[ES6] Arrow function syntax. Using 'super' in arrow function that declared out of the class should lead to Syntax error

• Component: JavaScriptCore
• Status: RESOLVED FIXED

[GSkachkov]

Using 'super' in arrow function that declared out of the class should lead to SyntaxError.
For instance:
var G = class G extends A {
     getParentValueWithError() {
       var f =  function () {
         return () => super.getValue();
       };
       return f()();
     }
 };

(new G()).getParentValueWithError(); 

Should be SyntaxError instead of  TypeError: undefined is not an object (evaluating 'super.getValue')

[GSkachkov]

Created attachment 269925 [details]
Patch

Patch

[Saam Barati]

Comment on attachment 269925 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=269925&action=review

r=me with comments

> Source/JavaScriptCore/parser/Parser.h:880
> +    ScopeRef currentNotArrowfunctionFunctionScope()

I propose calling this: "closestParentNonArrowFunctionNonLexicalScope"
We really don't need it to return a function scope (and it doesn't in some cases).
It just needs to return the closest non-arrow-function function or the global scope.

> Source/JavaScriptCore/parser/Parser.h:883
> +        ASSERT(i < m_scopeStack.size());

If you're making this assertion you should also assert that m_scopeStack.size() is non zero.

> Source/JavaScriptCore/parser/Parser.h:886
> +            ASSERT(i < m_scopeStack.size());

Style: I don't think this assert adds much here

[GSkachkov]

Created attachment 270010 [details]
Patch

Fix comments

[Saam Barati]

Should this be reviewed?
I'm confused b/c the r-

[GSkachkov]

(In reply to comment #4)
> Should this be reviewed?
> I'm confused b/c the r-

Previous patch does not cover case with eval("super()"). I'm finishing the new patch that cover this case. I hope that I'll manage to upload it today.

[GSkachkov]

Created attachment 270653 [details]
Patch

New patch that cover using super in eval

[GSkachkov]

Created attachment 270654 [details]
Patch

Fix build

[Saam Barati]

Comment on attachment 270654 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=270654&action=review

> Source/JavaScriptCore/ChangeLog:20
> +        Invoking super()/super in two or more evals will be part of the issue https://bugs.webkit.org/show_bug.cgi?id=153864

what kind of error should this be if it's not a SyntaxError?

> Source/JavaScriptCore/bytecode/EvalCodeCache.h:70
> +                    wrappedCodeBlockDerivedContextType = superBinding == SuperBinding::Needed
> +                        ? DerivedContextType::DerivedMethodContext
> +                        : derivedContextType;

style: make this one line or add "{}" to the else statement.

> Source/JavaScriptCore/bytecode/ExecutableInfo.h:67
> +    DerivedContextType wrappedCodeBlocContextType() const { return static_cast<DerivedContextType>(m_wrappedCodeBlocContextType); }

typo:
wrappedCodeBlocContextType => wrappedCodeBlockContextType

> Source/JavaScriptCore/parser/Parser.cpp:2070
> +    if (!m_lexer->isReparsingFunction()) {

Is this needed or just an optimization?

> Source/JavaScriptCore/parser/Parser.h:881
> +    // Find the closest not 'arrow function' function scope.

comment not needed.

> Source/JavaScriptCore/tests/stress/arrowfunction-lexical-bind-superproperty.js:-131
> -     getParentValueWithError() {
> -         var f =  function () {
> -             return () => super.getValue();
> -         };
> -         return f();
> -     }

you could keep this if you turned this into an "return eval(blah)"

[GSkachkov]

Created attachment 270826 [details]
Patch

Fix comments

[GSkachkov]

Comment on attachment 270654 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=270654&action=review

>> Source/JavaScriptCore/ChangeLog:20
>> +        Invoking super()/super in two or more evals will be part of the issue https://bugs.webkit.org/show_bug.cgi?id=153864
> 
> what kind of error should this be if it's not a SyntaxError?

I've made small change in this sentence in the last patch.  
It should be SyntaxError, but my patch only partly resolve this issue. I had to fix case with 'eval("super")' because some of the  arrow function tests invoke eval('super'). During the  additional testing I found out that my current patch does not cover case when we call eval('eval('super()')'), but I decided that it should not be part of this patch and created the issue

>> Source/JavaScriptCore/bytecode/EvalCodeCache.h:70
>> +                        : derivedContextType;
> 
> style: make this one line or add "{}" to the else statement.

Done

>> Source/JavaScriptCore/bytecode/ExecutableInfo.h:67
>> +    DerivedContextType wrappedCodeBlocContextType() const { return static_cast<DerivedContextType>(m_wrappedCodeBlocContextType); }
> 
> typo:
> wrappedCodeBlocContextType => wrappedCodeBlockContextType

Done

>> Source/JavaScriptCore/parser/Parser.cpp:2070
>> +    if (!m_lexer->isReparsingFunction()) {
> 
> Is this needed or just an optimization?

I added comment in the last patch, I hope that now it is more clear:
    // It unncecessary to check of using super during reparsing one more time. Also it can lead to syntax error
    // in case of arrow function becuase during reparsing we don't know that parse arrow function
    // inside of the constructor or method

>> Source/JavaScriptCore/parser/Parser.h:881
>> +    // Find the closest not 'arrow function' function scope.
> 
> comment not needed.

Removed

>> Source/JavaScriptCore/tests/stress/arrowfunction-lexical-bind-superproperty.js:-131
>> -     }
> 
> you could keep this if you turned this into an "return eval(blah)"

I've moved this test case to the LayoutTests/js/script-tests/arrowfunction-syntax-errors.js where I'm trying  to collect all case that are related to the syntax errors. 
See last case in 66 line in this file.

[GSkachkov]

Created attachment 270851 [details]
Patch

Remove all changes that are related to the using eval in arrow function. This will be implemented in separate patch

[Saam Barati]

Comment on attachment 270851 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=270851&action=review

> Source/JavaScriptCore/parser/Parser.h:886
> +        // When reaching the top level scope (it can be non function scope), we return it.

In a later patch, I'd remove this comment, or just say somethin like:
Note: it's okay to return the top most scope because it will be the global scope

[WebKit Commit Bot]

Comment on attachment 270851 [details]
Patch

Clearing flags on attachment: 270851

Committed r196261: <http://trac.webkit.org/changeset/196261>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[GSkachkov]

*** Bug 153977 has been marked as a duplicate of this bug. ***