Bug 18649 – SQUIRRELFISH: correctly handle exceptions in eval code

• Component: JavaScriptCore
• Status: RESOLVED FIXED

[Cameron Zwarich (cpst)]

Exceptions are broken in eval code. This leads to assertion failures of the following form when running JavaScriptCore tests:

ASSERTION FAILED: it != end
(./VM/Machine.cpp:431 bool KJS::Machine::unwindCallFrame(KJS::Register**, const KJS::Instruction*&, KJS::CodeBlock*&, KJS::JSValue**&, KJS::ScopeChainNode*&, KJS::Register*&))

ASSERTION FAILED: addressOffset < instructions.size()
(/Users/Cameron/sf/JavaScriptCore/VM/CodeBlock.cpp:541 bool KJS::CodeBlock::getHandlerForVPC(const KJS::Instruction*, KJS::Instruction*&, int&))

[Cameron Zwarich (cpst)]

The it != end assertion failure is hit when throwing an exception from eval in a global context. Global code is a special case at the beginning, because no unwind is necessary, and the rest of the function deals with the case of function code. If the eval is in a global context, then the end of the scope chain is also the top, hitting the assertion.

The addressOffset < instructions.size() assertion failure is hit when throwing an exception from eval in a function context. The top of the scope chain is an activation object, so Machine::unwindCallFrame() thinks it has succeeded when it really hasn't, passing the buck onto CodeBlock::getHandlerForVPC().

[Oliver Hunt]

Partial fix Committed r32331

[Oliver Hunt]

Created attachment 20743 [details]
Make eval setup a "native" callframe

[Geoffrey Garen]

Comment on attachment 20743 [details]
Make eval setup a "native" callframe

Need to fix the other eval case. Otherwise, code looks good. Clearing review flag.

[Geoffrey Garen]

Comment on attachment 20743 [details]
Make eval setup a "native" callframe

Sorry, the second version just calls the first, so this looks good.

[Oliver Hunt]

Committed r32361