TextDecoder: throw ERR_STRING_TOO_LONG instead of returning an empty string

src/jsc/bindings/bindings.cpp

@@ -3367,15 +3367,19 @@ JSC::EncodedJSValue JSC__JSValue__fromEntries(JSC::JSGlobalObject* globalObject,
 
     if (!clone) {
         for (size_t i = 0; i < initialCapacity; ++i) {
+            auto* value = Zig::toJSStringGC(values[i], globalObject);
+            RETURN_IF_EXCEPTION(scope, {});
             object->putDirect(
                 vm, JSC::PropertyName(JSC::Identifier::fromString(vm, Zig::toString(keys[i]))),
-                Zig::toJSStringGC(values[i], globalObject), 0);
+                value, 0);
             RETURN_IF_EXCEPTION(scope, {});
         }
     } else {
         for (size_t i = 0; i < initialCapacity; ++i) {
+            auto* value = Zig::toJSStringGC(values[i], globalObject);
+            RETURN_IF_EXCEPTION(scope, {});
             object->putDirect(vm, JSC::PropertyName(Zig::toIdentifier(keys[i], globalObject)),
-                Zig::toJSStringGC(values[i], globalObject), 0);
+                value, 0);
             RETURN_IF_EXCEPTION(scope, {});
         }
     }
@@ -3700,7 +3704,7 @@ __attribute__((__always_inline__)) VirtualMachine* JSC__JSGlobalObject__bunVM(JS
 
 JSC::EncodedJSValue ZigString__toValueGC(const ZigString* arg0, JSC::JSGlobalObject* arg1)
 {
-    return JSC::JSValue::encode(JSC::jsString(arg1->vm(), Zig::toStringCopy(*arg0)));
+    return JSC::JSValue::encode(Zig::toJSStringGC(*arg0, arg1));
 }
 
 void JSC__JSValue__toZigString(JSC::EncodedJSValue JSValue0, ZigString* arg1, JSC::JSGlobalObject* arg2)

src/jsc/bindings/helpers.h

@@ -6,6 +6,7 @@
 
 #include <JavaScriptCore/Error.h>
 #include <JavaScriptCore/Exception.h>
+#include <JavaScriptCore/ExceptionHelpers.h>
 #include <JavaScriptCore/Identifier.h>
 #include <JavaScriptCore/JSGlobalObject.h>
 #include <JavaScriptCore/JSString.h>
@@ -25,6 +26,14 @@ class GlobalObject;
 extern "C" size_t Bun__stringSyntheticAllocationLimit;
 extern "C" const char* Bun__errnoName(int);
 
+namespace Bun {
+namespace ERR {
+// Defined in ErrorCode.cpp; forward-declared here because ErrorCode.h cannot
+// be included from this header.
+JSC::EncodedJSValue STRING_TOO_LONG(JSC::ThrowScope& throwScope, JSC::JSGlobalObject* globalObject);
+}
+}
+
 namespace Zig {
 
 // 8 bit byte
@@ -191,6 +200,11 @@ static const WTF::String toStringCopy(ZigString str)
         return WTF::String::fromUTF8ReplacingInvalidSequences(std::span { untag(str.ptr), str.len });
     }
 
+    // This will fail if the string is too long. Let's make it explicit instead of an ASSERT.
+    if (str.len > Bun__stringSyntheticAllocationLimit || str.len > WTF::String::MaxLength) [[unlikely]] {
+        return {};
+    }
+
     if (isTaggedUTF16Ptr(str.ptr)) {
         std::span<char16_t> out;
         auto impl = WTF::StringImpl::tryCreateUninitialized(str.len, out);
@@ -240,9 +254,25 @@ static const JSC::JSString* toJSString(ZigString str, JSC::JSGlobalObject* globa
     return JSC::jsOwnedString(global->vm(), toString(str));
 }
 
+// Copies `str` into a GC-managed JSString. Throws ERR_STRING_TOO_LONG (or an
+// out-of-memory error) and returns nullptr when the string cannot be created.
 static JSC::JSString* toJSStringGC(ZigString str, JSC::JSGlobalObject* global)
 {
-    return JSC::jsString(global->vm(), toStringCopy(str));
+    auto wtfString = toStringCopy(str);
+    if (wtfString.isNull() && str.len > 0) [[unlikely]] {
+        // toStringCopy returns a null string when the input is over the string
+        // length limit or when allocation fails; surface an error instead of
+        // silently returning an empty string.
+        auto scope = DECLARE_THROW_SCOPE(global->vm());
+        size_t maxLength = std::min(Bun__stringSyntheticAllocationLimit, static_cast<size_t>(WTF::String::MaxLength));
+        if (str.len > maxLength) {
+            Bun::ERR::STRING_TOO_LONG(scope, global);
+        } else {
+            JSC::throwOutOfMemoryError(global, scope);
+        }
+        return nullptr;
+    }
+    return JSC::jsString(global->vm(), WTF::move(wtfString));
 }
 
 static const ZigString ZigStringEmpty = ZigString { (unsigned char*)"", 0 };

test/js/web/encoding/text-decoder.test.js

@@ -947,3 +947,40 @@
   },
   30_000,
 ); // 3072 decodes + repeated Bun.gc(true) take ~4s under ASAN.
+
+describe("TextDecoder.decode above the maximum string length", () => {
+  const expected = "threw:ERR_STRING_TOO_LONG:Cannot create a string longer than 2147483647 characters";
+  const script = size => `
+    try {
+      const s = new TextDecoder().decode(new Uint8Array(${size}));
+      console.log("returned:" + s.length);
+    } catch (e) {
+      console.log("threw:" + e.code + ":" + e.message);
+    }`;
+
+  // A zeroed input is all-ASCII, so the decoded string's length equals the
+  // input length. 2**31 exceeds WTF::StringImpl's maximum length (2^31-1);
+  // decode() used to return "" silently instead of throwing. The input is
+  // never written, so the 2 GiB Uint8Array stays untouched zero pages.
+  it("throws ERR_STRING_TOO_LONG for a 2**31 byte ASCII input", async () => {
+    await using proc = Bun.spawn({
+      cmd: [bunExe(), "-e", script("2 ** 31")],
+      env: bunEnv,
+    });
+    const [stdout, exitCode] = await Promise.all([proc.stdout.text(), proc.exited]);
+    expect(stdout.trim()).toBe(expected);
+    expect(exitCode).toBe(0);
+  }, 30_000); // scans 2 GiB of input; slow under debug + ASAN.
+
+  // Same path, exercised cheaply via the synthetic allocation limit so the
+  // guard is covered without a 2 GiB allocation.
+  it("throws ERR_STRING_TOO_LONG above the synthetic allocation limit", async () => {
+    await using proc = Bun.spawn({
+      cmd: [bunExe(), "-e", script("160 * 1024 * 1024")],
+      env: { ...bunEnv, BUN_FEATURE_FLAG_SYNTHETIC_MEMORY_LIMIT: "134217728" },
+    });
+    const [stdout, exitCode] = await Promise.all([proc.stdout.text(), proc.exited]);
+    expect(stdout.trim()).toBe(expected);
+    expect(exitCode).toBe(0);
+  });
+});