[analyzer] Should we invalidate the `this` pointer?

Clang Frontend Static Analyzer
#1

Hi all,

I recently encountered a assertion failure as shown below.

Assertion !InitValWithAdjustments.getAs() || Loc::isLocType(Result->getType()) || Result->getType()->isMemberPointerType()’ failed`

The code that will trigger this assertion failed.

#2

This assertion is pretty fundamental. Invalidation, if done correctly, should not have triggered it - after all, invalidation could occur for any other reason, not necessarily because of loop widening.

Invalidation of this-region contents (that is, not of the CXXThisObjectRegion of the current stack frame, but of the actual this-region which is a pointee of the CXXThisObjectRegion) sounds reasonable if the region is modified within the loop - which is going to often be the case.

#3

Hi Artem,

Thank you for your explanation1 You are right, invalidation of the region
contents of the class object is correct and common. However this pointer i
s no-lvalue and it’s a prvalue expression in c++17. IMHO, invalidation of CXXThisObjectRegion is incorrect and violates the C++ standard.

Given the code below:

#4

Yeah, this looks pretty broken. One does not simply overwrite his this-pointer using valid C++. Feel free to fix :slight_smile: