Cache native dependency lifecycle builds

.github/workflows/ci.yml

@@ -145,6 +145,9 @@ jobs:
       - name: Build workspace
         run: cargo build --workspace --locked
 
+      - name: Native build cache workflow (Linux sandbox)
+        run: cargo nextest run --locked -p lpm-workflows --test native_build_cache --no-fail-fast --status-level slow --final-status-level fail
+
       - name: Unit and integration tests (non-workflow, non-CLI)
         run: cargo nextest run --locked --workspace --exclude lpm-workflows --exclude lpm-cli --no-fail-fast --status-level slow --final-status-level fail
 

bench/perf-results/native-build-cache-20260712-macos-arm64-native-toolchain.json

@@ -0,0 +1,49 @@
+{
+  "generated_at": "2026-07-12T21:38:02.229Z",
+  "platform": "darwin-arm64",
+  "node": "v22.22.1",
+  "lpm_bin": "target/release/lpm-rs",
+  "fixture": "bench/audit-fixtures/native/sharp-image",
+  "samples": 10,
+  "artifact_count": 2,
+  "artifact_bytes": 503469,
+  "scenarios": {
+    "cache_miss": {
+      "samples": 10,
+      "wall_ms_median": 173.1,
+      "wall_ms_p95": 280,
+      "max_rss_bytes_median": 47169536,
+      "user_cpu_ms_median": 40,
+      "system_cpu_ms_median": 20,
+      "cache_hits": 0,
+      "cache_misses": 10,
+      "lifecycle_ms_avoided": 0
+    },
+    "local_hit": {
+      "samples": 10,
+      "wall_ms_median": 59.07,
+      "wall_ms_p95": 72.25,
+      "max_rss_bytes_median": 41631744,
+      "user_cpu_ms_median": 20,
+      "system_cpu_ms_median": 10,
+      "cache_hits": 10,
+      "cache_misses": 0,
+      "lifecycle_ms_avoided": 1049,
+      "speedup_vs_miss": 2.93,
+      "wall_reduction_percent": 65.88
+    },
+    "ci_warm_store_hit": {
+      "samples": 10,
+      "wall_ms_median": 58.49,
+      "wall_ms_p95": 61.69,
+      "max_rss_bytes_median": 41615360,
+      "user_cpu_ms_median": 10,
+      "system_cpu_ms_median": 10,
+      "cache_hits": 10,
+      "cache_misses": 0,
+      "lifecycle_ms_avoided": 1051,
+      "speedup_vs_miss": 2.96,
+      "wall_reduction_percent": 66.21
+    }
+  }
+}

bench/perf-results/native-build-cache-20260712-macos-arm64-native-toolchain.md

@@ -0,0 +1,27 @@
+# Native lifecycle-build cache benchmark — NativeToolchain
+
+- Platform: macOS arm64
+- Node: v22.22.1
+- Fixture: `bench/audit-fixtures/native/sharp-image`
+- Release-build samples per scenario: 10
+- Median build artifact size: 503,469 bytes
+
+| Scenario | Median wall | p95 wall | Median RSS | User CPU | System CPU | Result |
+|---|---:|---:|---:|---:|---:|---:|
+| Strict cache miss | 173.10 ms | 280.00 ms | 47,169,536 B | 40 ms | 20 ms | 10 misses |
+| Local artifact hit | 59.07 ms | 72.25 ms | 41,631,744 B | 20 ms | 10 ms | 2.93×; 65.88% lower wall time |
+| Stable-path CI warm-store hit | 58.49 ms | 61.69 ms | 41,615,360 B | 10 ms | 10 ms | 2.96×; 66.21% lower wall time |
+
+Every measured scenario used the same strict sandbox. Sharp follows the
+`NativeToolchain` cache path, so every hit validates the persisted compiler,
+SDK, package-database, pkg-config, Homebrew receipt, and Node-header snapshot.
+The first measured miss computed the complete host fingerprint in 126 ms;
+subsequent unchanged-host key derivations took 15–24 ms.
+Miss and local-hit samples rematerialized the same project. CI-like samples
+deleted and recreated a checkout at one stable path while retaining the warm
+LPM store.
+
+Across ten local hits, artifact metadata recorded 1,049 ms of lifecycle
+execution avoided. Across ten stable-path CI hits, it recorded 1,051 ms
+avoided. The measured hit remains beneficial after paying the conservative
+host-snapshot validation cost.

