
|
-
|
|
44
|
+
flat configuration enables ESLint for existing JavaScript/TypeScript files under that project's `src` and the files its
|
|
45
|
+
`tsconfig.test.json` declares, wherever they live. TypeScript resolves inherited `include`, `files` and `exclude`; the
|
|
46
|
+
same sources, the test declaration and its existing `extends` chain are explicit lint cache inputs. Both lists hold only
|
|
47
|
+
files in Nx's workspace file index (.gitignore and .nxignore applied), so a gitignored generated module that a build
|
|
48
|
+
left on disk is neither linted nor keyed, and the lint target is the same in every checkout of one tree. Rust-only
|
|
49
|
+
source trees keep their Cargo validation and manifest checks without invoking ESLint on nonexistent JavaScript inputs.
|
|
50
|
+
Test-config diagnostics retain TypeScript's parsed-command-line types through bootstrap emit; invalid declarations fail
|
|
51
|
+
with their diagnostic code and message, while an empty test program (TS18003) contributes no test files. Adding
|
|
52
|
+
JavaScript/TypeScript sources or changing the declared test program updates lint coverage automatically. The target's
|
|
53
|
+
`metadata.eslintFileSet` retains that exact command and workspace-relative file list for cache inspectors.
|
|
54
|
+
`@smoothbricks/nx-plugin/eslint-file-set` owns its type and generated validator; an inspector can resolve each file's
|
|
55
|
+
actual ESLint parser program instead of guessing from the project or target name.
|
|
56
|
+
|
|
57
|
+
Lint inputs include dependency production inputs because type-aware ESLint reads dependencies' declarations. This uses
|
|
58
|
+
`^production` unless the workspace opts into the versionless production input described below. Keep
|
|
59
|
+
`targetDefaults.lint` limited to shared cache policy or additive inputs such as `["...", "^production"]`. Smoo's
|
|
60
|
+
workspace policy removes static lint executors, commands, dependencies, outputs and replacement input lists because they
|
|
61
|
+
override source-aware inference for every project. Project-local `nx.targets.lint` declarations retain normal Nx
|
|
62
|
+
override precedence.
|
|
63
|
+
|
|
64
|
+
Smoo's workspace policy also sets `"cacheDirectory": ".nx/cache"` in the root `nx.json`, and `smoo monorepo check` fails
|
|
65
|
+
without it. With no `NX_*_DIRECTORY` variable set, as in a host shell, the pinned Nx otherwise moves the task cache and
|
|
66
|
+
its database to a per-user `~/.nx/<id>` keyed by the repository's remote, which every checkout of the repository shares.
|
|
67
|
+
A configured `cacheDirectory` keeps the cache in the checkout's `.nx/cache` and the task database in its
|
|
68
|
+
`.nx/workspace-data` (nx.json has no key for the database's directory). Cowshed links both into the checkout's build
|
|
69
|
+
volume, and `src/nx-state-location.test.ts` checks where this repository's host Nx resolves them.
|
|
40
70
|
|
|
41
71
|
The plugin's own `nx run nx-plugin:test` aggregates four bounded Bun targets. Each uses Bun's native
|
|
42
72
|
`--timings=test-timings.json --shard=N/4` partition to balance measured per-file work, while normal discovery still runs
|
|
43
73
|
every test file (including newly added files) exactly once. A failed shard fails the aggregate. The shard count and
|
|
44
|
-
standard 120-second target deadline stay unchanged.
|
|
45
|
-
|
|
46
|
-
|
|
74
|
+
standard 120-second target deadline stay unchanged. Bun cuts the path-sorted files into four contiguous runs and costs a
|
|
75
|
+
file the manifest does not name at the manifest's median, so files added without a refresh cluster into one shard and
|
|
76
|
+
can push it past its deadline. The inferred `shard-timings` target makes that loud instead: for any project whose
|
|
77
|
+
`bun test` targets carry both `--shard=N/M` and `--timings=<file>`, it replays Bun's discovery rules without running a
|
|
78
|
+
test and fails naming every file the shards would run that the manifest does not name, with the refresh command
|
|
79
|
+
`smoo nx refresh-test-timings <project>`. That command runs the suite once without `--shard`, adding `--update-timings`,
|
|
80
|
+
and then repeats the check. Refresh after adding or slowing a test file; the manifest is scheduling data, not a test
|
|
81
|
+
allowlist. The shards of one suite must share one cwd and every argument but `--shard`, pass flag values as
|
|
82
|
+
`--flag=value`, and use no shell syntax beyond quoting; a shard command that breaks this fails graph construction,
|
|
83
|
+
because the files it runs could only be guessed.
|
|
47
84
|
|
|
48
85
|
Real Nx fixtures own their cache and workspace-data directories and stop their daemon before removing the fixture.
|
|
49
|
-
|
|
50
|
-
|
|
51
|
-
|
|
86
|
+
Retirement also drains processes whose working directory is inside the fixture when no daemon record exists: startup may
|
|
87
|
+
not have published its record yet, so an absent record never proves the fixture is empty. Fixtures that intentionally
|
|
88
|
+
strip caller directory overrides declare `cacheDirectory: ".nx/cache"` in their own `nx.json`; this keeps the native
|
|
89
|
+
task database local too, rather than leaving a new `~/.nx/<repoKey>` behind. Offline, path-only Cargo fixtures likewise
|
|
90
|
+
own a temporary `CARGO_HOME`; production commands still keep the caller's Cargo home.
|
|
91
|
+
|
|
92
|
+
Fixture sockets and loopback listeners live where a Cowshed sandbox admits them, never at a literal `/tmp` path or on
|
|
93
|
+
port 0. A fixture's short socket parent alias lives beside the Nx socket directory the job supplies (the parent of
|
|
94
|
+
`NX_SOCKET_DIR`, the workspace's short runtime link), else under `COWSHED_RUNTIME_LINK`, else the temp directory. The
|
|
95
|
+
alias points into the fixture's own physical `.nx/sockets`, outside every checkout; its `nx` socket leaf is a real
|
|
96
|
+
directory. The foreign-workspace guard therefore sees the fixture as its owner, without a new grant or identity rule. A
|
|
97
|
+
shell's variable counts only when it belongs to this checkout, as `nx-socket-dir.sh` decides it; one that leads
|
|
98
|
+
elsewhere or cannot be looked at is passed over, and a refusal names it. The temp fallback admits a user-owned parent or
|
|
99
|
+
the root-owned public sticky directory with mode 1777, not an inaccessible sticky directory. A socket path over 95 UTF-8
|
|
100
|
+
bytes is refused with an error naming it. The shell owner retains a failed socket-directory creation's exit status; the
|
|
101
|
+
foreign-workspace provider refuses before running a target and preserves its complete diagnostic. `fixtureNxEnv` gives
|
|
102
|
+
each fixture its own `NX_SOCKET_DIR`; retirement stops its daemon before unlinking its verified parent alias and
|
|
103
|
+
removing the fixture root. A test whose subject is two daemons names a distinct real leaf for each. A listener another
|
|
104
|
+
process connects to (the counted Cargo broker, a remote-cache server) binds through `bindListener`: a port of the
|
|
105
|
+
workspace's `COWSHED_PORT_BASE`/`COWSHED_PORT_BLOCK_SIZE` block, lowest first, skipping only ports in use, and port 0
|
|
106
|
+
outside a block.
|
|
107
|
+
|
|
108
|
+
On macOS the working-directory holder census invokes `/usr/sbin/lsof` explicitly, independent of a test or devenv
|
|
109
|
+
`PATH`. A missing or inaccessible census tool fails cleanup with its command error; it never means no holders. Only
|
|
110
|
+
lsof's exit-1 process-disappearance race retains the partial inventory it produced.
|
|
111
|
+
|
|
112
|
+
The provider-real-Nx snapshot walk uses asynchronous children so its case timer can run while a producer or capture is
|
|
113
|
+
active. Its 60 s case guard covers one provider, daemon lifecycle and all provenance/refusal checks; it does not split
|
|
114
|
+
those checks into separately rebuilt fixtures.
|
|
115
|
+
|
|
116
|
+
Ignored-lockfile validity uses one daemon-successor walk: first refusal at the request boundary, then the successor's
|
|
117
|
+
request-free safety-poll invalidation. Both boundaries keep their original observations without a second fixture start.
|
|
118
|
+
|
|
119
|
+
Daemon-validity readiness waits for an actual successful PING under its existing 30 s guard, probing every 50 ms;
|
|
120
|
+
socket-directory events are not readiness because a bind/listen or closed-connection transition need not emit another
|
|
121
|
+
one. Expiry reports the expected and current daemon PID, observed process state and descendants, and the daemon log's
|
|
122
|
+
last 60 lines before fixture retirement removes them. The record-change and restart oracles remain event-driven.
|
|
52
123
|
|
|
53
124
|
## Cargo Workspace Layouts
|
|
54
125
|
|
|
@@ -56,20 +127,24 @@ The plugin discovers Cargo workspaces beside Nx project manifests at any depth:
|
|
|
56
127
|
|
|
57
128
|
- **Workspace owner:** the project beside a workspace `Cargo.toml` owns one `cargo-fetch`, `cargo-test-compile`,
|
|
58
129
|
`cargo-test-archive`, `cargo-lint`, cross-check, mutation, bench, and sweep target. Its `cargo-lint` and `cargo-test`
|
|
59
|
-
aggregates reach all workspace crates, including a root `[package]
|
|
130
|
+
aggregates reach all workspace crates, including a root `[package]`; `cargo-test` is the ordinary suite, and the
|
|
131
|
+
project's `test` also reaches the serial test groups (below).
|
|
60
132
|
- **Crate owner:** the deepest Nx project directory containing a crate owns `cargo-test-<crate>` targets. Its
|
|
61
133
|
`cargo-lint` delegates to the workspace owner's one verdict, and `lint` never depends on the workspace test suite.
|
|
62
134
|
Projects can own multiple crates. A nested independent workspace has its own prerequisites and does not inherit those
|
|
63
135
|
of an outer workspace.
|
|
64
136
|
- **Commands:** validation runs from the governing Cargo workspace directory. One `cargo fmt --all --check` and one
|
|
65
|
-
`cargo --frozen clippy --workspace --all-targets --
|
|
66
|
-
build of its whole closure, so a target directory per crate rebuilt every shared dependency once per
|
|
67
|
-
closures and 11 GiB on one repository — and no per-crate cache hit repaid it. Clippy writes the workspace's
|
|
68
|
-
`target/`, like every dev build: its check units carry their mode in the unit hash and never collide with build
|
|
69
|
-
and the host units a dev build compiled (build scripts and their host-only dependencies) are reused instead of
|
|
70
|
-
compiled a second time. Cargo's
|
|
71
|
-
|
|
72
|
-
|
|
137
|
+
`cargo --frozen clippy --workspace --all-targets --config 'build.warnings="deny"'` cover every member. Clippy of a
|
|
138
|
+
crate is a check build of its whole closure, so a target directory per crate rebuilt every shared dependency once per
|
|
139
|
+
crate — 49 closures and 11 GiB on one repository — and no per-crate cache hit repaid it. Clippy writes the workspace's
|
|
140
|
+
own `target/`, like every dev build: its check units carry their mode in the unit hash and never collide with build
|
|
141
|
+
units, and the host units a dev build compiled (build scripts and their host-only dependencies) are reused instead of
|
|
142
|
+
compiled a second time. Warnings are denied by Cargo's `build.warnings`, not `-- -D warnings`: arguments after `--`
|
|
143
|
+
are clippy-driver's `CLIPPY_ARGS`, which enter every member's fingerprint, so the gate and a plain
|
|
144
|
+
`cargo clippy --workspace --all-targets` would re-check each other's units. `build.warnings` also fails on warnings
|
|
145
|
+
replayed from Fresh units, so the two share one unit set and the gate still fails. Cargo's own lock serializes the
|
|
146
|
+
invocations that share the directory. `cargo-lint-cross` keeps `target/cargo-lint-cross`: a foreign triple's units are
|
|
147
|
+
distinct anyway, and its own lock keeps the cross check from queueing behind host builds.
|
|
73
148
|
- **Scheduling:** inference never throttles Nx. It sets no `parallelism: false` or `parallel: false`, no cap, and no
|
|
74
149
|
edge whose only purpose is keeping two cargo writers off one `target/`: Cargo owns that lock, and a second process
|
|
75
150
|
waits inside cargo. An edge states a real consumer — fetch before frozen cargo, the archive before the runners,
|
|
@@ -78,7 +153,9 @@ The plugin discovers Cargo workspaces beside Nx project manifests at any depth:
|
|
|
78
153
|
per-manifest `cargo fetch`, Biome and ESLint) stay in a `commands` list that Nx runs side by side.
|
|
79
154
|
- **Overrides:** normal Nx merging applies. Explicit `nx.targets` fields replace inferred fields, while omitted fields
|
|
80
155
|
retain their inferred base. Use `"dependsOn": ["...", "extra"]` to preserve inferred prerequisites when adding an
|
|
81
|
-
edge; replacing the array makes its author responsible for fetching before frozen Cargo commands
|
|
156
|
+
edge; replacing the array makes its author responsible for fetching before frozen Cargo commands and producing the
|
|
157
|
+
archive before any runner consumes it. This repository also declares the archive producer explicitly in both APFS
|
|
158
|
+
runners' owning project targets, so a cold checkout reaches it before extraction.
|
|
82
159
|
|
|
83
160
|
Member and exclude patterns may use `*` and `?` in any path segment (for example, `packages/*/crates/*`). Unsupported
|
|
84
161
|
patterns and member patterns that match no directory fail graph construction instead of silently omitting crates.
|
|
@@ -96,6 +173,13 @@ test binaries. A package-local target that builds a shipped artifact follows the
|
|
|
96
173
|
for an artifact family no other build compiles, such as a size-optimised wasm32 lane; a profile that compiles the same
|
|
97
174
|
crates for the same target as `dev` or `release` compiles every dependency a second time.
|
|
98
175
|
|
|
176
|
+
Rust LMAO's opt-in source-revision contract is defined in
|
|
177
|
+
[01f](../../specs/lmao/01f_arrow_table_structure.md#smoo/lmao!n/arrow-table-module-id.not-git). Its native and Wasm
|
|
178
|
+
targets declare `{ "env": "LMAO_GIT_REVISION" }` as an input. Default dev/test runs leave it unset; publishing jobs set
|
|
179
|
+
it to the dispatch commit, so those builds include the explicit revision in their key without making dev runs
|
|
180
|
+
commit-keyed. The `production` configuration selects the shipping profile; it does not discover a Git revision. The root
|
|
181
|
+
workspace's cached nextest archive hashes that variable too, because it contains the compiled LMAO test binaries.
|
|
182
|
+
|
|
99
183
|
### Cargo cache boundaries
|
|
100
184
|
|
|
101
185
|
`cargo-fetch` is uncached because the registry lives outside Nx outputs. `cargo-test-compile` is also uncached: it warms
|
|
@@ -103,12 +187,13 @@ Cargo's shared mutable build directory before the bounded runner, even if that d
|
|
|
103
187
|
successful run. Cargo's own incremental cache avoids recompiling unchanged code. Nx never restores or collects shared
|
|
104
188
|
Cargo build directories; lint and test **results**, and dedicated Wasm/N-API **outputs**, remain cacheable.
|
|
105
189
|
|
|
106
|
-
Validation inputs include the crate's files,
|
|
107
|
-
|
|
108
|
-
|
|
109
|
-
|
|
110
|
-
|
|
111
|
-
|
|
190
|
+
Validation inputs include the crate's files, every file its sources compile in through a literal `include_str!` or
|
|
191
|
+
`include_bytes!` path outside the crate's directory (a shipped skill or a shared corpus beside the crates), local
|
|
192
|
+
dependency closure, governing manifests and configuration, the declared toolchain pin, and the toolchain's own reported
|
|
193
|
+
versions. The pin is `devenv.lock`, hashed as an ordinary file input at both `{workspaceRoot}/devenv.lock` and
|
|
194
|
+
`{workspaceRoot}/tooling/direnv/devenv.lock` — devenv resolves it into the rustc, cargo, linker, C toolchain and SDK
|
|
195
|
+
every cargo command inherits, so bumping it invalidates every cached cargo target. Workspace test runners additionally
|
|
196
|
+
hash all workspace member manifests because those affect unified features, and the nextest executable/configuration.
