@vzn/vx 0.0.497 → 0.0.498

package/README.md

@@ -18,17 +18,15 @@
 
 <!-- bench:start — generated by packages/vx-bench/update-site.ts from results.json; do not hand-edit -->
 
-| 1,090 packages, 3,270 tasks      | vx         | Turborepo                   | Nx                        |
-| -------------------------------- | ---------- | --------------------------- | ------------------------- |
-| Cold build: time the runner adds | **9 s**    | 1 min 35 s (vx 11× faster)  | 31 min (vx 218× faster)   |
-| Cold build: CPU burned           | **34 s**   | 1 min 13 s (vx 2.1× faster) | 114 min (vx 199× faster)  |
-| Fully cached run                 | **476 ms** | 760 ms (vx 1.5× faster)     | 3.59 s (vx 7.5× faster)   |
-| Overhead per package             | **8 ms**   | 88 ms (vx 11× faster)       | 1,712 ms (vx 218× faster) |
+| 1,090 packages, 3,270 tasks      | vx         | Turborepo                  | Nx                     |
+| -------------------------------- | ---------- | -------------------------- | ---------------------- |
+| Cold build: time the runner adds | **2 s**    | 1 min 21 s (vx 34× faster) | 11 s (vx 4.7× faster)  |
+| Cold build: CPU burned           | **17 s**   | 21 s (vx 1.2× faster)      | 52 s (vx 3× faster)    |
+| Fully cached run                 | **393 ms** | 463 ms (vx 1.1× faster)    | 6.45 s (vx 16× faster) |
+| Overhead per package             | **2 ms**   | 74 ms (vx 34× faster)      | 10 ms (vx 4.7× faster) |
 
 vx N× faster: that tool takes N times as long as vx (theirs ÷ vx); N× slower: vx takes N times as long (vx ÷ theirs).
 
-Nx's column ran every task through npm run (~200 ms of CPU each; Turbo ran bun run), a harness fault since fixed; with bun, on a 4-core Linux box, Nx took 6m 59s cold and 4.50 s fully cached (vx 1.9× and 12× faster).
-
 Time added is the wall time over the tasks' own ideal schedule (3 min 38 s).
 Same graph, commands and concurrency: [how it is measured](https://vznjs.github.io/vx/benchmarks/).
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@vzn/vx",
-  "version": "0.0.497",
+  "version": "0.0.498",
   "description": "The fastest task runner and build cache for JavaScript monorepos.",
   "type": "module",
   "exports": {
@@ -21,10 +21,10 @@
     "node": ">=18"
   },
   "optionalDependencies": {
-    "@vzn/vx-linux-x64": "0.0.497",
-    "@vzn/vx-linux-arm64": "0.0.497",
-    "@vzn/vx-darwin-x64": "0.0.497",
-    "@vzn/vx-darwin-arm64": "0.0.497"
+    "@vzn/vx-linux-x64": "0.0.498",
+    "@vzn/vx-linux-arm64": "0.0.498",
+    "@vzn/vx-darwin-x64": "0.0.498",
+    "@vzn/vx-darwin-arm64": "0.0.498"
   },
   "dependencies": {
     "@anthropic-ai/sandbox-runtime": "^0.0.76"

package/src/exec/sandbox-runtime.ts

@@ -991,7 +991,16 @@ export function resolveSandboxConfig(
       (a.write ?? []).map((p) => {
         const real = resolve(p)
         assertWriteStaysHome(p, real, projectDir)
-        return real
+        // A directory a write glob matches is writable whole, as Linux's
+        // bind of the hit makes it: `<glob>/**` collapses below into the
+        // glob and its subtree. SRT compiles a glob as an exact regex, so
+        // `/tmp/pnpm-store-operation-locks-*/` covered the directory and
+        // not the lock file pnpm 12 opens inside it.
+        return process.platform !== 'linux' &&
+          !isMountableLiteral(real) &&
+          !/\/\*\*(?:\/\*)?$/.test(real)
+          ? `${real}/**`
+          : real
       }),
       walls,
       pending,