Bug 17936 – "ASSERTION FAILED: false" (GetterSetterImp::toObject is called)

• Component: JavaScriptCore
• Status: RESOLVED FIXED

[Jesse Ruderman]

testkjs doesn't like this script.

b = 3;
this.__defineSetter__("a", function() {});
this.__defineSetter__("b", function() {});
delete a;
b.c;


A Debug build of testkjs exits, saying

ASSERTION FAILED: false
(/Users/jruderman/WebKit/JavaScriptCore/kjs/internal.cpp:191 virtual KJS::JSObject* KJS::GetterSetterImp::toObject(KJS::ExecState*) const)

A Release build of testjks says "TypeError: Null value" when perhaps it should say "TypeError: Undefined value", but nothing really bad happens.

[Mark Rowe (bdash)]

<rdar://problem/5320744>

[Mark Rowe (bdash)]

JSObject::defineSetter calls putDirect(propertyName, gs, GetterSetter), where "GetterSetter" is an attribute.  If a property exists with the name, this ends up inside PropertyMap::put at the following code:

            // Put a new value in an existing hash table entry.
            m_u.table->entries()[entryIndex - 1].value = value;
            // Attributes are intentionally not updated.

The problem here is the fact that we do not update the attribute of the property, which will lead to the PropertyMap returning an incorrect value for containsGettersOrSetters() when it is called after the setter for "a" in the test case is deleted.

[Cameron Zwarich (cpst)]

It seems that this was fixed by the refactoring of the getter and setter code in r36016, the revision that introduced polymorphic inline caching. I'll write a test so we can close this bug.

[Cameron Zwarich (cpst)]

Created attachment 23134 [details]
Proposed patch adding layout tests

[Cameron Zwarich (cpst)]

Landed in r36035.