|
|
112
197
|
|
|
113
198
|
Nothing machine-local takes part. The ambient cargo environment and the global `$CARGO_HOME/config.toml` are
|
|
114
199
|
deliberately not hashed: their values carry absolute store paths and per-checkout state directories, which would make
|
|
@@ -119,6 +204,38 @@ build-script reads or custom environment variables that Cargo metadata cannot na
|
|
|
119
204
|
A target whose inputs a repository DECLARES replaces the inferred list, pin included. Name the two `devenv.lock` entries
|
|
120
205
|
in that declaration, or in the named input it uses, or a toolchain bump will not invalidate it.
|
|
121
206
|
|
|
207
|
+
Every cached cargo target names the npm packages its command runs as `externalDependencies`: none for cargo itself, and
|
|
208
|
+
`@smoothbricks/nx-plugin` for a target that runs its executor or reads its installed files (the nextest tool config, the
|
|
209
|
+
archive extractor). Without that input Nx keys a task whose executor is not an `@nx/` one on every package in the
|
|
210
|
+
lockfile, so a lockfile-only commit re-ran every cargo verdict. Inside this repository the plugin is a workspace
|
|
211
|
+
project, which Nx refuses as an external dependency, so the list is empty here.
|
|
212
|
+
|
|
213
|
+
No input declaration removes a project's own configuration from a task's key. Nx 23.2.1 adds `ProjectConfiguration` to
|
|
214
|
+
every task (`hash_planner.rs` `gather_self_inputs`), and `hash_project_config.rs` hashes the executor, outputs, options,
|
|
215
|
+
configurations and parallelism of every target in the project, plus its tags and named inputs (not `inputs`,
|
|
216
|
+
`dependsOn`, `cache` or `//`). An edit to one target's command therefore re-keys every target of that project, including
|
|
217
|
+
each inferred `cargo-test-*` runner. Only the project a target lives in decides that sensitivity.
|
|
218
|
+
|
|
219
|
+
Workspace policy defaults non-production tasks to `["{projectRoot}/**/*", "!{projectRoot}/**/*.md"]`, without a forced
|
|
220
|
+
CI workflow input. It migrates the old CI-only `sharedGlobals` reference out of `default` and never re-injects that
|
|
221
|
+
definition. Custom non-CI named globals and explicit workflow-target inputs remain repository-owned and unchanged;
|
|
222
|
+
workflow checks can still name their own CI inputs.
|
|
223
|
+
|
|
224
|
+
Policy excludes incidental Markdown through `!{projectRoot}/**/*.md` in `default` and `production`. A task that consumes
|
|
225
|
+
Markdown declares it through an input set without a matching negative: Nx exclusions win over positive paths in the same
|
|
226
|
+
fileset, so a later explicit path cannot re-include it. Inferred Cargo package filesets remain broad for arbitrary
|
|
227
|
+
compiled data, headers, schemas and build-script reads; incidental Markdown beneath a crate can therefore still
|
|
228
|
+
invalidate those tasks conservatively. Nx 23.2.1 hashes a cached target that declares no `inputs` over every file under
|
|
229
|
+
its project root and ignores the `default` named input, so `smoo monorepo check` refuses such a target: declare its
|
|
230
|
+
inputs, or give its target name a `targetDefaults` entry with `"inputs": ["default", "^default"]`.
|
|
231
|
+
|
|
232
|
+
TypeScript toolchain inputs exclude Markdown beneath workspace-root `patches/`, but retain every other patch byte and
|
|
233
|
+
pin, including extensionless `PIN` and `VERSION` files. Native content-hash regressions distinguish those inputs from
|
|
234
|
+
README and unrelated-source edits. A task that consumes Markdown declares it through an input set without a matching
|
|
235
|
+
negative: Nx exclusions win over positive paths in the same fileset, so a later explicit path cannot re-include it.
|
|
236
|
+
Inferred Cargo package filesets remain broad for arbitrary compiled data, headers, schemas and build-script reads;
|
|
237
|
+
incidental Markdown beneath a crate can therefore still invalidate those tasks conservatively.
|
|
238
|
+
|
|
122
239
|
Cargo keeps the caller's `CARGO_HOME`. Moving configuration into an isolated home can change relative paths or lose
|
|
123
240
|
source replacement, credentials, and toolchain settings; forwarding it with `--config` changes precedence. Registry
|
|
124
241
|
access may consequently wait on Cargo's package-cache lock, which Cargo arbitrates itself; Nx adds no ordering for it.
|
|
@@ -137,6 +254,11 @@ atomic rename; failed resolutions are never cached. A corrupt cache is reported
|
|
|
137
254
|
during resolution are retried at most three times, then refused with the cause intact. A WASM-only Nx host cannot
|
|
138
255
|
provide the native lock and receives a typed refusal rather than an unserialized fallback.
|
|
139
256
|
|
|
257
|
+
A persisted warm cache is checked before waiting for the cold-fill lock. Each Cargo resolution child and the
|
|
258
|
+
asynchronous wait for a foreign resolution lock have a 300-second bound. A refusal names the child or lock, the
|
|
259
|
+
workspace and elapsed bound; a timed-out child and its owned process group are killed. These bounds do not change the
|
|
260
|
+
native synchronous lock acquisition or Cargo's own package-cache arbitration.
|
|
261
|
+
|
|
140
262
|
A plugin worker's normal exit, host disconnect, or handled termination kills Cargo children it still owns; on POSIX it
|
|
141
263
|
kills the whole process group, including wrapper descendants. Standalone callers also forward unhandled termination
|
|
142
264
|
signals to their owned groups. An uncatchable `SIGKILL` cannot run an exit hook; the kernel still releases its file
|
|
@@ -145,11 +267,67 @@ lock.
|
|
|
145
267
|
The `@smoothbricks/nx-plugin:cargo-resolve` diagnostics channel reports `{ phase: "join", manifest }` when a caller
|
|
146
268
|
joins an existing flight. With no subscribers, inference neither allocates nor publishes an event. The cache regressions
|
|
147
269
|
use this readiness signal and Cargo control-channel events to prove concurrency and child teardown without sleep-based
|
|
148
|
-
timing assumptions; their outer hang guards report still-open test spans and process/gate state.
|
|
270
|
+
timing assumptions; their outer hang guards report still-open test spans and process/gate state. The counted Cargo
|
|
271
|
+
fixture delegates to the command selected before its shim was installed, with that original search path. An installed
|
|
272
|
+
jobserver launcher remains in the chain; it must not rediscover the counted shim and recursively enter the broker. The
|
|
273
|
+
real-producer regression covers both counted metadata and forwarded workspace discovery, with one metadata child.
|
|
274
|
+
|
|
275
|
+
Keep workspace-owned runtime state outside Nx's source index. The root `.nxignore` ignores every dot path with `.*` and
|
|
276
|
+
re-includes each tracked dot input by its dot-named prefix, such as `!/.github` or `!/packages/lmao/.cargo`: a dot-named
|
|
277
|
+
directory must be re-included to reach its tracked children. Nx's watcher reads only the root `.gitignore` and
|
|
278
|
+
`.nxignore`, not nested ignore files or Git's `.git/info/exclude`. `tooling:lint-dot-inputs` is an additive lint
|
|
279
|
+
dependency, keyed on the tracked file inventory as well as ignore-file contents. It fails when a tracked dot path lacks
|
|
280
|
+
its `!/` line or Nx's ignore checker excludes it, so a newly tracked dot input cannot silently disappear into `.*`.
|
|
281
|
+
`nx-dot-input-policy.test.ts` guards native index and file-hash parity for existing tracked inputs, unchanged project
|
|
282
|
+
graphs across the policy cutover, and actual daemon graph-pass silence for `.icm*` stores and hidden Bun staging. The
|
|
283
|
+
`.nxignore` file itself intentionally has new bytes. A tracked dot edit still reaches the daemon's watcher and graph
|
|
284
|
+
recomputation. Watching daemon logs, plugin sockets or job records instead turns each graph computation into another
|
|
285
|
+
file event and another graph computation.
|
|
286
|
+
|
|
287
|
+
### Runtime inputs
|
|
288
|
+
|
|
289
|
+
Every runtime input, inferred or declared, starts with `sh node_modules/@smoothbricks/nx-plugin/runtime-input.sh`.
|
|
290
|
+
Command-specific environment assignments follow the wrapper as `env NAME=value command ...`, so setup failures and
|
|
291
|
+
command failures use the same fail-closed path. For example:
|
|
292
|
+
|
|
293
|
+
```json
|
|
294
|
+
{ "runtime": "sh node_modules/@smoothbricks/nx-plugin/runtime-input.sh bun --version" }
|
|
295
|
+
```
|
|
296
|
+
|
|
297
|
+
Nx 23.2.1 hashes a runtime input's stdout and stderr and ignores its exit status, so a failed input becomes a valid key:
|
|
298
|
+
its error text, one string for every history, toolchain or file it could not read, and a result cached under it is
|
|
299
|
+
replayed for all of them. The script keys on the command's stdout alone and turns a nonzero exit into an Nx error: it
|
|
300
|
+
ends stdout with the byte 0xFF, and Nx refuses output that is not UTF-8
|
|
301
|
+
(`invalid utf-8 sequence of 1 bytes from index N`, exit 1). Nx drops a runtime input's stderr, so a failure that only
|
|
302
|
+
wrote to stderr would be invisible. The wrapper retains each failure under the checkout's
|
|
303
|
+
`.runtime-input-failures/failure.XXXXXXXXXX/`, outside Nx's state and the temporary directory. These private records
|
|
304
|
+
survive `nx reset`, task-cache cleanup and temporary-directory retirement; they remain until the operator removes them.
|
|
305
|
+
The reliable diagnostic after a daemon or headless gate's decoder error is:
|
|
306
|
+
|
|
307
|
+
```sh
|
|
308
|
+
cat .runtime-input-failures/*/report
|
|
309
|
+
```
|
|
149
310
|
|
|
150
|
-
|
|
151
|
-
|
|
152
|
-
|
|
311
|
+
Each readable `report` names the command, its indexed arguments, working directory, exit status and original stderr.
|
|
312
|
+
`argv` retains the exact NUL-delimited argument vector (including empty arguments), `stderr` retains the original bytes
|
|
313
|
+
(including trailing newlines), and `exit-status` retains the decimal status. The wrapper also attempts a stderr summary
|
|
314
|
+
naming the command and report path, and displays the report on `/dev/tty` when it can open a controlling terminal.
|
|
315
|
+
Neither display is guaranteed through Nx's daemon or a headless harness: rely on the retained report, not on Nx
|
|
316
|
+
forwarding stderr. A failed terminal write is retained in the record's `terminal-error`. Records can contain arguments
|
|
317
|
+
or stderr with credentials; keep them private and inspect before sharing. Evidence-storage errors name the problem and
|
|
318
|
+
the surviving stderr capture on stderr; they never remove the refusal byte or replace the command's status.
|
|
319
|
+
|
|
320
|
+
Successful commands still emit only their original stdout bytes, discard stderr and retain no failure record: no
|
|
321
|
+
evidence path, timestamp or diagnostic enters a successful key. `runtime-input.test.ts` compares that byte stream and
|
|
322
|
+
its hash with the previous success path, and plants a failure through real Nx (daemon and daemonless) to prove both
|
|
323
|
+
refusal and retained evidence.
|
|
324
|
+
|
|
325
|
+
An input that runs inside a cowshed workspace's sandboxed daemon gets the client's environment, a host shell's `HOME`
|
|
326
|
+
included, but not the client's filesystem view (cowshed `04_sandbox.md`). An input that reads the operator's home fails
|
|
327
|
+
there, so a runtime input reads only the checkout and its declared environment. A `git` command in one sets
|
|
328
|
+
`GIT_CONFIG_GLOBAL=/dev/null GIT_CONFIG_NOSYSTEM=1`: the key is a function of the repository, not of the operator's git
|
|
329
|
+
configuration. Git reads the global ignore file (`~/.config/git/ignore`) even then, so `smoo-nx-cargo-hash`, whose git
|
|
330
|
+
queries decide which package sources the key covers, also passes `-c core.excludesFile=/dev/null`.
|
|
153
331
|
|
|
154
332
|
### Cargo source runtime inputs
|
|
155
333
|
|
|
@@ -158,7 +336,7 @@ For inferred Cargo targets in a repository-root Cargo workspace, declare the ext
|
|
|
158
336
|
```json
|
|
159
337
|
{
|
|
160
338
|
"namedInputs": {
|
|
161
|
-
"externalRustCrates": [{ "runtime": "smoo-nx-cargo-hash" }]
|
|
339
|
+
"externalRustCrates": [{ "runtime": "sh node_modules/@smoothbricks/nx-plugin/runtime-input.sh smoo-nx-cargo-hash" }]
|
|
162
340
|
}
|
|
163
341
|
}
|
|
164
342
|
```
|
|
@@ -174,6 +352,14 @@ both the Nx and Cargo workspaces are left to the plugin's inferred file inputs.
|
|
|
174
352
|
require maintaining a second list of source roots. The command refuses missing dependencies or an unavailable locked
|
|
175
353
|
dependency cache instead of emitting a partial digest.
|
|
176
354
|
|
|
355
|
+
Each Cargo child and each acquisition of the cross-process resolve lock has a five-minute `CARGO_RESOLVE_BOUND_MS`
|
|
356
|
+
deadline. A warm content-keyed cache read takes no lock and spawns no child. A cold flight acquires the pinned Nx
|
|
357
|
+
binding's blocking kernel lock in one owned child; its parent kills and reaps that child if acquisition exceeds the
|
|
358
|
+
deadline. The named `CargoResolveLockError` reports the lock path, elapsed wait, bound and waiter pid (the holder pid is
|
|
359
|
+
unknown). A successful flight holds the lease until closing the child's stdin releases the lock and the child is reaped.
|
|
360
|
+
A timed-out Cargo command is killed, and resolution refuses instead of hanging graph construction or returning a stale
|
|
361
|
+
closure.
|
|
362
|
+
|
|
177
363
|
### Cargo closure input
|
|
178
364
|
|
|
179
365
|
Custom Cargo targets that replace inferred inputs need the source closure of the crates they build. The plugin infers it
|
|
@@ -202,21 +388,46 @@ the other directories the hash command skips), target sources outside it, its go
|
|
|
202
388
|
`Cargo.lock`. Nx hashes those from the file hashes it already keeps, so hashing these inputs starts no process. A graph
|
|
203
389
|
with unchanged resolution inputs starts no Cargo process either.
|
|
204
390
|
|
|
391
|
+
Literal `include_str!` and `include_bytes!` files also belong to the closure, including data inside a crate and tracked
|
|
392
|
+
files outside its directory (such as a shipped Markdown skill). One Rust lexer excludes comments and example strings;
|
|
393
|
+
computed paths and arbitrary build-script reads still need explicit inputs. External path-package digests include the
|
|
394
|
+
same literal data bytes. Inferred `napi-*` and repository-root `cargo-wasm` builds use this resolved closure rather than
|
|
395
|
+
every Rust file in the workspace, while retaining their toolchain pins, wrapper scripts, N-API configuration and CLI
|
|
396
|
+
version.
|
|
397
|
+
|
|
398
|
+
Runtime file identities are rooted in Cargo package name/version and the declared input role, not in the consuming
|
|
399
|
+
checkout's absolute path or depth. Source paths are package-relative; compiled-in data keeps its declaring source and
|
|
400
|
+
literal path as its identity. Identical main/shed inputs therefore share a key, while changed bytes, module names and
|
|
401
|
+
symlink targets still move it. Shared bytes are read once, and all identities are sorted before hashing. The Cargo
|
|
402
|
+
resolution memo schema retains package names/versions; refreshing that memo does not reset the Nx task database.
|
|
403
|
+
|
|
205
404
|
Members the file index cannot hold, outside the Nx workspace or under `node_modules`, are hashed by one runtime
|
|
206
405
|
`smoo-nx-cargo-hash --closure <projectRoot> <Cargo.toml>` entry, present only while such members exist. Its cached
|
|
207
406
|
resolution checks external manifests too, so a dependency introduced by an external manifest edit is still covered.
