Merge pull request #21684 from calixteman/fix/quad-regex-headers

Trim the response headers with a backward scan

src/display/network.js

@@ -25,6 +25,7 @@ import {
   ensureResponseOrigin,
   extractFilenameFromHeader,
   getResponseOrigin,
+  trimHeadersEnd,
   validateRangeRequestCapabilities,
 } from "./network_utils.js";
 import { endRequests } from "./transport_stream.js";
@@ -208,9 +209,7 @@ class PDFNetworkStreamReader extends BasePDFStreamReader {
     const rawResponseHeaders = fullRequestXhr.getAllResponseHeaders();
     const responseHeaders = new Headers(
       rawResponseHeaders
-        ? rawResponseHeaders
-            .trimStart()
-            .replace(/[^\S ]+$/, "") // Not `trimEnd`, to keep regular spaces.
+        ? trimHeadersEnd(rawResponseHeaders.trimStart())
             .split(/[\r\n]+/)
             .map(x => {
               const [key, ...val] = x.split(": ");

src/display/network_utils.js

@@ -32,6 +32,17 @@ function createHeaders(isHttp, httpHeaders) {
   return headers;
 }
 
+// Trim the trailing whitespace of the raw response headers, but keep the
+// regular spaces (hence no `trimEnd`). Scanning backwards keeps this linear,
+// whereas a `$`-anchored regex is quadratic in the length of the run.
+function trimHeadersEnd(str) {
+  let end = str.length;
+  while (end > 0 && str[end - 1] !== " " && /\s/.test(str[end - 1])) {
+    end--;
+  }
+  return str.slice(0, end);
+}
+
 function getResponseOrigin(url) {
   // Notably, null is distinct from "null" string (e.g. from file:-URLs).
   return URL.parse(url)?.origin ?? null;
@@ -117,5 +128,6 @@ export {
   ensureResponseOrigin,
   extractFilenameFromHeader,
   getResponseOrigin,
+  trimHeadersEnd,
   validateRangeRequestCapabilities,
 };

test/unit/network_utils_spec.js

@@ -17,6 +17,7 @@ import {
   createHeaders,
   createResponseError,
   extractFilenameFromHeader,
+  trimHeadersEnd,
   validateRangeRequestCapabilities,
 } from "../../src/display/network_utils.js";
 import { ResponseException } from "../../src/shared/util.js";
@@ -386,4 +387,38 @@ describe("network_utils", function () {
       testCreateResponseError(new URL("https://foo.com/bar.pdf"), 0, false);
     });
   });
+
+  describe("trimHeadersEnd", function () {
+    it("removes the trailing whitespace", function () {
+      expect(trimHeadersEnd("a: 1\r\nb: 2\r\n")).toEqual("a: 1\r\nb: 2");
+      expect(trimHeadersEnd("a: 1\n\n")).toEqual("a: 1");
+      expect(trimHeadersEnd("a: 1\t\r\n")).toEqual("a: 1");
+    });
+
+    it("keeps the regular spaces", function () {
+      expect(trimHeadersEnd("a:  1  ")).toEqual("a:  1  ");
+      expect(trimHeadersEnd("a: 1\r\n  ")).toEqual("a: 1\r\n  ");
+      expect(trimHeadersEnd("  ")).toEqual("  ");
+    });
+
+    it("handles strings without trailing whitespace", function () {
+      expect(trimHeadersEnd("")).toEqual("");
+      expect(trimHeadersEnd("a: 1")).toEqual("a: 1");
+      expect(trimHeadersEnd("\r\na: 1")).toEqual("\r\na: 1");
+    });
+
+    it("handles a long run of whitespace efficiently", function () {
+      // Removing the run with a `$`-anchored regex is quadratic in its length,
+      // and a server controls how long the headers are.
+      const run = "\t".repeat(100000);
+
+      const startTime = performance.now();
+      // The run is trailing, hence removed.
+      expect(trimHeadersEnd(`a: 1${run}`)).toEqual("a: 1");
+      // The run is followed by a non-whitespace, hence kept: this is the case
+      // which a regex has to backtrack over.
+      expect(trimHeadersEnd(`a: 1${run}b`)).toEqual(`a: 1${run}b`);
+      expect(performance.now() - startTime).toBeLessThan(1000);
+    });
+  });
 });