Enhancement: diagnose declaration or definition of reserved identifiers to avoid undefined behavior

[pmor13]

This code:

void __x(void);

compiled with:

-std=c11 -pedantic -Wall -Wextra

leads to:

<nothing>

instead of:

warning: declaration of reserved identifier `__x`

Per C11, 7.1.3 Reserved identifiers, 2:

If the program declares or defines an identifier in a context in which it is reserved (other than as allowed by 7.1.4), or defines a reserved identifier as a macro name, the behavior is undefined.

Hence, the enhancement is: diagnose declaration or definition of reserved identifiers (all possible cases, see C11, 7.1.3 Reserved identifiers) to avoid undefined behavior.

@AaronBallman Any comments / ideas / thoughts?

[AaronBallman]

This is what -Wreserved-identifier already does (there are some edge cases though, like we miss reserved macro names defined on the command line): https://godbolt.org/z/f3z7aPKTs

[pmor13]

Wow! Excellent! Does it cover all possible use cases? Why -Wreserved-identifier is not part of -Wall or -Wextra? Yes, there is -Weverything, however, it is not recommended for the end users.

[AaronBallman]

Does it cover all possible use cases?

It gets most of them, but not all (like I mentioned above, we miss macros defined via -D on the command line).

Why -Wreserved-identifier is not part of -Wall or -Wextra?

Chattiness of the diagnostic in practice. The real trouble is just how many bajillions of header guards are out there that do #define __MY_FANCY_HEADER_H__ without realizing that's a reserved identifier. Another big problem are third-party libraries that hide implementation details behind things like namespace __details. Those sorts of situations are valid to diagnose of course (they are hitting the UB of using a reserved identifier), but they're also the kinds of situations where the user getting the diagnostic can't actually address the diagnostic because it's triggered from code they don't own.

[pmor13]

@AaronBallman A quick relevant question.

In https://eel.is/c++draft/lex.name I see:

Each identifier that contains a double underscore __ or begins with an underscore followed by an uppercase letter is reserved to the implementation for any use.

Why "contains" and not "begins"? Where this "contains" came from? Is it a typo?

Here is C:

All identifiers that begin with an underscore and either an uppercase letter or another
underscore are always reserved for any use.

[rymiel]

It's not a typo. C++ is allowed to pick the rules it wants for the language, that's what the standard outlines. It was simply chosen to make those rules broader in the C++ standard

[AaronBallman]

Why "contains" and not "begins"? Where this "contains" came from? Is it a typo?

It's not a typo, C++ really does mean "contains". The first C++ compiler (CFront) would name mangle namespaces using __, so:

namespace Foo {
struct Bar {
  int i;
};
}

would mangle as Foo__Bar. When C++ was standardized, it had to go with "contains" instead of "begins with" so that CFront could continue to be a conforming implementation. Even though CFront is no longer a thing, some C++ implementations still support this mangling format.

[pmor13]

Wow! Thanks. I didn't know that Foo__Bar is reserved in C++. TIL.

[pmor13]

@EugeneZelenko I think that this issue can be closed as "not a problem". I didn't know that -Wreserved-identifier exist.

[EugeneZelenko]

@zygoloid: I added relevant label, since workaround exists already.