|
|
208
407
|
|
|
408
|
+
A runtime input that cannot be computed, whether from a manifest or locked dependency it cannot resolve or from
|
|
409
|
+
arguments it refuses, prints one fixed line (`cargo-input-unavailable`, `cargo-input-invalid-arguments`) and exits
|
|
410
|
+
nonzero, and `runtime-input.sh` turns that into an Nx error. Its stderr names the cause, Cargo's own message included,
|
|
411
|
+
with the checkout's absolute path written as a relative one and Cargo's timing-dependent lock-wait lines dropped;
|
|
412
|
+
refused arguments write nothing. The Cargo target, run by hand, still refuses the missing dependency or manifest with
|
|
413
|
+
Cargo's own cause. Only stdout supplies the cache-key result. Stderr retains the operational cause and may also contain
|
|
414
|
+
native-loader warnings; those diagnostics are not required to be byte-identical across checkouts.
|
|
415
|
+
|
|
209
416
|
Cargo resolution failures (no `cargo` on PATH, a stale `Cargo.lock`, or an unavailable locked dependency) fail graph
|
|
210
417
|
inference with `CargoMetadataError`, including the manifest path and Cargo's cause. A member inside the workspace
|
|
211
418
|
missing from Nx's file index, or another closure Nx cannot express precisely, fails with `CargoClosureInputError`.
|
|
212
419
|
Neither condition falls back to a whole-workspace runtime hash or silently reuses an older closure.
|
|
213
420
|
|
|
214
421
|
Resolution therefore needs the packages `Cargo.lock` pins already in `CARGO_HOME`, before any task can run: the graph's
|
|
215
|
-
own `cargo-fetch` target cannot supply them. The managed shell
|
|
216
|
-
`
|
|
217
|
-
`Cargo.toml
|
|
218
|
-
|
|
219
|
-
|
|
422
|
+
own `cargo-fetch` target cannot supply them. The managed shell's devenv task `smoo:cargo-fetch`
|
|
423
|
+
(`tooling/direnv/setup-environment.ts --cargo`, gated by its `--cargo --check` status) runs `cargo fetch --locked` for
|
|
424
|
+
every project whose `Cargo.toml` declares `[workspace]`, once per change to that workspace's `Cargo.toml`, `Cargo.lock`
|
|
425
|
+
or `.cargo/config[.toml]`. Its stamp lives in `$XDG_CACHE_HOME/smoo/cargo-fetched` (`~/.cache/smoo` when that is unset),
|
|
426
|
+
never in `CARGO_HOME`: a sandboxed shell may write only Cargo's own paths there. A stamp outside `CARGO_HOME` outlives
|
|
427
|
+
it, so the stamp also names the directory `CARGO_HOME` was when the fetch finished (device, inode and creation time): a
|
|
428
|
+
deleted and recreated `CARGO_HOME`, or another one, fetches again, while a copied checkout sharing the `CARGO_HOME` does
|
|
429
|
+
not. A failed fetch writes no stamp. In CI (`devenv tasks run`) a failed fetch fails; locally the task prints Cargo's
|
|
430
|
+
cause, the shell loads, and the next entry retries, while graph inference keeps refusing with that same cause.
|
|
220
431
|
|
|
221
432
|
In-workspace members follow Nx's file semantics, which differ from the command's own walk in two places: a symlinked
|
|
222
433
|
directory inside a package is not followed, and manifests and Cargo configuration above the Nx root (a
|
|
@@ -231,7 +442,9 @@ project declares its own named input and names its directory:
|
|
|
231
442
|
{
|
|
232
443
|
"namedInputs": {
|
|
233
444
|
"cargoSources": [
|
|
234
|
-
{
|
|
445
|
+
{
|
|
446
|
+
"runtime": "sh node_modules/@smoothbricks/nx-plugin/runtime-input.sh smoo-nx-cargo-hash --include-workspace --closure packages/example"
|
|
447
|
+
},
|
|
235
448
|
"{workspaceRoot}/Cargo.lock",
|
|
236
449
|
"{workspaceRoot}/devenv.lock",
|
|
237
450
|
"{workspaceRoot}/tooling/direnv/devenv.lock"
|
|
@@ -270,6 +483,11 @@ them by field instead: a `json` input with `excludeFields` for `package.json#ver
|
|
|
270
483
|
this costs no process. Everything else still counts: a dependency range, an export map, and the lockfile's resolution
|
|
271
484
|
table.
|
|
272
485
|
|
|
486
|
+
The workspace root manifest also drops `scripts`, `smoo` and `nx` from these three targets' hash. No check reads them:
|
|
487
|
+
scripts are commands a developer runs, `smoo.developerLinks` names links whose linked bytes reach a hash only through
|
|
488
|
+
their own inputs, and the root project's `nx` configuration is hashed as that project's configuration. Hashing them made
|
|
489
|
+
a root `postinstall` or developer-link edit re-key every lint and typecheck in the workspace.
|
|
490
|
+
|
|
273
491
|
Crate manifests are the one kind Nx cannot hash by field, because TOML has no such input. The plugin hashes them itself,
|
|
274
492
|
in the process that builds the graph and is already reading them: a single forward scan over the bytes feeds the hasher
|
|
275
493
|
every run it keeps, dropping `[package].version` and the `[workspace.package].version` members inherit and nothing else
|
|
@@ -301,6 +519,76 @@ installed. Every fallback costs cache hits and never trades away invalidation, b
|
|
|
301
519
|
reporting a failing one — a fileset that excluded a manifest with no digest replacing it would serve stale results
|
|
302
520
|
silently.
|
|
303
521
|
|
|
522
|
+
## Cowshed Build-State Output Policy
|
|
523
|
+
|
|
524
|
+
Consuming repositories enforce build-volume safety in their Nx lint, using the graph that lint already resolved:
|
|
525
|
+
`refuseOutputsUnderBuildState(graph, buildState.paths, workspaceRoot)` from
|
|
526
|
+
`@smoothbricks/nx-plugin/build-state-output-policy`. Obtain `buildState` from `cowshed build-state --json` in that
|
|
527
|
+
checkout; no graph is constructed by cowshed job admission or doctor.
|
|
528
|
+
|
|
529
|
+
The validator uses Nx output interpolation for every target and configuration, including resolved project overrides. It
|
|
530
|
+
refuses outputs at, beneath, or above a contributed path, and globs whose static prefix can cover one. Absolute outputs
|
|
531
|
+
are resolved against the caller's workspace root. `BuildStateOutputError.findings` names each project, target,
|
|
532
|
+
configuration, output, and contributed path. Keep cacheable artifacts under source-tree paths such as `dist/` and
|
|
533
|
+
`.cache/nextest/`; restoring an output under `target/` or `.nx/` can replace a build-volume link with a real directory.
|
|
534
|
+
|
|
535
|
+
## Test Walk Policy
|
|
536
|
+
|
|
537
|
+
A test pays its setup once and asserts after each step; two cases that test one path are one test. Consuming
|
|
538
|
+
repositories enforce this in their root Nx lint with the `smoo-nx-test-walks` bin. The bin judges only pairs with a case
|
|
539
|
+
changed since the upstream merge base. Among those pairs it reports each case's nearest refusing partner and its nearest
|
|
540
|
+
advisory partner separately, so an advisory pair never hides a refusal. Existing pairs therefore need no allowance
|
|
541
|
+
table. `smoo-nx-test-walks --base` prints that base for the target's runtime input, because the verdict depends on it
|
|
542
|
+
and no file hash names it. `upstreamBase` picks the upstream in this order: `NX_BASE` (refused when it names no commit),
|
|
543
|
+
`refs/remotes/main/main`, the local main's upstream branch, `refs/remotes/origin/main`, then `main`.
|
|
544
|
+
`findDuplicateScenarios(root, files, changes)`, `formatDuplicateScenario`, `diffScope` and `upstreamBase` are exported
|
|
545
|
+
from `@smoothbricks/nx-plugin/test-walk-policy`, and git failures come back as `GitAnswer` values.
|
|
546
|
+
|
|
547
|
+
The policy reads bun:test `test`/`it` cases in test files and `#[test]`-family fns in any Rust file. A body that is one
|
|
548
|
+
wrapper call, `await withHarness(memoryBackend, async (h) => { … })` or `runtime().block_on(async { … })`, walks the
|
|
549
|
+
call itself as its first step, with its body replaced by `$body`. It then walks the wrapper's statements, so a different
|
|
550
|
+
backend or runtime still tells two cases apart. A case's identity is part of the comparison, and only cases of one
|
|
551
|
+
identity are compared. Identity covers `test.failing` in TypeScript, and `#[should_panic]`, `#[ignore]`, `#[cfg(…)]` and
|
|
552
|
+
`#[cfg_attr(…)]` in Rust. It also covers the `before*`/`after*` hooks and the declarations of the enclosing `describe`
|
|
553
|
+
chain.
|
|
554
|
+
|
|
555
|
+
Each top-level statement of a case body is normalized several ways:
|
|
556
|
+
|
|
557
|
+
- `exact` renames the names the case declares to their declaration index.
|
|
558
|
+
- `shape` also abstracts every literal.
|
|
559
|
+
- `skeleton` also collapses the declared names.
|
|
560
|
+
- `flow` is `exact` with the declared names collapsed.
|
|
561
|
+
|
|
562
|
+
Field, method, key, struct-literal key, tuple-index and path names are never renamed. A pure assertion reads only what
|
|
563
|
+
earlier steps built. It is one of:
|
|
564
|
+
|
|
565
|
+
- an `expect`/`assert…` call that calls only methods and runs no closure, constructor, rejection or throw matcher;
|
|
566
|
+
- an `if (…) throw` guard;
|
|
567
|
+
- an `assert*!` that borrows nothing mutably and calls no method on a `let mut` local.
|
|
568
|
+
|
|
569
|
+
Whether a method reads or acts is not statically decidable. A method that acts inside an assertion, such as
|
|
570
|
+
`expect(await queue.drain())`, is treated as a read, and review decides. `TEST-ONE-WALK` refuses two shapes with
|
|
571
|
+
`error[TEST-ONE-WALK]` (exit 1):
|
|
572
|
+
|
|
573
|
+
- **same-path:** the cases share at least two leading statements exactly, and one of them only purely asserts beyond
|
|
574
|
+
them. Move its assertions into the other walk at the statement the diagnostic names.
|
|
575
|
+
- **literal-twin:** two or more statements equal once literals are abstracted; identical bodies are reported as verbatim
|
|
576
|
+
repeats. Make one table or property over the values. Keep each original case's label in its table row and include it
|
|
577
|
+
in assertion diagnostics; when rows need different numbers of checks, store all expected values per row and assert
|
|
578
|
+
every one in the common walk.
|
|
579
|
+
|
|
580
|
+
It advises on a third with `warning[TEST-ONE-WALK]`, which never fails the run. The warning is for the writer and the
|
|
581
|
+
reviewer, who decide whether the variants share one walk:
|
|
582
|
+
|
|
583
|
+
- **same-scenario:** one case's skeleton is the other's head (at least one statement) and tail around exactly one
|
|
584
|
+
statement of its own. The tail is the action and its outcome: at least three statements equal in flow, with at least
|
|
585
|
+
one action among them. The injected step differs, for example which crash window or precondition. Extra trailing pure
|
|
586
|
+
assertions on either side are more checks on the same outcome. Its precision is too low to refuse. Walk the variants
|
|
587
|
+
over one fixture, or make one table over the injected step.
|
|
588
|
+
|
|
589
|
+
A separate case stays only for a distinct boundary or edge: its setup or path diverges, or it needs a state the walk's
|
|
590
|
+
earlier steps destroy. Each diagnostic names the file, both cases, the shared statement count and that remedy.
|
|
591
|
+
|
|
304
592
|
## Nx Target Naming
|
|
305
593
|
|
|
306
594
|
Target names are `{tool}-{output}` names. Use names like `tsc-js`, `tsdown-js`, and `cargo-wasm`; `build` and `lint` are
|
|
@@ -311,58 +599,117 @@ Concrete targets come from concrete files:
|
|
|
311
599
|
- `tsc-js` emits transformed JavaScript through a temporary JS-only `ttsc` config, restores execute bits on outputs
|
|
312
600
|
declared by `package.json#bin`, then emits declarations and declaration maps from the original project with native
|
|
313
601
|
`tsc`. Direct project references and every cached output lane are preserved.
|
|
314
|
-
- `typecheck` is inferred from `tsconfig.lib.json` and runs native `
|
|
315
|
-
|
|
316
|
-
|
|
602
|
+
- `typecheck` is inferred from `tsconfig.lib.json` and runs the native compiler `SMOO_TYPECHECK_TSC` names (built by
|
|
603
|
+
`devenv.smoo.nix` from a pinned TypeScript revision) as `-p tsconfig.lib.json --noEmit`, directly rather than through
|
|
604
|
+
`ttsc`. Transform plugins' diagnostics are enforced by `tsc-js`, which runs them.
|
|
605
|
+
- `typecheck-tests` is inferred from `tsconfig.test.json` and runs the same compiler on
|
|
606
|
+
`-p tsconfig.test.json --noEmit`. It first rebuilds the current package's `tsc-js` output (or its non-TypeScript
|
|
607
|
+
`build`) so self-imports resolve after `clean`. Transform calls in test files are checked when the test runner
|
|
608
|
+
transforms them.
|
|
317
609
|
- `typecheck-tests:watch` is inferred from `tsconfig.test.json` and runs the same typecheck in watch mode.
|
|
318
610
|
- `test-watch` is inferred when the package already defines an explicit Bun or Vitest `test` command. The plugin derives
|
|
319
611
|
the corresponding watch command and makes it depend on `typecheck-tests`.
|
|
320
612
|
- A workspace-root `Cargo.toml` provides `cargo-test`, `test`, `cargo-lint`, `mutation`, and `bench`.
|
|
321
613
|
`cargo-test-compile` warms one workspace `cargo test --no-run`. `cargo-test-archive` runs one
|
|
322
|
-
`cargo --frozen nextest archive --workspace` into
|
|
323
|
-
it produces a file rather than a mutable build tree.
|
|
324
|
-
|
|
325
|
-
|
|
326
|
-
|
|
327
|
-
|
|
614
|
+
`cargo --frozen nextest archive --workspace` into `.cache/nextest/archive.tar.zst` and is the only cached cargo BUILD:
|
|
615
|
+
it produces a file outside the build-volume links rather than a mutable build tree. Nx cache restoration must never
|
|
616
|
+
replace Cargo's `target` link. Its inputs, like every cargo-derived fileset, are anchored at `{workspaceRoot}`: Nx
|
|
617
|
+
resolves a `{projectRoot}` glob against only the files its project owns, and a nested package owns the crates under
|
|
618
|
+
it, so a root workspace's archive spelled that way hashes none of its members and replays stale binaries against
|
|
619
|
+
edited crates. Each member crate gets a cached `cargo-test-<package>` run (30s slow-test warning like
|
|
620
|
+
`bun test --timeout=30000`, no per-test kill) whose inputs include that crate and its path dependencies, and whose
|
|
621
|
+
command runs
|
|
328
622
|
`cargo --frozen nextest run --binaries-metadata … --cargo-metadata … --target-dir-remap … --workspace-remap . -E 'package(<crate>)'`
|
|
329
|
-
from the archive's one extraction. Runners therefore execute rather than build: they
|
|
623
|
+
from the archive's one extraction. Runners therefore execute rather than build: they compile nothing in cargo's build
|
|
330
624
|
tree, and fan out instead of chaining. They do not unpack the archive each: before the run, `smoo-nx-nextest-extract`
|
|
331
|
-
hashes it and extracts it into
|
|
625
|
+
hashes it and extracts it into `<target_directory>/nextest-extracted/<sha256>` unless a runner already did — one
|
|
332
626
|
extracts, the others wait — so a workspace with ~200 test binaries and ~25 crates unpacks them once per archive
|
|
333
627
|
instead of once per crate. The directory is keyed by the archive's bytes, so a rebuilt or cache-restored archive never
|
|
334
|
-
runs an older extraction's binaries, and publishing one removes the others
|
|
335
|
-
|
|
336
|
-
|
|
337
|
-
|
|
338
|
-
|
|
339
|
-
tree's
|
|
340
|
-
|
|
341
|
-
|
|
342
|
-
|
|
343
|
-
|
|
344
|
-
|
|
345
|
-
|
|
346
|
-
|
|
347
|
-
|
|
348
|
-
|
|
349
|
-
|
|
350
|
-
|
|
351
|
-
|
|
352
|
-
`
|
|
353
|
-
|
|
354
|
-
|
|
628
|
+
runs an older extraction's binaries, and publishing one removes the others. The target directory is the one
|
|
629
|
+
`cargo metadata` reports for the runner's Cargo workspace, so a `build.target-dir` setting moves the extractions with
|
|
630
|
+
the build tree. An extraction is per-checkout tool state like the rest of that tree, never an Nx output: it lives in
|
|
631
|
+
Cargo's build tree, which a checkout with a build volume keeps off its source tree, and it never sits under a declared
|
|
632
|
+
output such as `.cache/nextest`. `cargo clean` removes it. `--workspace-remap` is required — an archive records the
|
|
633
|
+
producing tree's absolute paths, and without it a restored archive hands tests another checkout's
|
|
634
|
+
`CARGO_MANIFEST_DIR`. With it the archive is relocatable, so one cache entry serves every checkout of the same commit;
|
|
635
|
+
nextest re-points `CARGO_BIN_EXE_<name>`/`NEXTEST_BIN_EXE_<name>` at the extracted binaries at runtime, but a test
|
|
636
|
+
that reads them through the compile-time `env!` macro keeps the producing tree's path. Per-crate runners accept an
|
|
637
|
+
empty nextest selection because a valid workspace member may have no tests and a hash partition may legitimately be
|
|
638
|
+
empty. A crate in the project that builds the debug cdylib runs after `napi-debug`, because its tests load that addon
|
|
639
|
+
from `target/debug`. A crate declaring `[package.metadata.smoothbricks.wasm-bindgen]` also receives the cacheable
|
|
640
|
+
`cargo-wasm` output target in its owning project.
