[clang] Diagnose potential size confusion with VLA params [contd.] (PR #181550)

[Martin Uecker via cfe-commits]

uecker wrote:

> > > > Note that this confusion in principle applies to any use of any identifier, not just constants but also others in sizeof / typeof.
> > > 
> > > 
> > > It does, but I think `-Wshadow` is the general purpose diagnostic used to find those kinds of potential naming surprises.
> > > I see the purpose to this diagnostic as being more specific to situations where potential changes to lookup rules (allowing for late parsing of function parameter names) might cause behavioral changes for users. But I can see why that might be too narrow of a focus.
> > > Where do you see the line being drawn between the two given that they're related?
> > 
> > 
> > I don't know. I hope we never see such breaking changes. If make breaking changes, the exact warning would depend on the exact nature of the changes. But I also think the need for the warning already shows that changing the lookup rules is a bad idea.
> 
> FWIW, `-Wshadow` already warns about all these cases.

I think the existing -Wshadow is basically useless and I am not sure who even uses it.   I think the model to look at would be GCC's -Wchanges-meaning which warns about the only similar case I know of, which is in C++ where use before redeclaration as a class member is IFNDR: https://godbolt.org/z/ee8KejbGa

> 
> That said, I think adding a specific flag for array parameters would be valuable because those are the specific scenarios we want to explore as potential constraint violations, with the possibility of changing the behavior in the future.
> 
> @AaronBallman By the way, should the name be something like `-Wshadow-array-parameter-potential-size-confusion` since we're also covering const sizes?



https://github.com/llvm/llvm-project/pull/181550