[Parser] Remove Variable::is_possibly_eval.

src/parsing/parser.cc

Index: src/parsing/parser.cc
diff --git a/src/parsing/parser.cc b/src/parsing/parser.cc
index 0d16469f3190e582d399752cf5387926922d268a..90c162e38ea6efa3409365a2d5a0f2f8c6ac7707 100644
--- a/src/parsing/parser.cc
+++ b/src/parsing/parser.cc
@@ -788,7 +788,7 @@ Expression* ParserTraits::GetIterator(Expression* iterable,
       factory->NewProperty(iterable, iterator_symbol_literal, pos);
   Zone* zone = parser_->zone();
   ZoneList<Expression*>* args = new (zone) ZoneList<Expression*>(0, zone);
-  return factory->NewCall(prop, args, pos);
+  return factory->NewCall(prop, args, pos, Call::NOT_EVAL);
 }
 
 
@@ -3199,7 +3199,7 @@ Expression* Parser::BuildIteratorNextResult(Expression* iterator,
   ZoneList<Expression*>* next_arguments =
       new (zone()) ZoneList<Expression*>(0, zone());
   Expression* next_call =
-      factory()->NewCall(next_property, next_arguments, pos);
+      factory()->NewCall(next_property, next_arguments, pos, Call::NOT_EVAL);
   Expression* result_proxy = factory()->NewVariableProxy(result);
   Expression* left =
       factory()->NewAssignment(Token::ASSIGN, result_proxy, next_call, pos);
@@ -5558,12 +5558,14 @@ Expression* Parser::CloseTemplateLiteral(TemplateLiteralState* state, int start,
     Expression* call_site = factory()->NewCallRuntime(
         Context::GET_TEMPLATE_CALL_SITE_INDEX, args, start);
 
+    Call::PossiblyEval is_possibly_eval = CheckPossibleEvalCall(tag, scope());

[rmcilroy, 2016/08/11 21:27:59]

This call updates the scopes to record the eval call - I'm not sure if this is
safe here but doesn't seem to cause any issues on the test. Let me know if we
should do something else.

[adamk, 2016/08/11 21:42:31]

There used to be a call to CheckPossibleEvalCall here, but I removed it in
https://codereview.chromium.org/1580463003/. I wish I'd written a better CL
description, though, as I can't remember what caused me to make that change. My
guess is I was worried about the fact that this call only appears in the parser,
not preparser. So a function could get different "uses eval" bits when
preparsing than when doing full parsing. If you're going to add it back in, I
think it ought to also get set when preparsing.

Does anything fail if you leave this out, I'm curious? As noted in my CL, using
eval as a template tag is harmless: it's a call to a runtime function that
returns its argument.

[rmcilroy, 2016/08/12 09:18:20]

The reason I added this is that when I took it out the behavior of
Call::is_possibly_eval was different from Variable::is_possibly_eval for the
tests in mjsunit/es6/template.js. Taking it out doesn't seem to have anything to
fail though, so I've taken it out now.

+
     // Call TagFn
     ZoneList<Expression*>* call_args =
         new (zone()) ZoneList<Expression*>(expressions->length() + 1, zone());
     call_args->Add(call_site, zone());
     call_args->AddAll(*expressions, zone());
-    return factory()->NewCall(tag, call_args, pos);
+    return factory()->NewCall(tag, call_args, pos, is_possibly_eval);
   }
 }
 
@@ -6276,7 +6278,8 @@ Expression* ParserTraits::RewriteYieldStar(
     Expression* input_proxy = factory->NewVariableProxy(var_input);
     auto args = new (zone) ZoneList<Expression*>(1, zone);
     args->Add(input_proxy, zone);
-    Expression* call = factory->NewCall(next_property, args, nopos);
+    Expression* call =
+        factory->NewCall(next_property, args, nopos, Call::NOT_EVAL);
     Expression* output_proxy = factory->NewVariableProxy(var_output);
     Expression* assignment =
         factory->NewAssignment(Token::ASSIGN, output_proxy, call, nopos);