|
|
641
|
+
- The plugin's own `nextest.toml` is passed as
|
|
642
|
+
`--tool-config-file "smoo:$PWD/<to workspace root>/node_modules/@smoothbricks/nx-plugin/nextest.toml"`, so it is a
|
|
643
|
+
layer BENEATH the repository's `<cargo-workspace>/.config/nextest.toml` rather than a replacement for it: a repository
|
|
644
|
+
can raise a timeout or declare `archive.include` for a cdylib or fixture the archived test binaries need, and its
|
|
645
|
+
settings win. That file is an input of both the archive and every runner, so changing it invalidates the verdicts it
|
|
646
|
+
governs. `$PWD` keeps the command text identical across checkouts while satisfying nextest's requirement that a tool
|
|
647
|
+
config path be absolute, and naming the file through the workspace root's `node_modules` link — never the plugin's
|
|
648
|
+
real location — keeps it identical for a plugin linked from a checkout outside the workspace, whatever depth each
|
|
649
|
+
workspace sits at; `--user-config-file none` still holds, because a developer's `~/.config/nextest` must not decide a
|
|
650
|
+
cached verdict. Runtime hash inputs use that same workspace-relative installed path, never the physical plugin
|
|
651
|
+
checkout. Both the runtime output and the owning target's command declaration therefore remain portable.
|
|
652
|
+
- A crate whose suite is better spread over several runners declares `[package.metadata.smoothbricks.test] shards = N`,
|
|
653
|
+
and gets `cargo-test-<package>-shard1..N`, each running `--partition hash:i/N` under the same progress bound. nextest
|
|
654
|
+
assigns a test to a shard by hashing its name, so the shards stay an exact partition of the crate as tests and test
|
|
655
|
+
binaries are added, and a stale `N` can only make a target slow — never drop a test. Omitting the key means one
|
|
656
|
+
target, as before.
|
|
355
657
|
- Tests that a `nextest.toml` override singles out run separately in `cargo-test-<package>-exceptions-shard1..N`, using
|
|
356
|
-
the same declared partition count. Ordinary shards run the complement
|
|
357
|
-
once. The plugin's own
|
|
358
|
-
tests; lifting them out keeps their
|
|
359
|
-
every target inside its
|
|
360
|
-
|
|
361
|
-
|
|
362
|
-
|
|
363
|
-
|
|
364
|
-
|
|
365
|
-
|
|
658
|
+
the same declared partition count. Ordinary shards run the complement of that class and of every declared serial group
|
|
659
|
+
(below), so the union still covers every test exactly once. The plugin's own override raises the `slow-timeout`
|
|
660
|
+
warning period for compile-fail tests; lifting them out keeps their long bodies from delaying an ordinary shard, and
|
|
661
|
+
partitioning the class keeps every target inside its progress bound without relaxing it. The remaining non-group
|
|
662
|
+
exceptional tests stay normally schedulable. Every piece runs `--no-fail-fast`, so a failing test does not stop the
|
|
663
|
+
rest of its piece. The exceptional filter is the union of the overrides in the plugin's own `nextest.toml`, so adding
|
|
664
|
+
one there is the whole change to that class. Serial groups are classified separately, by `test-group` overrides in the
|
|
665
|
+
workspace's `.config/nextest.toml`; the plugin's file declares none. Only sharded crates need these targets; an
|
|
666
|
+
unsharded crate already runs all its non-group tests in one process. Empty partitions pass, including platforms where
|
|
667
|
+
the tests are `cfg`-ed out. `nextest-partitions.test.ts` executes the inferred archive and shard commands over 29
|
|
668
|
+
tests and asserts that the ordinary shards ran exactly the 13 ordinary tests once and the exceptional shards exactly
|
|
669
|
+
the 12 exceptional ones. The suite builds one shared fixture and archive, then runs separate bounded cases through the
|
|
670
|
+
real default serial-group executor: a successful count and run with the group's probes exactly once each, three
|
|
671
|
+
refusals that run nothing and clear the task's previous verdict, and a failing run whose ordinary failures still run
|
|
672
|
+
the group's remaining probes. The shared fixture first produces one real passing verdict, before any case runs. Each
|
|
673
|
+
refusal copies that seed and removes its own marker log, so neither a timed-out success case nor a prior refusal that
|
|
674
|
+
wrongly ran can become another case's precondition. The success case still runs and verifies the real executor
|
|
675
|
+
independently. Drivers run the built production Node executor, supplied by the test target's build dependency, rather
|
|
676
|
+
than starting a native TypeScript compiler for every case inside the shard's progress bound. Each case settles its
|
|
677
|
+
owned processes before the next case starts; the fixture is removed only after the final settlement.
|
|
678
|
+
- A repository's serial test groups each run in one runner, `cargo-test-group-<group>`, owned by the cargo workspace
|
|
679
|
+
root beside `cargo-test-archive`. A group is declared once, in `<cargo-workspace>/.config/nextest.toml`: a
|
|
680
|
+
`[test-groups.<group>]` table with `max-threads = 1`, and default-profile overrides that name it with `test-group` and
|
|
681
|
+
`platform = 'cfg(target_os="macos")'`. The runner's filter is the union of those overrides, so the runner and the
|
|
682
|
+
per-crate exclusions cannot drift; every crate piece excludes it with `not (…)`, so each test is selected by exactly
|
|
683
|
+
one target. The runner is not inferred off macOS, where nextest applies the group nowhere and no piece excludes its
|
|
684
|
+
tests. It is a `bounded-exec` run of the group's selection under the same progress bound as every test target, and
|
|
685
|
+
asks for one CPU token, matching the group's one runnable test process instead of reserving idle host capacity. The
|
|
686
|
+
run carries `--no-fail-fast`: an ordinary test failure does not stop the group's remaining tests, and a failed verdict
|
|
687
|
+
lists the failed ones. It carries `--no-tests=fail`, so a filter that selects no test fails the group. Ignored tests
|
|
688
|
+
keep their explicit lane. A crate named `group-*` is refused, since its runner name would collide. A repository's
|
|
689
|
+
serial group may measure each test's code time against a class with a nextest run-wrapper from the plugin's source
|
|
690
|
+
(`src/nextest-code-bound.ts`), which warns of a test over its class and never fails or kills it; the group's target
|
|
691
|
+
hashes that file as an input, so changing it reruns the group (`specs/cowshed/08_testing.md`, "Integration tests (real
|
|
692
|
+
substrate)").
|
|
693
|
+
- `cargo-test` is the ordinary suite alone: it depends on the crate pieces and never on a serial group, so a run of it
|
|
694
|
+
never reaches a group's real-image tests. The inferred `test` of the cargo workspace root and of every project that
|
|
695
|
+
owns crates depends on `cargo-test` alone. Explicit repository config owns host coverage: this repository's root
|
|
696
|
+
`host-e2e` depends on both real-APFS groups, and `gate` depends on `host-e2e` plus every project's `lint`, `test` and
|
|
697
|
+
`build`. `nx run-many -t gate` therefore covers the host boundary without pulling it into ordinary `test`; each group
|
|
698
|
+
replays from cache unless its own inputs changed.
|
|
699
|
+
- A group's runner hashes only the crates its filter can select tests from, with everything they link, every member
|
|
700
|
+
manifest and the repository's nextest config, so a change to an unrelated crate leaves it cached. The filter is read
|
|
701
|
+
as nextest reads it (its precedence, blanks and predicate arguments; `filterScope`): a plain `package(<name>)` or
|
|
702
|
+
`package(=<name>)` bounds a clause to that package, `and`, `&` and `-` keep the narrower side or the left side, and
|
|
703
|
+
`or`, `|` and `+` take every clause. Anything that bounds nothing scopes the group to every member, as the archive is
|
|
704
|
+
hashed: a clause with no package, a glob, substring or regex name, `not`, `deps` or `rdeps`, an argument with a
|
|
705
|
+
backslash or a comma, a filter nextest refuses, or packages that exclude each other. A scope that names a package
|
|
706
|
+
which is not a member of the Cargo workspace is refused when the graph is built.
|
|
707
|
+
- Inference orders nothing between two serial groups: each is serial only within its own runner. A repository that needs
|
|
708
|
+
two never concurrent states the edge as explicit target config, `"dependsOn": ["...", "cargo-test-group-<other>"]`
|
|
709
|
+
under `nx.json` `targetDefaults` or the owning project's `nx.targets`, and Nx starts the later group once the earlier
|
|
710
|
+
one has finished. Explicit target declarations are not platform-conditional: dependency-only target nodes and their
|
|
711
|
+
ordering edges remain off macOS even though no executable group runner is inferred. The ordinary `test` and
|
|
712
|
+
`cargo-test` aggregates must still reach neither group.
|
|
366
713
|
- Canonical `napi` package metadata provides a host `cargo-napi` target and named release targets for each configured
|
|
367
714
|
triple. Linux `--use-napi-cross` targets compile C/C++ dependencies with Clang; the NAPI CLI supplies its downloaded
|
|
368
715
|
GNU sysroot and toolchain flags. This avoids the bundled GCC's unsupported diagnostics-color flag without disabling
|
|
@@ -428,15 +775,28 @@ the tasks it aggregates answer in its place, and any outputs it declares are ver
|
|
|
428
775
|
uncacheable task is a miss on every call, because only running it can say whether its side effects are current.
|
|
429
776
|
|
|
430
777
|
The probe and runner use the workspace's installed Nx and its normal cache configuration. A build performed by the Nx
|
|
431
|
-
CLI can satisfy the next wrapper invocation without re-executing tasks;
|
|
778
|
+
CLI can satisfy the next wrapper invocation without re-executing tasks; uncached current inputs go through Nx's runner.
|
|
432
779
|
Both invocations must use the same cache location. Relative `NX_CACHE_DIRECTORY` and `NX_WORKSPACE_DATA_DIRECTORY`
|
|
433
780
|
values resolve against the Nx workspace root, so they remain checkout-local when inherited by another shell. Absolute
|
|
434
781
|
paths remain caller-controlled, including deliberately shared CI locations; an absolute path inherited from another
|
|
435
782
|
checkout does not become local merely because the working directory changed.
|
|
436
783
|
|
|
437
|
-
Before hashing, it compares a native snapshot
|
|
438
|
-
|
|
439
|
-
|
|
784
|
+
Before hashing, it compares a fresh native workspace snapshot with the daemon's file table and sends writes the watcher
|
|
785
|
+
has not delivered. It can reuse the exact previously awaited daemon graph only through a witness in the same Nx
|
|
786
|
+
workspace-data directory: filesystem-to-daemon reconciliation, an awaited authoritative graph generation, and identical
|
|
787
|
+
pre/post native file hashes and membership establish the witness. Every new or changed node is rehashed through Nx,
|
|
788
|
+
using device, inode, mode, length and nanosecond modification/change times to reject the native archive's whole-second
|
|
789
|
+
mtime ambiguity. The witness also binds the environment and installed Nx/configured plugin package resolution and owned
|
|
790
|
+
dependency-module/file closure (physical targets, aliases, package manifests and public outputs), plus the
|
|
791
|
+
workspace-resolved TypeScript analyzer. A package-declared dist directory link is proved by its physical tree, including
|
|
792
|
+
every node and the link itself; unreadable required module bytes fail loudly. Unknown plugin input contracts, unresolved
|
|
793
|
+
required packages, directory links whose external graph inputs are unproved, missing or unreadable witnesses, and any
|
|
794
|
+
mismatch require the authoritative graph RPC. An RPC need not publish a replacement witness when its required pre/post
|
|
795
|
+
evidence is unavailable or changed. No cache-file age or existence decides currentness. Workspace file membership,
|
|
796
|
+
nx.json, project/package manifests and lockfiles have no output exclusions; output writes can therefore refresh this
|
|
797
|
+
conservative graph witness too. Only Nx's current task hashes and matching artifacts decide whether anything runs: an
|
|
798
|
+
unrelated edit does not force cached-log replay, and an actual input edit cannot hit an old source snapshot. Argument
|
|
799
|
+
normalization, per-task environments and plugin pre/post hooks still run.
|
|
440
800
|
|
|
441
801
|
The daemon's output records are lossy: they are held in memory, tracked per collapsed directory, and erased by writes it
|
|
442
802
|
processes more than two seconds late — including, on a busy daemon, a restore's own writes. When it cannot vouch for a
|
|
@@ -446,6 +806,45 @@ text (never followed), file permission bits and bytes. A tree the restore would
|
|
|
446
806
|
daemon's record is re-armed as Nx's runner does after a restore. A missing or empty artifact, any difference, a read
|
|
447
807
|
error, or a write observed during the comparison is a miss.
|
|
448
808
|
|
|
809
|
+
A successful byte comparison may retain a cache-bound node inventory in that same Nx workspace-data directory. It binds
|
|
810
|
+
the exact task hashes, declared outputs and artifact paths to every compared working-tree/cache node and both sets of
|
|
811
|
+
real parent directories. Reuse requires the same device, inode, type/permission mode, length and nanosecond mtime/ctime
|
|
812
|
+
for all those nodes. It is evidence of the same previously compared bytes, never source identity or an mtime freshness
|
|
813
|
+
rule. Missing/unreadable evidence or moved nodes require a full comparison; mutations to either the cache or working
|
|
814
|
+
tree, parent symlink replacement, and directory membership changes cannot borrow an old equality verdict.
|
|
815
|
+
|
|
816
|
+
`probeBuilt({ target, cwd })` is the distinct typed read-only provider operation. It returns `EnsureBuiltProbeResult`: a
|
|
817
|
+
`miss` with its `MissReason` without starting Nx's runner or writing producer outputs; the existing operational socket
|
|
818
|
+
failure; a loud `provider-proof` failure; or a `hit` with `EnsureBuiltReceipt`. `ensureBuilt` is build/stream-only and
|
|
819
|
+
its CLI callers retain the existing hit/built/failed result. With the daemon disabled, `probeBuilt` returns a no-daemon
|
|
820
|
+
miss instead of running the target. Both operations preserve Nx argument normalization, hooks, per-task environments and
|
|
821
|
+
local cache parity; the read-only operation additionally requires exact-cache artifact equality. A managed consumer
|
|
822
|
+
requires the installed module's `probeBuilt` function before making any producer call; a missing operation refuses
|
|
823
|
+
loudly without falling back to `ensureBuilt` or writing provider outputs.
