Do not propagate exception inside HTMLRewriter parser

Author: ryanking13

src/workerd/api/html-rewriter.c++

@@ -1263,8 +1263,14 @@ jsg::Ref<Response> HTMLRewriter::transform(jsg::Lock& js, jsg::Ref<Response> res
   //   after we know that nothing else (like invalid encoding) could cause an exception.
 
   // Drive and flush the parser asynchronously.
-  ioContext.addTask(ioContext.waitForDeferredProxy(
-      KJ_ASSERT_NONNULL(maybeInput)->pumpTo(js, kj::mv(rewriter), true)));
+  ioContext.addTask(
+      ioContext
+          .waitForDeferredProxy(KJ_ASSERT_NONNULL(maybeInput)->pumpTo(js, kj::mv(rewriter), true))
+          .catch_([](kj::Exception&& e) {
+    // Errors in pumpTo() are already propagated to the destination stream. We don't want to
+    // throw them from here since it'll cause an uncaught exception to be reported via taskFailed(),
+    // which would poison the IoContext even though the application may have handled the error.
+  }));
 
   // TODO(soon): EW-2025 Make Rewriter a proper wrapper object and put it in hidden property on the
   //   response so the GC can find the handlers which Rewriter co-owns.

src/workerd/api/tests/htmlrewriter-test.js

@@ -908,6 +908,64 @@ export const exceptionPropagation = {
   },
 };
 
+// handled HTMLRewriter errors must not poison the IoContext.
+export const continueAfterException = {
+  async test() {
+    const errorResponse = new HTMLRewriter()
+      .on('*', {
+        element() {
+          throw new Error('intentional error for testing');
+        },
+      })
+      .transform(new Response('<div>test</div>'));
+
+    await rejects(errorResponse.text(), {
+      message: 'intentional error for testing',
+    });
+
+    const successResponse = new HTMLRewriter()
+      .on('div', {
+        element(el) {
+          el.setInnerContent('success');
+        },
+      })
+      .transform(new Response('<div>original</div>'));
+
+    strictEqual(await successResponse.text(), '<div>success</div>');
+  },
+};
+
+export const continueAfterException2 = {
+  async test() {
+    const errorResponse = new HTMLRewriter()
+      .on('*', {
+        element() {
+          throw new Error('intentional error for pipeTo testing');
+        },
+      })
+      .transform(new Response('<div>test</div>'));
+
+    const { writable } = new TransformStream();
+
+    await rejects(errorResponse.body.pipeTo(writable), {
+      message: 'intentional error for pipeTo testing',
+    });
+
+    const successResponse = new HTMLRewriter()
+      .on('div', {
+        element(el) {
+          el.setInnerContent('success after pipeTo');
+        },
+      })
+      .transform(new Response('<div>original</div>'));
+
+    strictEqual(
+      await successResponse.text(),
+      '<div>success after pipeTo</div>'
+    );
+  },
+};
+
 export const sameToken = {
   async test() {
     const obj = {};