fix(unplugin): bound per-delivery watch-input derivation; ci(release): fail-fast publish timeouts

.github/workflows/nestia.yml

@@ -90,10 +90,18 @@ jobs:
           name: ttsc-tarballs
           path: experimental/tarballs
 
-      - name: Clone nestia
-        run: |
-          rm -rf experimental/nestia
-          git clone --depth=1 --branch master https://github.com/samchon/nestia experimental/nestia
+      # Keep this integration gate attributable to ttsc. nestia master became
+      # red in its own sdk lane after this revision (PR #1594, openapi_v2), so a
+      # moving clone made every ttsc change inherit an unrelated upstream
+      # failure. Advance the pin only to a revision with a reviewed green full
+      # suite; 3b27e69b is the merge of nestia #1593 and its complete test run
+      # 29959327169 passed.
+      - name: Check out verified nestia integration revision
+        uses: actions/checkout@v4
+        with:
+          repository: samchon/nestia
+          ref: 3b27e69b69dea3f102315042dce87c18d81be74a
+          path: experimental/nestia
 
       - name: Use local ttsc tarballs
         working-directory: experimental/nestia

.github/workflows/release.yml

@@ -10,6 +10,8 @@ permissions:
 jobs:
   publish:
     runs-on: ubuntu-latest
+    # One full build-plus-publish cycle takes ~35 minutes end to end.
+    timeout-minutes: 60
     steps:
       # `pnpm run build` cross-compiles every platform package in one job: seven
       # GOOS/GOARCH pairs, each producing a tsgo-sized ttsc and ttscserver, each
@@ -78,13 +80,23 @@ jobs:
           ! printf '%s\n%s\n' "$ttsc_out" "$server_out" | grep -q '0.0.0-dev'
           ! printf '%s\n%s\n' "$ttsc_out" "$server_out" | grep -q 'commit dev'
       - name: Publish VS Code Marketplace extension
+        # Fail fast on a stalled Marketplace call: the healthy step takes
+        # under a minute (11-28 s observed), and the default 360-minute job
+        # timeout would otherwise hold the npm publish hostage for hours
+        # while looking like a slow build (#1008).
+        timeout-minutes: 10
         run: SKIP_BUILD=1 bash scripts/publish-vscode-marketplace.sh
         env:
           VSCE_PAT: ${{ secrets.VSCE_PAT }}
       - name: Publish to npm
+        # Same fail-fast contract as the Marketplace step: the healthy step
+        # takes ~5.5 minutes; the default 360-minute job timeout would hold a
+        # stalled publish silently for hours (#1008).
+        timeout-minutes: 20
         run: pnpm run package:latest:publish
   wasm-smoke:
     runs-on: ubuntu-latest
+    timeout-minutes: 15
     needs: publish
     steps:
       - uses: actions/setup-node@v6
@@ -102,6 +114,7 @@ jobs:
           node -e "if(!require('@ttsc/wasm'))process.exit(1)"
   vscode-smoke:
     runs-on: ubuntu-latest
+    timeout-minutes: 15
     needs: publish
     steps:
       - uses: actions/checkout@v4
@@ -146,6 +159,7 @@ jobs:
           NODE
   release:
     runs-on: ubuntu-latest
+    timeout-minutes: 15
     needs: [wasm-smoke, vscode-smoke]
     permissions:
       contents: write

experimental/unplugin-perf/README.md

@@ -3,12 +3,15 @@
 This experiment drives the **real** `@ttsc/unplugin` Rollup plugin object over a synthetic project of `N` TypeScript files and measures, per simulated build:
 
 - **plugin runs** — how many times the whole project is re-transformed (native plugin spawns). A correct per-build cache transforms the project **once**.
-- **`fs.readFileSync` calls / bytes** — the file-system work the adapter performs while serving the `N` modules. A correct cache walks the tree a constant number of times; the current code re-hashes the entire project on every module.
+- **`fs.readFileSync` calls / bytes** — the file-system work the adapter performs while serving the `N` modules. A correct cache walks the tree a constant number of times per build.
+- **fs identity probes** — the `existsSync`/`realpathSync.native` call volume the macOS `pathIdentityKey` branch would pay, counted under a simulated `process.platform = "darwin"` so it is observable on any host. Correct watch-input derivation pays it once per distinct graph path per generation, not once per module delivery.
 
-The guarded invariant is **`plugin runs == 1`**: a build must transform the whole project exactly once and serve every other module from the per-build cache. The harness exits non-zero if any build exceeds one transform.
+The guarded invariants are **`plugin runs == 1`** and, for the graph scenario, **bounded probes per module** — the harness exits non-zero when either breaks.
 
 - **Scenario A — output keys under the project root.** The cache hits, so the project is transformed once. (`reads` still grow with `N`: validating a cache hit re-hashes the project to detect a sibling-file change — bounded work that the existing invalidation contract requires.)
 - **Scenario B — one output key outside the validator's directory walk** (a `node_modules/**` path, exactly what the native host emits for program dependencies). Before the fix the store-time and validate-time hash key sets diverged, the cache _never_ hit, and the whole project was re-transformed once per module (`plugin runs == N`); now the cache hits and `plugin runs == 1`.
+- **Scenario C — a graph-bearing envelope (the typia >= 13.1.19 shape).** The sidecar stamps `graph` (every module edges to its next sibling plus `K` external `node_modules` declarations, and one missing resolution candidate) and per-file `dependencies`. Before the #1007 fix every cache-hit delivery re-walked the whole graph: probes per module grew linearly with the edge count (6.7k/12k/22.8k probes per module at E=2.6k/5.1k/10.1k, ~76-95 s for 99 deliveries), the O(modules x edges) syscall storm behind the #970 residual stall on macOS. The gate is 64 probes per module; the fixed code derives once per generation and stays flat regardless of `E`.
+- **Scenario D — the same envelope without a build boundary** (the Vite development server's persistent-validation mode). Measurement only: it quantifies the deliberate per-delivery complete-input validation against a large external input set. The path is linear and is the #980 freshness contract, so no invariant gate applies.
 
 The adapter source is bundled on the fly with esbuild (with `ttsc` and `unplugin` kept external), so the production code path runs unmodified — no rebuilt `lib` required.