|
|
824
|
+
|
|
825
|
+
A probe receipt names the real provider workspace, Git HEAD and HEAD's tree, every task id/current Nx hash in the
|
|
826
|
+
selected target's dependency graph, and an independently read census of every Nx-expanded public output. Files carry
|
|
827
|
+
SHA-256; symlinks carry their text and mode without following them; directories carry mode, ordered membership and a
|
|
828
|
+
child-census digest, including empty directories. Source/revision/output nodes must remain unchanged across capture.
|
|
829
|
+
Missing, unreadable or stale output evidence refuses, never starts a build and never signs daemon-state identity.
|
|
830
|
+
|
|
831
|
+
Git provenance covers Nx's planned FILE inputs and the owning project/dependency manifests and root graph configuration.
|
|
832
|
+
Dirty matched inputs (including new or index-only changes) refuse. The native inspector cannot prove deleted membership
|
|
833
|
+
or a custom hasher's opaque reads: a deleted provider source or a bounded runtime/env input therefore uses conservative
|
|
834
|
+
any-dirty-provider-source refusal, excluding public outputs rather than treating them as committed source. Runtime/env
|
|
835
|
+
values remain explicitly labelled `nx-task-hash-only`, not Git provenance. Only the existing toolchain generator's exact
|
|
836
|
+
version-only recipe with declared cwd, a declared toolchain FILE pin and proved installed-module closure is bounded;
|
|
837
|
+
arbitrary shell/file/runtime commands and unavailable custom/native input proofs categorically refuse receipts. A caller
|
|
838
|
+
probing multiple targets uses the existing single-target API and rejects differing provider HEAD/tree identities before
|
|
839
|
+
publishing its immutable snapshot. This producer evidence binds a managed linked-package snapshot to one current build;
|
|
840
|
+
Nx alone decides whether its current keys and cache/output bytes match. A caller from another workspace never lends its
|
|
841
|
+
socket, database or cache overrides to the provider. Inside a managed shell (`DEVENV_RUNTIME`), `ensureBuilt` asks the
|
|
842
|
+
packaged `nx-socket-dir.sh` owner for the provider's binding: the root's verified cowshed runtime link when present,
|
|
843
|
+
otherwise its canonical checkout spelling under that shell runtime. Outside a managed shell, Nx retains its normal
|
|
844
|
+
defaults. A socket-owner refusal returns `disposition: 'failed'` with `reason.kind: 'workspace-socket'`, the selected
|
|
845
|
+
`taskId`, `workspaceRoot` and operational `detail`, before any task or daemon client starts. The owner's actual exit or
|
|
846
|
+
signal is preserved; the wrapper must not report that the target ran.
|
|
847
|
+
|
|
449
848
|
```typescript
|
|
450
849
|
import { ensureBuilt } from '@smoothbricks/nx-plugin/ensure-built';
|
|
451
850
|
|
|
@@ -466,7 +865,8 @@ smoo-nx-exec my-cli:build -- ./packages/my-cli/dist/my-cli argument
|
|
|
466
865
|
|
|
467
866
|
It is deliberately not `nx run <target> && exec`:
|
|
468
867
|
|
|
469
|
-
-
|
|
868
|
+
- An unchanged witnessed graph avoids graph reconstruction/transfer; current task hashing, daemon file/output probes and
|
|
869
|
+
local cache reads remain. This is not a measured end-to-end latency guarantee and does not start an Nx CLI.
|
|
470
870
|
- A hit is silent. `nx run` replays the cached log and prints its `[local cache]` banner, which a CLI wrapper must not
|
|
471
871
|
print.
|
|
472
872
|
- It execs the binary, preserving its argv, exit status, and signals. `nx run` has no notion of handing the process over
|
|
@@ -496,20 +896,40 @@ stderr.
|
|
|
496
896
|
|
|
497
897
|
## Nx 23.2.1 Runtime Patch
|
|
498
898
|
|
|
499
|
-
The repository installs Nx 23.2.1
|
|
500
|
-
|
|
501
|
-
|
|
502
|
-
|
|
503
|
-
uncompressed tar's sha256, and verifies the served asset on every later run. A tarball, unlike a `patchedDependencies`
|
|
504
|
-
entry, stays in Bun's read-only global store (`~/.bun/install/cache/links`), so a sandbox can run Nx without being able
|
|
505
|
-
to write it.
|
|
899
|
+
The repository installs Nx 23.2.1 as an ordinary Bun `patchedDependencies` entry: the root `package.json` lists
|
|
900
|
+
`nx@23.2.1` and `@nx/js@23.2.1`, and applies [`patches/nx@23.2.1.patch`](../../patches/nx@23.2.1.patch) to the registry
|
|
901
|
+
`nx` package. Bun installs a patched package project-local under `node_modules/.bun/`, so the `nx` the repository runs
|
|
902
|
+
is the patched one.
|
|
506
903
|
|
|
507
904
|
`@nx/js` imports `nx` (`nx/release`, `nx/src/…`) without declaring it
|
|
508
|
-
([nrwl/nx#34087](https://github.com/nrwl/nx/issues/34087))
|
|
509
|
-
|
|
510
|
-
|
|
511
|
-
|
|
512
|
-
|
|
905
|
+
([nrwl/nx#34087](https://github.com/nrwl/nx/issues/34087)). `patches/@nx%2Fjs@23.2.1.patch` declares the peer, so Bun
|
|
906
|
+
binds the same pinned Nx through the package's normal dependency namespace.
|
|
907
|
+
|
|
908
|
+
Native APFS fixtures install this same pinned package and patch offline into their own checkout, using Bun's local
|
|
909
|
+
cache, and report each installation's elapsed time. They do not borrow another checkout's `node_modules`: a patched
|
|
910
|
+
package's dependency links can traverse lexical `.bun` symlinks outside an explicit sandbox read grant even when the
|
|
911
|
+
resolved payload is granted. The workspace's own read authority covers those local links; only shared dependency
|
|
912
|
+
payloads receive additional bounded store grants. Grants are derived from the fixture's installed package namespace, not
|
|
913
|
+
the repository's different private patched namespace. Installation runs once per suite template, which fixtures clone.
|
|
914
|
+
Installer-restoration witnesses clone that real installed product into `packages/nx` of a distinct private developer
|
|
915
|
+
checkout, outside every consumer `node_modules`. Its root declares `workspaces: ["packages/nx"]` and the dependency
|
|
916
|
+
`nx: "workspace:*"`, following the ordinary in-repo workspace binding pattern. Pinned offline Bun produces that root's
|
|
917
|
+
lock and production namespace; no hand-made self alias substitutes for the install. The unchanged Nx CLI's global
|
|
918
|
+
resolution must resolve `nx/bin/nx.js` from that producer-owned checkout, not from `/private/tmp`. Ordinary pinned Bun
|
|
919
|
+
first generates this package's own lock offline with the existing `--lockfile-only` producer, then installs its
|
|
920
|
+
manifest-owned production dependencies offline into its own `node_modules`, without changing its manifest or maintained
|
|
921
|
+
program bytes. The fixture records the actual install, lock and product digests. The process launcher keeps `PWD` equal
|
|
922
|
+
to its actual `current_dir`, as the supervisor launcher does; the fixture observes Bun's real root before installation
|
|
923
|
+
and requires the lock at that root. Production mode implies a frozen lock and does not write a missing one, so a
|
|
924
|
+
production-only run is not lock-production evidence. Private lexical dependency aliases therefore stay inside the owned
|
|
925
|
+
developer checkout; reached namespaces and canonical shared targets use the same single closure traversal, bounded by
|
|
926
|
+
the exact caller-owned developer checkout and the actual shared store. Only reached package and namespace paths become
|
|
927
|
+
grants, never the whole checkout or `.bun` root. Traversal deduplicates exact package identities, not read-grant
|
|
928
|
+
prefixes: a covered local `@nx` scope can still contain a binding to the pinned native platform package in the shared
|
|
929
|
+
store. A normal Bun store entry is never presented as a developer binding. Local file providers are checkout-relative
|
|
930
|
+
directories that travel with the clone.
|
|
931
|
+
Foreign-daemon retirement witnesses retain the daemon's own standalone root, state and short socket throughout the
|
|
932
|
+
landing; its actual produced record is only a nomination, and its owner performs cleanup after the identity proof.
|
|
513
933
|
|
|
514
934
|
The Nx patch repairs upstream Nx runtime behavior, separately from this plugin's workspace-owner checks:
|
|
515
935
|
|
|
@@ -517,10 +937,17 @@ The Nx patch repairs upstream Nx runtime behavior, separately from this plugin's
|
|
|
517
937
|
daemon can have frozen a different database namespace before a later client supplies workspace-data overrides. The
|
|
518
938
|
patch removes the obsolete history RPCs and retains native errors, including post-run failures. See
|
|
519
939
|
[upstream Nx #37268](https://github.com/nrwl/nx/pull/37268).
|
|
940
|
+
- **A raw-input query resolves only its requested task's plan.** Nx still builds the full dependency graph to plan
|
|
941
|
+
dependent outputs, but does not walk every producer's own inputs and then discard those results. The plural inspector
|
|
942
|
+
keeps its full-closure default. The installed-product regression compares all root input buckets and configurations
|
|
943
|
+
against closure inspection and rejects the old all-plan selection. See
|
|
944
|
+
[root-only raw-input inspection](../../patches/README.md#root-only-raw-input-inspection).
|
|
520
945
|
- **The default cache limit uses the cache's own filesystem.** It takes ten percent of `statfs(cacheDir)` capacity, or
|
|
521
946
|
its nearest existing ancestor when the directory has not been created. This avoids synchronously inventorying every
|
|
522
947
|
mounted disk. Explicit `NX_MAX_CACHE_SIZE` and `nx.json.maxCacheSize` retain their precedence; filesystem errors other
|
|
523
948
|
than a missing directory still fail the operation. See [upstream Nx #37269](https://github.com/nrwl/nx/pull/37269).
|
|
949
|
+
Cowshed exports an explicit caller-overridable default of configured build-volume capacity divided by ten for host
|
|
950
|
+
shells and jobs. The hosted setup action keeps its existing separate `20GB` ceiling.
|
|
524
951
|
- **The daemon keeps graph plugin workers running.** Nx stops an isolated plugin worker after the last phase it has
|
|
525
952
|
hooks for, which for a graph-only plugin is every graph; the daemon recomputes the graph on every tracked file change,
|
|
526
953
|
so each change spawned and loaded every graph plugin's worker again (median 199 ms against 63 ms from edit to graph in
|
|
@@ -528,21 +955,141 @@ The Nx patch repairs upstream Nx runtime behavior, separately from this plugin's
|
|
|
528
955
|
long as the daemon; one-shot clients and task-only plugins keep the eager shutdown. See
|
|
529
956
|
[upstream Nx #37271](https://github.com/nrwl/nx/pull/37271).
|
|
530
957
|
- **Nx resolves typescript and release version actions from the workspace.** Nx 23.2.1 loads both with a plain `require`
|
|
531
|
-
from its own location
|
|
532
|
-
|
|
533
|
-
|
|
534
|
-
|
|
535
|
-
|
|
536
|
-
|
|
537
|
-
|
|
538
|
-
|
|
539
|
-
|
|
540
|
-
|
|
541
|
-
|
|
542
|
-
|
|
543
|
-
|
|
544
|
-
|
|
545
|
-
|
|
958
|
+
from its own location. Where that location cannot see `typescript`, its dependency analysis skips every source import
|
|
959
|
+
without a warning (the first install that exposed this lost 12 of 44 graph edges), and `release version` cannot
|
|
960
|
+
resolve `@nx/js`'s version actions. The patch resolves both through `getNxRequirePaths` (the workspace first), falling
|
|
961
|
+
back to Nx's own location. See [upstream Nx #37272](https://github.com/nrwl/nx/pull/37272).
|
|
962
|
+
- **A task's dependencies do not depend on which other tasks the run asks for.** A target that depends on `^build`
|
|
963
|
+
through a project with no `build` target gets a dummy task for it, and Nx flattens each dummy into the real tasks
|
|
964
|
+
behind it except a dummy it believes sits in a cycle. `findCycles` reported every task on the depth-first path that
|
|
965
|
+
first reached a cycle, and only the cycles reached before another traversal marked their tasks visited, so the answer
|
|
966
|
+
followed the order of the graph's keys. In a workspace whose projects import each other, a shard requested alone
|
|
967
|
+
(`nx run app:test-4`) and the same shard requested through its aggregate (`nx run-many -t test`) got different
|
|
968
|
+
dependencies and, because `dependentTasksOutputFiles` keys on the dependency tasks' outputs, different hashes: a
|
|
969
|
+
pre-run of one never hit the cache of the other. The patch answers with the tasks that lie on a cycle (Tarjan's
|
|
970
|
+
strongly connected components), so a task's dependencies are a property of the graph behind it. Not yet proposed
|
|
971
|
+
upstream: Nx `master` carries the same `findCycles`.
|
|
972
|
+
- **A target's output has one owner across clients of the checkout.** A per-project/target lease spans cache lookup and
|
|
973
|
+
restore, execution, output publication, output-hash recording and cache capture. A producer replacing and retiring a
|
|
974
|
+
directory cannot invalidate a concurrent cache walk, and same-hash clients cannot remove one another's artifact while
|
|
975
|
+
publishing it. Waiting clients recheck the cache under ownership; independent targets retain concurrency. See
|
|
976
|
+
[`patches/README.md` §Cross-client task output ownership](../../patches/README.md#cross-client-task-output-ownership)
|
|
977
|
+
for the kernel lock, worker lifecycle and real-client regression.
|
|
978
|
+
- **A restored output is newer than the output it replaces.** Nx restores a cached output with `std::fs::copy`, which on
|
|
979
|
+
macOS clones the file and keeps its timestamps, so a restored output carried the mtime it had when its cache entry was
|
|
980
|
+
written. Build from input A, rebuild from B, return to A, and the restored A was older than the B output it replaced:
|
|
981
|
+
every consumer that judges freshness by mtime (a Cargo build script's `rerun-if-changed`, make, a file watcher) kept
|
|
982
|
+
what it built from B. The patch stamps each restored file with the time of the restore, as running the task would
|
|
983
|
+
have: after `copyFilesFromCache` for a local hit, and after the native `applyRemoteCacheResults` for a remote hit,
|
|
984
|
+
which untars the artifact with its stored mtimes and restores it with the same copy. Restoring one directory output of
|
|
985
|
+
10,000 1 KiB files took a median 3.4 s for the native remove and copy and 0.79 s more for the stamp, on a machine at
|
|
986
|
+
load average 70–120. Not yet proposed upstream.
|
|
987
|
+
- **A file rewritten within the second it was hashed in is hashed again.** Every Nx context (each run without the
|
|
988
|
+
daemon, and the daemon when it starts) hashes the workspace's files through `<workspace-data>/nx_files.nxt`, which on
|
|
989
|
+
Unix keeps each file's hash beside its mtime in whole seconds, and reuses an archived hash while that second is
|
|
990
|
+
unchanged; size and nanoseconds are not compared. A file written again within the second it was hashed in keeps its
|
|
991
|
+
mtime and hashes as the bytes it replaced: a revert replays the cache entry of the edit it reverts. The patch applies
|
|
992
|
+
Git's racy-index rule before the native context reads the archive: when the archive holds an mtime from ten seconds
|
|
993
|
+
before its own write (the longest a hashing pass is assumed to run) or later, it is removed and every file is hashed
|
|
994
|
+
again. A full pass over 6,370 files measured 288–753 ms at load average 70–110. The check reads each entry's mtime
|
|
995
|
+
from the archive's rkyv layout as Nx 23.2.1's binary writes it (the root's entry count and entry pointer in the last
|
|
996
|
+
12 bytes, 24-byte entries with the i64 mtime at +8), in under 1 ms for 6,396 files. In 134 synthetic workspaces of 1
|
|
997
|
+
to 400 files dated 2001 it kept every archive, and it removed every one once a file was written in the pass. An
|
|
998
|
+
archive whose root does not fit is removed. A file dated ahead of the archive keeps it distrusted, so every run hashes
|
|
999
|
+
everything until the clock passes that date. The check cannot see a pass longer than ten seconds, nor an archive
|
|
1000
|
+
another Nx process rewrites between the check and the native read; on Windows, whose native mtime is a FILETIME, it
|
|
1001
|
+
does nothing. The native fix records every entry dated from the second its pass started on as racy and has none of
|
|
1002
|
+
these limits. Not yet proposed upstream.