bench/perf-results/native-build-cache-20260712-macos-arm64.json

@@ -0,0 +1,49 @@
+{
+  "generated_at": "2026-07-12T13:19:13.671Z",
+  "platform": "darwin-arm64",
+  "node": "v22.22.1",
+  "lpm_bin": "target/release/lpm-rs",
+  "fixture": "bench/audit-fixtures/native/esbuild-prebuilt",
+  "samples": 10,
+  "artifact_count": 2,
+  "artifact_bytes": 134444,
+  "scenarios": {
+    "cache_miss": {
+      "samples": 10,
+      "wall_ms_median": 138.03,
+      "wall_ms_p95": 449.94,
+      "max_rss_bytes_median": 52248576,
+      "user_cpu_ms_median": 50,
+      "system_cpu_ms_median": 20,
+      "cache_hits": 0,
+      "cache_misses": 10,
+      "lifecycle_ms_avoided": 0
+    },
+    "local_hit": {
+      "samples": 10,
+      "wall_ms_median": 32.33,
+      "wall_ms_p95": 37.27,
+      "max_rss_bytes_median": 41467904,
+      "user_cpu_ms_median": 10,
+      "system_cpu_ms_median": 10,
+      "cache_hits": 10,
+      "cache_misses": 0,
+      "lifecycle_ms_avoided": 1356,
+      "speedup_vs_miss": 4.27,
+      "wall_reduction_percent": 76.58
+    },
+    "ci_warm_store_hit": {
+      "samples": 10,
+      "wall_ms_median": 29.5,
+      "wall_ms_p95": 30.12,
+      "max_rss_bytes_median": 41418752,
+      "user_cpu_ms_median": 10,
+      "system_cpu_ms_median": 10,
+      "cache_hits": 10,
+      "cache_misses": 0,
+      "lifecycle_ms_avoided": 1053,
+      "speedup_vs_miss": 4.68,
+      "wall_reduction_percent": 78.63
+    }
+  }
+}

bench/perf-results/native-build-cache-20260712-macos-arm64.md

@@ -0,0 +1,24 @@
+# Native lifecycle-build cache benchmark
+
+- Platform: macOS arm64
+- Node: v22.22.1
+- Fixture: `bench/audit-fixtures/native/esbuild-prebuilt`
+- Release-build samples per scenario: 10
+- Median build artifact size: 134,444 bytes
+
+| Scenario | Median wall | p95 wall | Median RSS | User CPU | System CPU | Result |
+|---|---:|---:|---:|---:|---:|---:|
+| Strict cache miss | 138.03 ms | 449.94 ms | 52,248,576 B | 50 ms | 20 ms | 10 misses |
+| Local artifact hit | 32.33 ms | 37.27 ms | 41,467,904 B | 10 ms | 10 ms | 4.27×; 76.58% lower wall time |
+| Stable-path CI warm-store hit | 29.50 ms | 30.12 ms | 41,418,752 B | 10 ms | 10 ms | 4.68×; 78.63% lower wall time |
+
+Every measured scenario used the same strict sandbox. Miss and local-hit
+samples rematerialized the same project. CI-like samples deleted and recreated
+a checkout at one stable path while retaining the warm LPM store; observable
+lifecycle paths are part of the cache key, so moving a checkout intentionally
+invalidates the artifact.
+
+Across ten local hits, artifact metadata recorded 1,356 ms of lifecycle
+execution avoided. Across ten stable-path CI hits, it recorded 1,053 ms
+avoided. This fixture has a short lifecycle command; native compilation
+workloads should have a substantially larger absolute hit benefit.

bench/scripts/README.md

@@ -1,5 +1,22 @@
 # Install benchmark scripts
 
+## Native lifecycle-build cache
+
+Build the release CLI, then compare strict cache misses, local artifact hits,
+and CI-like stable-path checkout recreation against sharp's NativeToolchain
+cache path. Every measured scenario uses the same strict sandbox, so cache-hit
+timings include host compiler, SDK, package database, and pkg-config
+fingerprinting:
+
+```bash
+cargo build --release --locked -p lpm-cli --bin lpm-rs
+node bench/scripts/native-build-cache-benchmark.mjs --self-test
+node bench/scripts/native-build-cache-benchmark.mjs --samples 10
+```
+
+The harness writes raw rows plus JSON and Markdown summaries under
+`bench/perf-results/native-build-cache-<timestamp>/`.
+
 ## Production-readiness harness
 
 `run-install-readiness.mjs` is the current install-readiness harness. It is