[regexp] Experimental support for regexp named captures

src/runtime/runtime-regexp.cc

Index: src/runtime/runtime-regexp.cc
diff --git a/src/runtime/runtime-regexp.cc b/src/runtime/runtime-regexp.cc
index a8133d349538e4f0742f25042653b793d0f8b877..01c6f31c0b77d2c451623925efad07f6323f7adf 100644
--- a/src/runtime/runtime-regexp.cc
+++ b/src/runtime/runtime-regexp.cc
@@ -797,7 +797,6 @@ RUNTIME_FUNCTION(Runtime_RegExpSource) {
   return regexp->source();
 }
 
-
 RUNTIME_FUNCTION(Runtime_RegExpConstructResult) {
   HandleScope handle_scope(isolate);
   DCHECK(args.length() == 3);
@@ -1014,6 +1013,38 @@ RUNTIME_FUNCTION(Runtime_RegExpExecReThrow) {
   return isolate->ReThrow(exception);
 }
 
+RUNTIME_FUNCTION(Runtime_RegExpResultAttachNamedCaptures) {
+  HandleScope handle_scope(isolate);
+  DCHECK(args.length() == 2);
+  CONVERT_ARG_CHECKED(JSArray, result, 0);
+  CONVERT_ARG_CHECKED(JSRegExp, regexp, 1);
+
+  if (regexp->TypeTag() != JSRegExp::IRREGEXP ||

[Yang, 2016/06/13 10:54:53]

We don't actually have to check this, right? We can assert this in the
then-block, but for non-IRREGEXP the capture name map should not be a fixed
array anyways.

[jgruber, 2016/06/13 13:10:00]

I think we need to check, as ATOM JSRegExps have a shorter data array (see
Factory::SetRegExpAtomData) and we'd have an out of bounds access.

+      !regexp->DataAt(JSRegExp::kIrregexpCaptureNameMapIndex)->IsFixedArray())

[Yang, 2016/06/13 10:54:53]

Please add brackets around the then-block.

[jgruber, 2016/06/13 13:10:00]

Done.

+    return isolate->heap()->undefined_value();
+
+  FixedArray* captures =
+      FixedArray::cast(regexp->DataAt(JSRegExp::kIrregexpCaptureNameMapIndex));
+  DCHECK(captures->length() > 0);
+
+  Handle<JSArray> result_handle = handle(result, isolate);
+
+  for (int i = 0; i < captures->length(); i += 2) {
+    Handle<String> name = handle(String::cast(captures->get(i)), isolate);
+    const int ix = Smi::cast(captures->get(i + 1))->value();

[Yang, 2016/06/13 10:54:53]

Accessing captures like this is not GC-safe. You need to put captures into a
handle.

[jgruber, 2016/06/13 13:10:00]

Done.

+
+    MaybeHandle<Object> existing = Object::GetProperty(result_handle, name);
+    if (!existing.ToHandleChecked()->IsUndefined()) {
+      continue;  // Skip conflicts.

[Yang, 2016/06/13 10:54:53]

I don't think this can actually happen unless the property is on the array
prototype. Let's not care about that and just always add the property. We can
still change this once the spec takes shape.

[jgruber, 2016/06/13 13:10:00]

This happens for regexp properties 'index' and 'input'. There's two test cases
covering this. If we remove this check, those properties could be overwritten.

[Yang, 2016/06/13 13:38:00]

I see. Can we use JSReceiver::HasOwnProperty for this?

[jgruber, 2016/06/14 07:53:12]

Sounds good, will change this when we look at adding properties for named
captures.

+    }
+
+    Handle<Object> value =
+        Object::GetElement(isolate, result_handle, ix).ToHandleChecked();
+    JSObject::AddProperty(result_handle, name, value, NONE);
+  }
+
+  return isolate->heap()->undefined_value();
+}
 
 RUNTIME_FUNCTION(Runtime_IsRegExp) {
   SealHandleScope shs(isolate);