|
|
1003
|
+
- **A failed run exits 1, not 130, when it skips the tasks behind a failure.** Nx skips every task that depends on a
|
|
1004
|
+
failed one, and tells no life cycle about a skipped task, so the results `nx` reads its exit status from had none for
|
|
1005
|
+
it; `didCommandComplete` takes a discrete task without a result for one the run never finished, and `nx` exited 130,
|
|
1006
|
+
the status of an interrupted run. A gate then reported `Failed to run "nx" due to exit code 130` for an ordinary test
|
|
1007
|
+
failure. The patch has `didCommandComplete` also accept a task the runner reports skipped, so the exit status is 1.
|
|
1008
|
+
With `--nx-bail=true`, pending tasks have no status and active tasks can be stopped by bailout cleanup; the CLI uses
|
|
1009
|
+
the orchestrator's actual bailout and interruption flags, so a known failure still exits 1, but a signal or forced TUI
|
|
1010
|
+
stop wins and exits 130. The terminal output's "Tasks not run" list is unchanged. Only that check sees the runner's
|
|
1011
|
+
skipped tasks: `invokeTasksRunner` and the plugins' `postTasksExecution` hooks still get Nx's own results, with none
|
|
1012
|
+
for a skipped task, so a caller that replays Nx's skip rule over them counts it once. (An earlier version of this hunk
|
|
1013
|
+
added a skipped result to those results, and such a caller counted no task as skipped.) Not yet proposed upstream: Nx
|
|
1014
|
+
`master` reports skipped tasks the same way.
|
|
1015
|
+
- **A workspace has one daemon, and a second one leaves it alone.** Every daemon overwrote `server-process.json` with
|
|
1016
|
+
its own pid, so two clients that each found no daemon started two, and the later one displaced the earlier. A client
|
|
1017
|
+
of the displaced daemon lost its socket mid-request (`EPIPE` on `RECORD_OUTPUTS_HASH_BATCH`) after the run had printed
|
|
1018
|
+
its summary. Under a configured `NX_SOCKET_DIR`, every daemon binds the same path, and closing a listening Unix socket
|
|
1019
|
+
unlinks its path. The displaced daemon's shutdown therefore removed its successor's socket, the next client found no
|
|
1020
|
+
daemon and started another, and one workspace saw a new daemon every five seconds. Before it does anything else, the
|
|
1021
|
+
patched daemon creates the record with link(2), which never replaces a file. If a live daemon of the same Nx version
|
|
1022
|
+
that serves the same root holds the record, the new one waits for that daemon to answer and then exits. A record whose
|
|
1023
|
+
holder is dead, runs another Nx version, or serves another root (a copied workspace) is replaced. A daemon whose
|
|
1024
|
+
socket, at the path it bound, has been removed or rebound exits within 20 ms, since no client can reach it. It leaves
|
|
1025
|
+
a path another daemon has bound alone, so that daemon's socket is not unlinked. A holder starts when it claims, so a
|
|
1026
|
+
live one that has not answered by the time its record is a minute old (a pid reused after a crash) is stale and
|
|
1027
|
+
replaced. On a filesystem without hard links the record is created exclusively instead. Not yet proposed upstream.
|
|
1028
|
+
- **A run that throws part way through does not report success.** The default tasks runner ends its life cycle in a
|
|
1029
|
+
`finally`, so the static summary also prints when the orchestrator throws mid-run. It printed "Successfully ran"
|
|
1030
|
+
whenever no task had failed or stopped, and counted each task the run never reported as skipped: a focused `run-many`
|
|
1031
|
+
whose daemon socket died while recording output hashes printed success with "10 skipped", then `Error: write EPIPE`.
|
|
1032
|
+
The orchestrator reports every task it finishes except one it skips, and it skips a task only behind a failed or
|
|
1033
|
+
stopped one, so the patch prints success only when every task was reported. Otherwise the summary says the run did not
|
|
1034
|
+
complete and names the unreported tasks: under "Tasks not run" after a failure, as before, and under "Tasks the run
|
|
1035
|
+
ended without finishing" when none failed. Nx `master` prints the same summary. Not yet proposed upstream.
|
|
1036
|
+
- **Nx refuses an environment set up for another workspace.** A shell that moves into another checkout keeps the
|
|
1037
|
+
`NX_WORKSPACE_ROOT_PATH`, `NX_WORKSPACE_DATA_DIRECTORY`, `NX_CACHE_DIRECTORY` and `NX_SOCKET_DIR` of the workspace it
|
|
1038
|
+
was set up for, and Nx takes `NX_WORKSPACE_ROOT_PATH` verbatim as its root: `nx daemon --stop` run in the second
|
|
1039
|
+
checkout stopped the first one's daemon and deleted its socket directory. The `nx` CLI now finds the current
|
|
1040
|
+
directory's workspace by its `nx.json`, `nx` or `nx.bat` alone, ignoring the override, and refuses any command, before
|
|
1041
|
+
it reaches a daemon, socket or file, when one of those variables, `NX_DAEMON_SOCKET_DIR` or
|
|
1042
|
+
`NX_PROJECT_GRAPH_CACHE_DIRECTORY` names another workspace. The error names each such variable with its value and both
|
|
1043
|
+
workspaces, and says to run under the current workspace's own environment (`direnv exec <workspace> nx ...`).
|
|
1044
|
+
`NX_WORKSPACE_ROOT_PATH` names the directory it holds; a data, cache or socket directory belongs to the workspace
|
|
1045
|
+
whose root contains it, found by walking the directory's ancestors to a workspace marker (`nx.json`, `nx` or
|
|
1046
|
+
`nx.bat`). Workspaces are compared by the device and inode of their root directory, not by path, so every path to the
|
|
1047
|
+
current workspace is that workspace: a symlink, a bind mount or a firmlink. CI depends on it: `setup-devenv`
|
|
1048
|
+
bind-mounts the checkout at `/work/...` for devenv's evaluation cache, the environment names that mount, and the steps
|
|
1049
|
+
run in the checkout. Compared as resolved paths the two were different workspaces, and every `nx` of the job refused.
|
|
1050
|
+
A current directory inside no workspace is not refused. A directory outside every workspace, such as a socket
|
|
1051
|
+
directory under `/tmp`, is left unchecked: only the daemon behind it could say whose it is, and reaching that daemon
|
|
1052
|
+
is what the check prevents. The daemon, plugin workers and tasks do not start through the CLI and are not checked; an
|
|
1053
|
+
`nx` a task runs from a project directory resolves to the task's own workspace and passes. A cache or data directory
|
|
1054
|
+
that several checkouts share on purpose has to lie outside every workspace, or it is refused. Not yet proposed
|
|
1055
|
+
upstream.
|
|
1056
|
+
- **An idle daemon releases its workers after five minutes.** The patched `SERVER_INACTIVITY_TIMEOUT_MS` is 300000 ms,
|
|
1057
|
+
not stock Nx's three hours. Normal requests and watched source changes reset that timer; an open request connection,
|
|
1058
|
+
file watcher or graph listener retains the daemon even when silent. The last request connection's close starts a fresh
|
|
1059
|
+
five-minute idle interval. Cowshed's existing usage-aware keeper remains authoritative for sheds whose tree changes
|
|
1060
|
+
without a job, session or task-database holder. No daemon is disabled and the next normal client starts one again. The
|
|
1061
|
+
non-default, uncached host proof is `bun nx run tooling:nx-daemon-idle-proof` from the checkout root. Its declared
|
|
1062
|
+
compile dependency emits the Typia-transformed product before Bun runs it. The proof holds a real silent client beyond
|
|
1063
|
+
five minutes, disconnects, and records the kernel and daemon-log evidence of the next five-minute departure. It stays
|
|
1064
|
+
outside the bounded Bun shards; their fast regression checks the actual installed policy export.
|
|
1065
|
+
- **The daemon stops for an ignore file only when its watcher's rules may be stale.** Nx's native watcher reads its
|
|
1066
|
+
ignore files once, in `watch()`, and the daemon stopped itself on every watcher event for a `.gitignore` or
|
|
1067
|
+
`.nxignore`, whatever the file then held. A rewrite with the same bytes (a checkout, a formatter) reports one, and so
|
|
1068
|
+
does FSEvents for a write made shortly before the daemon's stream started, when fseventsd numbers it after that start.
|
|
1069
|
+
The stop destroys every client socket at once, so a file-watcher client lost the notification of the batch that
|
|
1070
|
+
carried the event, and a watcher waiting to judge an edit in that batch waited forever. The patch keeps the content
|
|
1071
|
+
hash of each ignore file the workspace context read, but only where neither the file nor its directory changed since
|
|
1072
|
+
just before the watcher's read (judged on the filesystem's own clock, by stamping a probe in the workspace data
|
|
1073
|
+
directory): those are the bytes the watcher's rules came from. An event stops the daemon unless its file is one of
|
|
1074
|
+
those and still hashes the same; a file changed between the watcher's read and the context's, or one the context never
|
|
1075
|
+
held, stops it as before.
|
|
1076
|
+
- **A failure before execution does not crash a completion lifecycle.** The
|
|
1077
|
+
[run-information repair](../../patches/README.md#run-information-after-a-pre-start-failure) owns the completion,
|
|
1078
|
+
history and timing contract.
|
|
1079
|
+
|
|
1080
|
+
Publishing or installing `@smoothbricks/nx-plugin` alone does **not** change a consumer's Nx. `smoo monorepo update`
|
|
1081
|
+
distributes the managed `patches/nx@23.2.1.patch` and maintains the root registry override `nx = "23.2.1"` plus its
|
|
1082
|
+
matching `patchedDependencies` entry. Run ordinary `bun install`, commit its lock changes, then prove a frozen install
|
|
1083
|
+
and the affected normal Nx gates. Do not replace the registry dependency with a local link or hide a failure by
|
|
1084
|
+
resetting the database or disabling the daemon.
|
|
1085
|
+
|
|
1086
|
+
The patch is keyed to `nx@23.2.1`; a different Nx version needs its own patch file. On an Nx upgrade, remove each hunk
|
|
1087
|
+
only when the installed upstream release contains that repair and the task-history namespace, cache-bound,
|
|
1088
|
+
resident-worker, store-resolution, task-graph, restore-time, racy-archive, skipped-task exit, daemon-claim,
|
|
1089
|
+
incomplete-run summary, foreign-environment, ignore-file and run-information regressions pass; preserve any repair not
|
|
1090
|
+
yet released. The restore-time regression fails without its hunk only on macOS, where the copy clones (on Linux
|
|
1091
|
+
`std::fs::copy` writes a fresh mtime), so run it on macOS before dropping the hunk. When every hunk is upstream, remove
|
|
1092
|
+
the `nx@23.2.1` and `@nx/js@23.2.1` entries from `patchedDependencies` and delete their patch files.
|
|
546
1093
|
|
|
547
1094
|
## Bun Test Tracing Generator
|
|
548
1095
|
|
|
@@ -566,8 +1113,227 @@ What it wires:
|
|
|
566
1113
|
|
|
567
1114
|
## Bounded Test Targets
|
|
568
1115
|
|
|
569
|
-
`@smoothbricks/nx-plugin:bounded-exec` runs a shell command
|
|
570
|
-
this executor so hung test processes fail predictably instead of blocking Nx
|
|
1116
|
+
`@smoothbricks/nx-plugin:bounded-exec` runs a shell command under a no-progress bound or a wall-clock ceiling, with a
|
|
1117
|
+
force-kill grace period. Test targets use this executor so hung test processes fail predictably instead of blocking Nx
|
|
1118
|
+
indefinitely.
|
|
1119
|
+
|
|
1120
|
+
A test target is bounded by progress alone: `idleTimeoutMs`, 180 s (`BOUNDED_TEST_IDLE_TIMEOUT_MS`, the hang backstop of
|
|
1121
|
+
`specs/cowshed/08_testing.md`), with no `timeoutMs`. The bound re-arms whenever the runner reports a finished test: a
|
|
1122
|
+
nextest status line carrying its run counter (`PASS [ 0.438s] (1/4) …`), a libtest `test … ... ok` line, or a bun
|
|
1123
|
+
result line or dot. Output that finishes no test is not progress: nextest's `SLOW` warning, repeated every slow-timeout
|
|
1124
|
+
period for a test still running, and a hung test's heartbeat both keep printing. A suite that keeps finishing tests
|
|
1125
|
+
therefore runs as long as it needs, however slow the host, and a test that never returns ends the run 180 s after the
|
|
1126
|
+
last test that finished. bun prints nothing per passing test when it believes an agent runs it, so the executor asks a
|
|
1127
|
+
lone `bun test` for `--dots`, one dot per finished test. A command that reports no tests makes progress only by exiting.
|
|
1128
|
+
nextest's slow-timeout and the serial groups' code-time classes only warn; no per-test bound kills a test or fails one
|
|
1129
|
+
that passed. `timeoutMs`, an absolute ceiling, is for commands that report no tests; a target names at least one of the
|
|
1130
|
+
two.
|
|
1131
|
+
|
|
1132
|
+
The final-five-second pre-signal snapshot attributes the runner's process tree and observed resources, with explicit
|
|
1133
|
+
incomplete/raced results and no delay to termination. Its timing, visibility and evidence contract is defined in
|
|
1134
|
+
`specs/cowshed/08_testing.md`, "Integration tests (real substrate)".
|
|
1135
|
+
|
|
1136
|
+
On a macOS host, every `bounded-exec` task runs with its own `TMPDIR` lease: a directory on one RAM-backed APFS volume
|
|
1137
|
+
per user, so test temp files never reach the SSD. The volume is created lazily by the first task that needs it and
|
|
1138
|
+
detached when the last live lease ends, since its pages return only on detach. It is a legacy DiskImages RAM disk, so it
|
|
1139
|
+
spends none of the AppleDiskImages2 attach budget cowshed counts (`specs/cowshed/01a_apfs.md`, "How the APFS host
|
|
1140
|
+
degrades"). A lease whose task died is reclaimed by the next task, unless an image is still attached from a file below
|
|
1141
|
+
it or a mount sits below it: those belong to whoever attached them, so the lease and the volume are kept and every new
|
|
1142
|
+
task names them. A target's own `env.TMPDIR` wins over the lease. Inside a cowshed sandbox, which cannot write the lock
|
|
1143
|
+
in `/private/tmp`, the task keeps its inherited `TMPDIR` and says so once. A failed task that left the volume nearly
|
|
1144
|
+
full names the volume.
|
|
1145
|
+
|
|
1146
|
+
Outside macOS, the executor never provisions an APFS RAM disk: it keeps the inherited `TMPDIR`, or uses the target's
|
|
1147
|
+
`env.TMPDIR` override, without reclaiming either directory. The host temp-directory proof runs the real default executor
|
|
1148
|
+
for that contract on Linux; on macOS it exercises RAM-volume recovery, shared leases and final detach.
|
|
1149
|
+
|
|
1150
|
+
The volume's kernel `O_EXLOCK` admits ledger mutations and releases on process death. It is held only for cheap lease
|
|
1151
|
+
and operation claims and their publication: host-wide process and disk-inventory probes, directory cleanup, provisioning
|
|
1152
|
+
and detach run outside it. A caller re-takes the lock and revalidates the volume, lease identity and its owned claim
|
|
1153
|
+
before publishing a mutation; another caller can take a lease while a dead lease is being inspected. An acquire that
|
|
1154
|
+
fails after taking its lease directory gives it back, and detaches the volume if that was the last lease, as the last
|
|
1155
|
+
release would have.
|
|
1156
|
+
|
|
1157
|
+
Checkouts of this plugin from before the ledger lease under the kernel lock alone and never read the claims, so the two
|
|
1158
|
+
protocols must never share a volume. The protocol's revision (2) names the volume, its lock and its ledger, and adds as
|
|
1159
|
+
many sectors to the RAM disk: the creator's stray reclaim matches user and size, so neither revision detaches the
|
|
1160
|
+
other's disk, mounted or still attaching. On a host running both, each revision has its own volume.
|
|
1161
|
+
|
|
1162
|
+
A creation or detach claim is void once its owner's pid is gone. A crashed owner whose pid was reused holds every caller
|
|
1163
|
+
waiting on the volume, and macOS gives no cheap way to tell the new process from the old without running `ps` under the
|
|
1164
|
+
lock. So every 15 s that a caller waits on the same claim, stderr names its kind, owner pid, token, ledger and the wait
|
|
1165
|
+
so far.
