fix(ci): keep boundary output tails UTF-8 safe (#109167)

* fix(ci): keep boundary output tails UTF-8 safe

* fix(ci): preserve split UTF-8 output tails

Co-authored-by: qingminlong <qing.minlong@xydigit.com>

---------

Co-authored-by: Peter Steinberger <steipete@gmail.com>

scripts/run-additional-boundary-checks.mjs

@@ -183,6 +183,15 @@ export function formatCommand({ command, args }) {
   return [command, ...args].join(" ");
 }
 
+function decodeUtf8Tail(buffer) {
+  let start = 0;
+  while (start < buffer.length && (buffer[start] & 0b1100_0000) === 0b1000_0000) {
+    start += 1;
+  }
+  // Appends are complete JS strings; only a byte slice's leading boundary can be partial.
+  return buffer.subarray(start).toString("utf8");
+}
+
 /**
  * Keeps only the tail of noisy check output so failure logs stay bounded.
  */
@@ -197,7 +206,7 @@ export function createBoundedOutputBuffer(maxBytes = DEFAULT_OUTPUT_MAX_BYTES) {
     const textBytes = Buffer.byteLength(text);
     if (textBytes >= limit) {
       const buffer = Buffer.from(text);
-      const tail = buffer.subarray(buffer.length - limit).toString("utf8");
+      const tail = decodeUtf8Tail(buffer.subarray(buffer.length - limit));
       chunks.splice(0, chunks.length, tail);
       bytes = Buffer.byteLength(tail);
       truncated = true;
@@ -218,7 +227,7 @@ export function createBoundedOutputBuffer(maxBytes = DEFAULT_OUTPUT_MAX_BYTES) {
       }
 
       const buffer = Buffer.from(first);
-      const tail = buffer.subarray(overflow).toString("utf8");
+      const tail = decodeUtf8Tail(buffer.subarray(overflow));
       chunks[0] = tail;
       bytes = chunks.reduce((total, chunk) => total + Buffer.byteLength(chunk), 0);
       truncated = true;

test/scripts/run-additional-boundary-checks.test.ts

@@ -155,6 +155,21 @@ describe("run-additional-boundary-checks", () => {
     expect(output.read()).toBe("[output truncated to last 12 bytes]\nsecond-line\n");
   });
 
+  it("drops split UTF-8 prefixes when one chunk exceeds the output byte cap", () => {
+    const output = createBoundedOutputBuffer(5);
+    output.append("old😀new");
+
+    expect(output.read()).toBe("[output truncated to last 5 bytes]\nnew");
+  });
+
+  it("drops split UTF-8 prefixes when older buffered output overflows", () => {
+    const output = createBoundedOutputBuffer(5);
+    output.append("old😀");
+    output.append("new");
+
+    expect(output.read()).toBe("[output truncated to last 5 bytes]\nnew");
+  });
+
   it("parses and applies CI shard specs", () => {
     expect(parseShardSpec("2/4")).toEqual({ count: 4, index: 1, label: "2/4" });
     expect(parseShardSelection("2/4,3/4")).toEqual([
@@ -280,6 +295,30 @@ describe("run-additional-boundary-checks", () => {
     expect(result.output).toContain("timed out after 50ms");
   });
 
+  it("preserves UTF-8 split across process output chunks before trimming the byte tail", async () => {
+    const script = [
+      'const bytes = Buffer.from("old😀new");',
+      "process.stdout.write(bytes.subarray(0, 5));",
+      "setTimeout(() => process.stdout.end(bytes.subarray(5)), 20);",
+    ].join("");
+    const result = await runSingleCheck(
+      {
+        label: "split-utf8",
+        command: process.execPath,
+        args: ["-e", script],
+      },
+      {
+        checkTimeoutMs: 2_000,
+        cwd: process.cwd(),
+        env: process.env,
+        outputMaxBytes: 5,
+      },
+    );
+
+    expect(result.code).toBe(0);
+    expect(result.output).toBe("[output truncated to last 5 bytes]\nnew");
+  });
+
   it("clamps oversized check timers before scheduling", async () => {
     const result = await runSingleCheck(
       {