C++: update test output following extractor template fix

[nickrolfe]

No description provided.

[jbj]

It seems fine to me that a type argument to a template is normalised slightly to remove a pointless const. I suppose the same will now happen for a const int argument.

What happens if there's also a G<const int *> instantiation? Does that become a separate class but with the same template argument? Or will it be the same class as G<const int *const> and have two names? And how many bodies?

[nickrolfe]

If you add a G<const int *> instantiation you do indeed get a separate class with the same template argument.

[jbj]

That's probably the least bad of the two possibilities.

[dave-bartolomeo]

Wait, what's the extractor change that causes this? G<const int*const> and G<const int*> are two different specializations of the template in C++. It's function parameters that ignore top-level const, not template arguments.

For a real-world example, std::unique_ptr<int> is an int* that gets deleted when it goes out of scope, and std::unique_ptr<const int> is a const int* that gets deleted when it goes out of scope.

[nickrolfe]

This is not a desirable change, it's just an unfortunate side-effect of us struggling to work around an issue extracting instantiations like a<decltype(f)> - see internal PR 29618.

[dave-bartolomeo]

OK, it sounds like we're still keeping them as separate classes, but with top-level const incorrectly stripped from the template argument. That's still not good, but it's probably less likely to cause problems than having them as the same class. Is there a Jira ticket still open to figure out the right extractor fix post-1.19?

Objection removed

[jbj]

Should this be closed now?

[nickrolfe]

Yes, @ian-semmle figured out the correct extractor fix that keeps the correct output for this test.