|
|
1166
|
+
|
|
1167
|
+
Each reclaim pass reads at most one I/O Registry and mount-table snapshot, not one per lease. A dead lease held only by
|
|
1168
|
+
mounted filesystems keeps a positive verdict with stable mount and lease identities. A valid positive still names and
|
|
1169
|
+
retains it without another host-wide probe; an unmount exposes the underlying RAM device and triggers a fresh probe.
|
|
1170
|
+
Image-backed and otherwise unkeyable holds stay uncached: device-node reuse has no proven cheap incarnation key. Every
|
|
1171
|
+
pass that claims such a lease probes it again, with a process census and the pass's inventory, which take seconds on a
|
|
1172
|
+
busy host; that latency falls on the claiming task, outside the lock. A negative verdict is never cached, probe errors
|
|
1173
|
+
never stand in for an empty holder set, and held directories stay unchanged.
|
|
1174
|
+
|
|
1175
|
+
Contended lock opens also use `O_NONBLOCK` and yield for 50 ms, avoiding a separate latent filesystem-worker deadlock
|
|
1176
|
+
when one Node process makes five concurrent requests with four workers. The observed gate failures instead came from
|
|
1177
|
+
cross-process waits behind long host-wide probes under the RAM lock. Real nested and sibling default-executor proofs run
|
|
1178
|
+
with background single-lease processes to cover that shape. No case or target timeout is raised.
|
|
1179
|
+
|
|
1180
|
+
Once the process and holder checks permit removal, cleanup restores owner read/write/search permission on directories
|
|
1181
|
+
through no-follow directory handles before deleting the lease. Read-only fixture directories cannot strand a dead lease
|
|
1182
|
+
and block later tasks. File modes and symlink targets stay untouched; a directory on another device refuses cleanup
|
|
1183
|
+
instead of changing an attached volume. Held leases undergo no permission changes.
|
|
1184
|
+
|
|
1185
|
+
The real RAM-volume regression walks one volume's lifecycle in a single test: unmarked debris and a live-lease refusal,
|
|
1186
|
+
recovery over the debris, concurrent leases, failed-acquire lease/claim rollback, read-only cleanup, a held dead lease,
|
|
1187
|
+
owner detach and reclaim, and the final detach. External symlink targets and held directory modes stay unchanged,
|
|
1188
|
+
without duplicate provision and teardown; its 60 s case bound is unchanged and its target has the progress bound.
|
|
1189
|
+
|
|
1190
|
+
The host target names both owned files with `./` paths: Bun otherwise interprets a bare path as a filter over recursive
|
|
1191
|
+
test discovery. This changes no selected case or timeout. The failed-provisioning regression keeps its exact injected
|
|
1192
|
+
refusal and rollback exit-status assertions, then checks the named volume and both mountpoints before test teardown. Its
|
|
1193
|
+
trace records command start/end times, refusal, inventory and separate teardown. Commands, refusal and inventory are
|
|
1194
|
+
also written to stderr immediately, because a runner timeout can end the process before the harness flushes its trace. A
|
|
1195
|
+
reported rollback failure must not be mistaken for a successful detach that left a volume behind.
|
|
1196
|
+
|
|
1197
|
+
A refused rollback also records the kernel mount and read-only filesystem/device `lsof` observations before teardown.
|
|
1198
|
+
Each observation has a one-second wait bound and retains its status and stderr; an incomplete probe is not an empty
|
|
1199
|
+
holder set. These observations do not prove which holder caused `EBUSY`, and change no case or target bound.
|
|
1200
|
+
|
|
1201
|
+
The native lifecycle walk also observes its real working-directory census through the existing process-host seam,
|
|
1202
|
+
retaining start/end times and the original error's exit code, signal and killed flag before rethrowing it unchanged.
|
|
1203
|
+
This locates whether a census was outstanding at a case timeout without substituting a fake process inventory. Its real
|
|
1204
|
+
host-command delegate records the same timestamps from case entry, retaining the actual argv, requested deadline and
|
|
1205
|
+
result without changing them; this covers the work before the first process census. Observation before delegation is
|
|
1206
|
+
timed separately from the delegate's wait, so logging cost is never presented as child execution time.
|
|
1207
|
+
|
|
1208
|
+
Whatever still works in a lease when its task ends is a leak, and is stopped before the lease goes. The usual one is a
|
|
1209
|
+
fixture's Nx daemon: Nx detaches it into a session of its own, so the process-group kill that ends a timed-out command
|
|
1210
|
+
never reaches it, and a test process that is killed runs no teardown. Deleting its root asks the daemon to leave only by
|
|
1211
|
+
deleting its record, which an idle daemon notices within a second and a wedged one never does, and a process left on the
|
|
1212
|
+
volume keeps it from detaching. So the task's own release, and the next task's reclaim of a dead lease, first hold
|
|
1213
|
+
discovered processes with SIGSTOP and rescan their descendants and lease working directories until a stopped census is
|
|
1214
|
+
stable. A process can fork after a table read but before its signal; only a table read after it is stopped closes that
|
|
1215
|
+
gap. Discovery failures attempt to resume every held process; one denied SIGCONT cannot strand the rest. Errors retain
|
|
1216
|
+
the original discovery failure and name any denied continuation with its PID and command. The settled census gets
|
|
1217
|
+
SIGTERM then SIGCONT, SIGKILL if anything has not exited five seconds later, and names each PID once on stderr
|
|
1218
|
+
(`stopped <pid> (<command>), which the command left working in its lease`). A lease with nothing in it costs no scan.
|
|
1219
|
+
Discovery follows stopped-census readiness under the enclosing operation's existing bound; the five-second exit grace
|
|
1220
|
+
does not include discovery or fixture readiness. The lease's working directories are read again only after the same
|
|
1221
|
+
census is stopped in consecutive process tables, including an empty one, which recovers children reparented out of the
|
|
1222
|
+
tree. This avoids host-wide cwd scans while stopped-table quiescence is still being established. The line is the leak
|
|
1223
|
+
check: a suite that prints it has a fixture to fix.
|
|
1224
|
+
|
|
1225
|
+
The volume mounts at `/Volumes/smoo-ram2-<uid>`, where DiskArbitration puts it: asking for any other mountpoint
|
|
1226
|
+
escalates to an administrator dialog that blocks every `diskutil` on the host. That mount is `noowners`, and launchd
|
|
1227
|
+
refuses a plist from it, so a test that bootstraps a launchd job keeps the plist outside `TMPDIR`.
|
|
1228
|
+
|
|
1229
|
+
The volume is created under its final name, `smoo-ram2-<uid>`, and is never renamed. It is mounted and marked at a
|
|
1230
|
+
private staging directory, then unmounted and placed at `/Volumes/smoo-ram2-<uid>` by DiskArbitration, so a creator that
|
|
1231
|
+
dies part way never leaves an unmarked volume at that mountpoint for the next one to dodge as `smoo-ram2-<uid> 1`.
|
|
1232
|
+
Before creating, a task detaches every RAM disk of the same user and size that is mounted nowhere but at those two
|
|
1233
|
+
paths: the debris of a creator that was killed or ran into its deadline. One that still carries a lease directory, or
|
|
1234
|
+
has something attached below it, is named in the task's error instead.
|
|
1235
|
+
|
|
1236
|
+
Every `hdiutil` and `diskutil` execution goes through `cowshed disk-command`, including host-test inventory and cleanup.
|
|
1237
|
+
The cowshed is the installed one on `PATH`, which must provide this command, unless `COWSHED_EXECUTABLE` names another.
|
|
1238
|
+
A target may set it in its `env`, workspace-relative or absolute: the task's own RAM volume work and its command then
|
|
1239
|
+
both run that cowshed, given to the command as an absolute path. This repository sets it to
|
|
1240
|
+
`packages/cowshed/bin/cowshed`, after `cowshed:build`, on the targets that exercise the facade (both real-APFS groups
|
|
1241
|
+
and `nx-plugin:test-host`), so they test this checkout's cowshed, not an installed one that may predate a command the
|
|
1242
|
+
checkout added. Cowshed alone owns disk-lifecycle leasing. Safe disk commands use
|
|
1243
|
+
`cowshed disk-command diskutil --deadline-ms <positive integer> -- <args>`: the CLI owns that total monotonic budget
|
|
1244
|
+
across lease admission and child execution, never starts a child after it expires, and kills and reaps an overrun safe
|
|
1245
|
+
child before returning. Node waits for the facade's actual terminal exit without a second timer or lease. Only protected
|
|
1246
|
+
image commands (any `hdiutil`, `diskutil image`, and `diskutil eject`) use `cowshed disk-command <tool> -- <args>` and a
|
|
1247
|
+
caller-bounded stop-wait: cowshed starts them detached from caller signals, with file-backed output, and never kills
|
|
1248
|
+
them, because killing an attach or detach can leak its kernel device until reboot. A still-running protected command
|
|
1249
|
+
retains its lease and is named in the timeout error. That timeout status is the wrapper's stop-wait outcome, not an
|
|
1250
|
+
actual disk-child exit or signal; native termination diagnostics come from the CLI's owned child wait. Other helpers
|
|
1251
|
+
retain their process-group deadline. The shared default remains 120 s. A process whose lease waits are recorded
|
|
1252
|
+
(`COWSHED_LEASE_WAIT_RECORD`) asks for its phase `alone`; a gateway from before `alone` refuses the field at once and is
|
|
1253
|
+
asked shared instead, for a minute. Cowshed's runner refuses `diskutil rename` and `renameVolume` for every caller; the
|
|
1254
|
+
RAM volume is created under its final name and never renamed.
|
|
1255
|
+
|
|
1256
|
+
`bun node_modules/@smoothbricks/nx-plugin/src/disk-command-policy.ts <repository...>` checks tracked and nonignored
|
|
1257
|
+
source, tests and scripts for direct disk-tool process calls. SmoothBricks runs it from `tooling:lint-disk-commands`;
|
|
1258
|
+
consumers wire the same guard into their root lint. Rust callers use `SystemCommandRunner` and `CommandRequest`; script
|
|
1259
|
+
callers use the CLI facade rather than introducing another disk-command runner.
|
|
1260
|
+
|
|
1261
|
+
Without an inherited parent pool, a `bounded-exec` command takes CPU tokens from cowshed's host budget before it starts,
|
|
1262
|
+
when the gateway answers on the same socket (`specs/cowshed/05_gateway.md`, "Host CPU budget"), so concurrent gates on
|
|
1263
|
+
one host keep their runnable work near the core count instead of each sizing its runners to the whole machine. It asks
|
|
1264
|
+
for the runner's requested parallelism — nextest's `--test-threads` or one per core, `bun test --parallel=N` N, a cargo
|
|
1265
|
+
build one per core, anything else one, or the target's `parallelism` option — bounded by positive inherited
|
|
1266
|
+
`NEXTEST_TEST_THREADS`/`CARGO_BUILD_JOBS` ceilings. A configured count can lower an inherited ceiling, never raise it.
|
|
1267
|
+
The runner uses `min(bounded request, granted tokens)`: nextest's environment and named flag, Bun's parallel flag, and
|
|
1268
|
+
Cargo's jobs are sized to that limit. Cargo's `RUST_TEST_THREADS` retains its independent inherited/configured ceiling.
|
|
1269
|
+
`BOUNDED_EXEC_CPU_TOKENS`, the sub-ledger and telemetry still name the actual held grant. The gateway shares tokens
|
|
1270
|
+
fairly between checkouts (Nx workspace roots). `timeoutMs` and `idleTimeoutMs` start at the grant. The tokens go back
|
|
1271
|
+
when the command exits, or with the executor's socket when it is killed. The wait prints one
|
|
1272
|
+
`cowshed: cpu-tokens wait done elapsed=… tokens=<granted>/<asked>` line. With no gateway listening, the command runs
|
|
1273
|
+
exactly as configured and nothing is said; a gateway from before the budget runs it unbudgeted and stderr says so.
|
|
1274
|
+
|
|
1275
|
+
A granted executor serves a private `BOUNDED_EXEC_CPU_POOL` with its actual token count. A nested default executor takes
|
|
1276
|
+
from that immediate parent pool, capped at `min(asked, inherited, ceil(cores/2))`, instead of taking from the host while
|
|
1277
|
+
the parent holds its grant. The pool accounts for siblings together; each admitted child serves only its own allocation
|
|
1278
|
+
to grandchildren. Closing a child's connection returns its local allocation, never the parent's upstream grant. Upstream
|
|
1279
|
+
loss revokes the downstream pool; losing a borrowed grant stops the command and records an ordinary failure, not a
|
|
1280
|
+
timeout. A bare count or absent endpoint follows normal host admission. A connected parent pool's refusal or protocol
|
|
1281
|
+
failure instead fails admission: taking from the host then could wait on the parent's own grant. `options.env` cannot
|
|
1282
|
+
inject either marker. Parent-pool grants record `cpu.inherited: true` and their actual local wait; the span also says
|
|
1283
|
+
`inherited=true`. Pool setup, listener and cleanup errors fail rather than bypassing accounting. Command bounds stay
|
|
1284
|
+
unchanged.
|
|
1285
|
+
|
|
1286
|
+
The absent-parent fixture uses the executor's injectable budget with a real, controlled outer pool as its backing. It
|
|
1287
|
+
proves a positive allocation and re-acquires it after the command ends, rather than depending on free tokens in the
|
|
1288
|
+
host-wide gateway while the test runner itself holds an allocation. A bare count or unbudgeted success is not proof. A
|
|
1289
|
+
land gate enters the existing task server with
|
|
1290
|
+
`cowshed exec <workspace> --class gate --host --want N [--wait-ms N] -- COMMAND [ARGS...]` before daemon-stop and Nx
|
|
1291
|
+
gate work. It requires a real gateway jobserver `class: "gate"` grant and serves that exact count as an existing live
|
|
1292
|
+
parent CPU pool; no environment-only class or unbudgeted fallback is accepted. The default **120000 ms** admission wait
|
|
1293
|
+
is explicitly overridable per invocation. A queued gate stops new background root admissions and takes whole tokens as
|
|
1294
|
+
in-flight grants end. Running background tests/FIFOs are never shrunk, frozen or killed to manufacture tokens. An
|
|
1295
|
+
expired wait returns `gate-admission-timeout` naming the wait and current checkout holders; it changes no task wall/idle
|
|
1296
|
+
bound. Requested/granted count and actual queue wait are logged, and a child's nonzero exit is preserved exactly.
|
|
1297
|
+
Compiling Cargo under the live pool borrows a sibling-accounted local allocation and receives its own private child
|
|
1298
|
+
pool, so grandchildren never compete with their occupied ancestor. See `specs/cowshed/05_gateway.md`, "Gate admission by
|
|
1299
|
+
whole-grant drain", for the guarantee and its deliberate limits.
|
|
1300
|
+
|
|
1301
|
+
Every `bounded-exec` task records why it ended in `.nx/workspace-data/bounded-exec/<task>/verdict.json`, keyed by the
|
|
1302
|
+
task id and the hash it ran at:
|
|
1303
|
+
|
|
1304
|
+
- `bound`: nothing failed but a wall-clock bound. Either the command outlived `timeoutMs`, or every failing test in the
|
|
1305
|
+
runner's JUnit report failed on its runner's per-test timeout (`cargo-nextest` writes `type="test timeout"` where a
|
|
1306
|
+
configuration still sets `terminate-after`, `bun test` writes `type="TimeoutError"`). A loaded host produces exactly
|
|
1307
|
+
these. When `timeoutMs` fires, the tests `cargo-nextest` was still running are reported as failures of
|
|
1308
|
+
`type="test abort"`, `process aborted with signal 15 (SIGTERM)` (SIGKILL once the force-kill follows); they are the
|
|
1309
|
+
bound's casualties and do not make the run `failed`. A test that dies of any other signal, or that is aborted with no
|
|
1310
|
+
bound to explain it, does.
|
|
1311
|
+
- `wedged`: `idleTimeoutMs` fired: no test finished for that long. A host that is only slow still finishes tests, so
|
|
1312
|
+
this is a hang, never load. Its `tests` name the tests still running: those the runner said were running (nextest's
|
|
1313
|
+
`SLOW`, libtest's `has been running for over`) and never said finished, and those the JUnit report says it aborted
|
|
1314
|
+
when the bound stopped it.
