mjhjmtu

Consider the following piece of code:

struct Foo 
 void* p; 
 Foo() : p(class Bar*)0 
;

Bar* bar;

The latest versions of GCC (8.2) and Clang (7.0.0) fail to compile it. So does ICC (19.0.1).

However MSVC (v19.16) compiles it cleanly.

The error from GCC is: error: 'Bar' does not name a type; did you mean 'char'?

Clang and ICC issue similar messages.

Conformance viewer for all four compilers at godbolt.

So which of the compiler(s) is correct as per the standard?

• [basic.lookup.elab] ... If the elaborated-type-specifier is introduced by
  the class-key and this lookup does not find a previously declared type-name ... the elaborated-type-specifier is a declaration that introduces the class-name as described in [basic.scope.pdecl]



• 
  

  [basic.scope.pdecl] - for an elaborated-type-specifier of the form

  
  
  

  class-key identifier

  
  
  

  if the elaborated-type-specifier is used in the decl-specifier-seq or parameter-declaration-clause of a
  function defined in namespace scope [does not apply because of scope], ... otherwise, except as a friend declaration, the identifier is declared in the
  smallest namespace or block scope that contains the declaration.

  
  

Now, the tricky bit. Is the member initialiser list "contained" in the scope of the constructor or not? If not, then the smallest block or namespace scope is the global namespace and the program would be well-formed. If yes, then the smallest scope is the block scope of the constructor, and thus the class would not be declared in the global scope.

As far as I can tell, there is no rule saying that mem-init-list is "contained in the block scope of the constructor". It is outside the curly brackets that delimit the scope. As such, the program is well-formed.

Mem-init-list is part of the constructor body [dcl.fct.def.general], but that body is neither a block scope nor a namespace scope, so it is not relevant to the rule in [basic.scope.pdecl].

This is about declaration rules and scope rules; (elaborated type specifier may also declare a class-name in an expression).

scope.declarative/3:
The names declared by a declaration are introduced into the scope in
which the declaration occurs, except that the presence of a friend
specifier, certain uses of the elaborated-type-specifier
([dcl.type.elab]), and using-directives ([namespace.udir]) alter this
general behavior.

a constructor has a scope (as eeroika's answer mentions), likewise anything declared therein. That is why this should be valid

struct Foo 
 void* p; 
 Foo() : p(class Bar*)0 
 Bar* bar;
 
;

https://godbolt.org/z/m3Tdle

But not:

struct Foo 
 void* p; 
 Foo() : p(class Bar*)0 
 //Scope of Bar only exists here
 
;

Bar* bar;

EDIT:
More details about this:

The so-called "forward declaration" in C++ is technically one of multiple "forms" of elaborated-type-specifier; And only two forms may introduce a name. (Emphasis mine)

The point of declaration of a class first declared in an
elaborated-type-specifier is as follows:

• 
  

  (7.1) for a declaration of the form

  
  
  

  class-key attribute-specifier-seq_opt identifier; //<-- Note the semi colon

  
  
  

  the identifier is declared to be a class-name in the scope that contains
  the declaration, otherwise

  
  


• 
  

  (7.2) for an elaborated-type-specifier of the form

  
  
  

  class-key identifier

  
  
  

  if the elaborated-type-specifier is used in the decl-specifier-seq or parameter-declaration-clause of a function defined in namespace scope, the identifier is declared as a class-name in the namespace that contains the declaration; otherwise, except as a friend declaration, the identifier is declared in the smallest namespace or block scope that contains the declaration. [ Note: These rules also apply within templates. —- end note ] [ Note:
  Other forms of elaborated-type-specifier do not declare a new name,
  and therefore must refer to an existing type-name. See
  [basic.lookup.elab] and [dcl.type.elab]. —- end note ]

  
  

Here's another interesting bit about it;

dcl.type.elab/1 An
attribute-specifier-seq shall not appear in an elaborated-type-specifier unless the latter is the sole constituent of
a declaration...

That is why this (below) is valid, See it here https://godbolt.org/z/IkmvGn;

void foo()
 void* n = (class Boo*)(0);
 Boo* b;


class [[deprecated("WTF")]] Mew;

But this, (below) is incorrect¹; See it here https://godbolt.org/z/8X1QKq;

void foo()
 void* n = (class [[deprecated("WTF")]] Boo*)(0);


class [[deprecated("WTF")]] Mew;

• 
  

  ¹Strangely, GCC accepts it, but gives this warning:

  
  
  

  attributes ignored on elaborated-type-specifier that is not a forward declaration [-Wattributes]
  

  
  

To see all the other "forms" of elaborated-type-specifier see dcl.type.elab

Edit 2

To double-check my understanding, I asked further, and @Simon Brand made me realise some edge cases, and that is partly hinted by basic.scope/declarative-4.note-2. But chiefly by the second quote in this answer basic.scope/pdecl-7.2

An elaborated-type-specifier within a function-parameter-scope will leak its class-name to the enclosing namespace (note: this is not class-scope).

This is valid https://godbolt.org/z/Fx5B83:

struct Foo 
 static void foo(void* = (class Bat*)0); //Leaks it past class-scope
;

void moo()
 Bat* m;

Even if you hide it in nested classes https://godbolt.org/z/40Raup:

struct Foo 
 class Moo
 class Mew 
 void foo(void* = (class Bat*)0); //
 ;
 ;
;

void moo()
 Bat* m;

... However, the name stops leaking at the closest namespace https://godbolt.org/z/YDljDo .

namespace zoo 
 struct Foo 
 class Moo
 class Mew 
 void foo(void* = (class Bat*)0);
 ;
 ;
 ;

void moo()
 zoo::Bat* m;

Finally,

mem-init-list is a compound-statement and does not have a function-parameter-scope; So, if you want to achieve, the leakage, do the declaration in the function-parameter-scope. See https://godbolt.org/z/CqejYS

struct Foo 
 void* p; 
 Foo(void* = (class Zoo*)(0)) 
 : p(class Bar*)0 
 Bar* bar;
 
;

Zoo* m; //Zoo* leaked
//Bar* n //Bar* did not leak