|
|
1315
|
+
- `failed`: a test failed on its own, or the command exited non-zero with no report naming a timeout.
|
|
1316
|
+
- `passed`.
|
|
1317
|
+
|
|
1318
|
+
The report needs no target configuration. When the command runs exactly one `bun test` or `nextest run`, the executor
|
|
1319
|
+
asks that runner for a JUnit report in the task's directory: `bun test` through `--reporter=junit --reporter-outfile`,
|
|
1320
|
+
nextest through a one-key tool config (nextest takes the report path only from configuration). A `bun` is any unquoted
|
|
1321
|
+
command word whose basename is `bun`, so `../bun-runtime/.runtime/bun test` is the runner too, and the rewrite keeps the
|
|
1322
|
+
binary the target named. A script that spawns the runner itself passes
|
|
1323
|
+
`--reporter=junit --reporter-outfile="$BOUNDED_EXEC_JUNIT"` on; the variable is set for every task. A command with no
|
|
1324
|
+
runner, or with two (they would write one report), is judged by its exit alone.
|
|
1325
|
+
|
|
1326
|
+
`smoo-nx-bound-failures` reads the workspace's last Nx run (`run.json` in Nx's `cacheDir`, the directory
|
|
1327
|
+
`StoreRunInformationLifeCycle` writes: the checkout's own `.nx/cache` under smoo's workspace policy) against those
|
|
1328
|
+
records and prints one line per failed task. It exits 0 only when the run failed and every failed task's record, for
|
|
1329
|
+
that task's hash, is a `bound`; a failed task without one is not. This diagnostic is not land acceptance. The land
|
|
1330
|
+
wrapper, `tooling/land-host.sh`, requires `nx run-many -t gate` to exit zero: bounds, assertions and infrastructure
|
|
1331
|
+
failures all stop the land without a retry or waiver. `smoo-nx-failures OUTPUT` reports unique failing tasks and
|
|
1332
|
+
deduplicated exact diagnostics from that invocation's retained output, not a shared last-run record. The portable
|
|
1333
|
+
`@smoothbricks/nx-plugin/failure-summary` export owns `summarizeFailures(output)` and `runFailureSummary(args)`;
|
|
1334
|
+
repository entrypoints such as `tooling/land-summary.ts` call that owner without copying the parser. Its zero exit means
|
|
1335
|
+
reporting succeeded, never that the failed gate may land. Known diagnostic sections omit fixture progress and daemon
|
|
1336
|
+
debug dumps while preserving multiline assertions and source locations; unknown formats retain their output.
|
|
571
1337
|
|
|
572
1338
|
The shared policy API is exported from `@smoothbricks/nx-plugin/bounded-test-policy` for generators or other workspace
|
|
573
1339
|
tools that need to normalize package JSON consistently.
|
|
@@ -581,10 +1347,81 @@ The generator rewrites `package.json` so `nx.targets.test` uses:
|
|
|
581
1347
|
- executor `@smoothbricks/nx-plugin:bounded-exec`
|
|
582
1348
|
- command preserved from an existing `nx:run-commands` test target or direct `scripts.test`
|
|
583
1349
|
- `cwd: "{projectRoot}"`
|
|
584
|
-
- `
|
|
1350
|
+
- `idleTimeoutMs: 180000`
|
|
585
1351
|
- `killAfterMs: 10000`
|
|
586
1352
|
- package script alias `nx run <project>:test --outputStyle=stream`
|
|
587
1353
|
|
|
1354
|
+
Workspace updates preserve no-op test targets rather than adding a second test runner. A no-op `test` aggregate passes
|
|
1355
|
+
the check when it reaches at least one local test execution leg and every such leg is bounded, including through nested
|
|
1356
|
+
no-op aggregates: `bounded-exec` with a command, a `cwd`, a positive `idleTimeoutMs` and `killAfterMs`, and no
|
|
1357
|
+
`timeoutMs`. Upstream and non-test prerequisites such as `^build`, `build`, `typecheck-tests`, and Cargo
|
|
1358
|
+
compilation/archive targets are ignored by this test-leg check. Each bounded Bun leg must declare a positive per-test
|
|
1359
|
+
`--timeout` no greater than its own `idleTimeoutMs`; the migration's 30-second default is not a limit on explicitly
|
|
1360
|
+
bounded lanes. Only a leg that runs `bun test` must start in the project root or its `src/`, where Bun's test-file
|
|
1361
|
+
discovery is cheap; any other command (a cargo workspace's per-crate nextest legs run from the workspace root, a gate
|
|
1362
|
+
may run a script that lives in another project) may use any `cwd`.
|
|
1363
|
+
|
|
1364
|
+
## Test Fixtures That Run Nx
|
|
1365
|
+
|
|
1366
|
+
The executor's real-process tests acknowledge handler installation and process-group membership before delivering a
|
|
1367
|
+
timeout signal. Cooperative termination requires its handler receipt and no force kill; descendant lifetime pipes prove
|
|
1368
|
+
actual closure. The abort-report walk requires the runner's named cancellation record as well as total-bound
|
|
1369
|
+
classification, so an executor-only expiry cannot stand in for a report the runner never wrote.
|
|
1370
|
+
|
|
1371
|
+
Most tests never start Nx. Plugin behaviour is the value `createNodesV2` returns, so a test asserts on that value, and
|
|
1372
|
+
where it needs Nx's verdict on it, it calls the Nx function that gives the verdict, in process: `createTaskGraph` for
|
|
1373
|
+
the order a run-many runs tasks in, `getOutputsForTargetAndConfiguration` for what a cache hit restores,
|
|
1374
|
+
`globWithWorkspaceContextSync` for the files a fileset input hashes, and a `runtime` input's command under `sh -c` from
|
|
1375
|
+
the workspace root for the value Nx keys on. `smoo-nx-exec`'s decisions are functions of Nx-shaped data — selector and
|
|
1376
|
+
arguments, the environment it hands Nx, the daemon's file table and output verdicts, the cache records, the artifact
|
|
1377
|
+
tree a restore would leave — and are called directly; its process handling (stdio, exit status, signals, cwd, `execve`)
|
|
1378
|
+
runs against a stand-in `nx` script with the daemon off, which loads no Nx at all. A test starts `nx` only when that
|
|
1379
|
+
process is the subject: a fresh daemon serving `smoo-nx-exec`'s probe (outputs it holds no record of, and the socket it
|
|
1380
|
+
binds for a root entered from another workspace's shell), daemon-hosted plugin workers, task history, a hunk of the
|
|
1381
|
+
repository's Nx patch, and fixture teardown.
|
|
1382
|
+
|
|
1383
|
+
A test that runs real Nx in a temp workspace runs it daemonless unless the daemon is the behaviour under test: a
|
|
1384
|
+
daemonless Nx leaves nothing running once it exits. A daemon a fixture does start works in the fixture root, idling and
|
|
1385
|
+
watching files until something stops it, so `@smoothbricks/nx-plugin/testing` (used by this package's and the CLI's
|
|
1386
|
+
tests) owns those roots:
|
|
1387
|
+
|
|
1388
|
+
- `ownedFixtureRoot(parentDirectory, suite, prefix)` creates a root in the run directory of the test process that owns
|
|
1389
|
+
it, `<parentDirectory>/smoothbricks-fixtures/<suite>/run-<pid>`. The caller always names the parent: `tmpdir()`, or
|
|
1390
|
+
the directory of the volume a fixture has to lie on. The parent is resolved through every link first, so one spelled
|
|
1391
|
+
two ways is one run, and a process holds one run for each parent and suite. The first root a process creates for a
|
|
1392
|
+
suite under a parent first reclaims every run of that suite there whose owner pid is gone: each process working in it
|
|
1393
|
+
gets SIGTERM (SIGKILL if it outlives 10 s); once all have exited, every socket alias naming a workspace of that run
|
|
1394
|
+
(see `publishFixtureSocketDirectory`) is unlinked, whether that workspace still exists or not, then the run is
|
|
1395
|
+
deleted. A run whose owner is alive is never touched by another process, and other entries of the socket namespace are
|
|
1396
|
+
left alone. A killed process therefore leaves what it owned in a `run-<pid>` of its own, inside that parent's
|
|
1397
|
+
`smoothbricks-fixtures` directory, never as a bare prefix among whatever else the parent holds, and the next process
|
|
1398
|
+
to use that parent and suite reclaims it. The owner itself, on SIGTERM, SIGINT or SIGHUP (a bounded test leg's timeout
|
|
1399
|
+
sends SIGTERM, then SIGKILL 10 s later), first stops whatever still works in every run it holds, all at once, and
|
|
1400
|
+
makes no new root meanwhile; a repeated signal waits for that. Then it dies of the first signal it got.
|
|
1401
|
+
- `reclaimDeadFixtureRuns(parentDirectory, suite)` runs that sweep on demand, for a test that has just watched a
|
|
1402
|
+
fixture-owning process die.
|
|
1403
|
+
- `fixtureSocketDirectory(workspace)` names, without creating anything, the `nx` socket leaf below a short parent alias
|
|
1404
|
+
in the sandbox's socket namespace (`socketNamespace`). The alias is named for a hash of the workspace's physical path,
|
|
1405
|
+
so every client and daemon stop of a workspace names the same leaf, and no two fixtures share one. A namespace too
|
|
1406
|
+
small for Nx's longest socket name is refused here, before anything is published.
|
|
1407
|
+
- `publishFixtureSocketDirectory(workspace)` makes that alias and returns the leaf. The workspace must be a physical
|
|
1408
|
+
directory in a run this process holds (`ownedFixtureRoot`), so the run's sweep can reclaim the alias if the process
|
|
1409
|
+
dies; anything else is refused. The alias points at the fixture's own `.nx/sockets` (mode 0700, never through a
|
|
1410
|
+
symlinked `.nx`); Nx makes the real `nx` leaf. This package's `withNxFixture` publishes its workspace and passes the
|
|
1411
|
+
directory to the body, and the CLI's `runFixtureNx` publishes each root it runs in. Both `fixtureNxEnv`s take the
|
|
1412
|
+
directory as an argument and publish nothing.
|
|
1413
|
+
- `ownedWorkspaceSocketDirectory(workspace, slot)` creates the real leaf `a` or `b` beside `nx` below a published alias,
|
|
1414
|
+
for tests of multiple daemon socket directories. Its caller stops the daemon before removing the leaf.
|
|
1415
|
+
- `retireFixtureSocketDirectory(workspace)` unlinks only the verified parent alias after the daemon exits, named from
|
|
1416
|
+
the workspace's path alone, so a deleted workspace retires too. A colliding non-symlink or retargeted alias refuses
|
|
1417
|
+
instead of removing another owner's entry; removing the fixture root reclaims its physical socket directories.
|
|
1418
|
+
- `stopNxDaemon(workspace, stop)` runs the caller's `nx daemon --stop` and waits until the daemon recorded in
|
|
1419
|
+
`<workspace>/.nx/workspace-data/d/server-process.json`, and its descendants, has exited, then ends every process still
|
|
1420
|
+
working in the workspace. That final census also catches editor probes reparented when the daemon exits.
|
|
1421
|
+
|
|
1422
|
+
A fixture retires on every exit path, a throwing body included: stop its daemon, unlink its socket alias, then delete
|
|
1423
|
+
its root. A daemon that outlives the stop keeps the alias and the root, so the next run's sweep still finds both.
|
|
1424
|
+
|
|
588
1425
|
## Managed workspace files
|
|
589
1426
|
|
|
590
1427
|
`nx generate @smoothbricks/nx-plugin:managed-files` stages the same managed files as `smoo monorepo update`, without
|
|
@@ -595,6 +1432,46 @@ The plugin owns the packaged templates, pure rendering and content-preservation
|
|
|
595
1432
|
files through Nx `Tree` and uses Nx's resolved project graph for inferred targets. The CLI only supplies the filesystem
|
|
596
1433
|
and process boundary. There is no separate serialized change plan.
|
|
597
1434
|
|
|
1435
|
+
The idempotence witness compares the second pass's managed-target roster with the populated first pass before checking
|
|
1436
|
+
unchanged actions, an empty staged diff and retained bytes. An empty result is not evidence that generation inspected
|
|
1437
|
+
the managed files.
|
|
1438
|
+
|
|
598
1439
|
Files with a `# smoo-local` tail or `# smoo-local-begin` / `# smoo-local-end` blocks retain those sections. Matching
|
|
599
1440
|
source symlinks are preserved; conflicting local blocks and broken or external links are reported instead of
|
|
600
1441
|
overwritten. The generator does not remove a repository-owned file when a capability is disabled.
|
|
1442
|
+
|
|
1443
|
+
### Workflow security and bounded execution
|
|
1444
|
+
|
|
1445
|
+
Managed action references use full commit SHAs with their release tags in comments. Read-only jobs explicitly set
|
|
1446
|
+
`persist-credentials: false`; checkout includes the required history and blobs so later private historical reads do not
|
|
1447
|
+
need credentials left in Git configuration. Jobs that push release or managed-file refs retain credentials explicitly,
|
|
1448
|
+
with their reason beside checkout. A platform producer that fetches a private release-candidate ref also requires
|
|
1449
|
+
authenticated Git transport; its checkout names that reason.
|
|
1450
|
+
|
|
1451
|
+
| Job | Persist checkout credentials | Reason |
|
|
1452
|
+
| ----------------------------------------------------------------- | ---------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
|
|
1453
|
+
| Validate, E2E, cross-test execution, deploy-only, preview cleanup | No | Complete history and blobs make historical reads local; these jobs do not push or fetch later private refs. |
|
|
1454
|
+
| Linux release candidate | No | The candidate is versioned locally and transferred as a Git bundle, not pushed. |
|
|
1455
|
+
| macOS / cross-platform release producer | Yes | `releaseCollectPlatformOutputs` fetches private candidate/repair refs after checkout (`packages/cli/src/release/index.ts:335`); a full initial clone cannot contain refs created later. |
|
|
1456
|
+
| Release publisher and pending-release healer | Yes | Publishing and repair fetch and push release refs. |
|
|
1457
|
+
| Managed-file healer | Yes | It pushes the regenerated review branch. |
|
|
1458
|
+
|
|
1459
|
+
Job timeouts remain generous and finite: Validate and compact release jobs use 45 minutes; Apple/cross-platform
|
|
1460
|
+
production and combined platform releases use 90 minutes to accommodate cold native builds; production deployment and
|
|
1461
|
+
managed-file healing use 30 minutes; deployed E2E and preview cleanup use 15 minutes. Cross-target test execution uses
|
|
1462
|
+
30 minutes. These are job hang bounds, not replacements for package test deadlines.
|
|
1463
|
+
|
|
1464
|
+
Forgejo artifact steps use managed local JavaScript actions, not GitHub clients or absolute `uses:` URLs. They fetch and
|
|
1465
|
+
verify the pinned GHES-gate-free upstream commit before running its original bundled entry point under the runner's Node
|
|
1466
|
+
runtime. This preserves artifact inputs, outputs and execution conditions while allowing the unchanged jactionlint
|
|
1467
|
+
default rules to validate the same workflow grammar for both providers.
|
|
1468
|
+
|
|
1469
|
+
A configured flattened producer runs its preflight even when no platform family matches. Its declared environment is
|
|
1470
|
+
merged directly into every emitted step of its producer job; step-local fields take precedence. Producer names are
|
|
1471
|
+
encoded as YAML strings so identifiers such as `TRUE` and `NULL` retain their exact spelling, not scalar coercions.
|
|
1472
|
+
Runner-dependent values are evaluated in the valid step-level context, not job env. No producer value is exported
|
|
1473
|
+
through `GITHUB_ENV`: GitHub
|
|
1474
|
+
[forbids setting `NODE_OPTIONS` through that environment file](https://docs.github.com/en/actions/reference/workflows-and-actions/workflow-commands#setting-an-environment-variable).
|
|
1475
|
+
If a declaration name cannot be carried, generation fails with that name in the diagnostic instead of silently dropping
|
|
1476
|
+
it. Conditional preflight and platform-build steps are numbered from the actual emitted job sequence, so every following
|
|
1477
|
+
step anchor remains exact.
|