clang doesn't diagnose some invalid code

[llvmbot]

Bugzilla Link	14375
Version	trunk
OS	All
Reporter	LLVM Bugzilla Contributor
CC	@DougGregor,@zygoloid

Extended Description

http://gcc.gnu.org/ml/gcc/2012-11/msg00210.html and followups: GCC produces an error with diagnostic for the invalid code snippets posted below, while clang 3.0 and ToT (tested with r168286) don't, even with e.g. -Weverything.

According to the followups, compilers are not required to produce diagnostics for that situation, but it would nevertheless be nice for users to have them :)
The followups mention 3.3.6/1 and [dcl.typedef]/6.

===== begin snippet 1 (original testcase) =====
class A {};
class B : public A {

typedef A super;

public:

class X {};

};

class C : public B {

typedef B super;

class X : public super::X {

   typedef super::X super;
};

};
===== end snippet 1 =====

===== begin snippet 2 (reduced testcase in first followup) =====
typedef struct { typedef int type; } s1;
struct S2 { s1::type s1; };
===== end snippet 2 =====

[llvmbot]

typedef struct { } s1;
struct S2 { s1 s1; };

[llvmbot]

struct S1 { };
struct S2 { S1 S1; };

[llvmbot]

@llvm/issue-subscribers-clang-frontend

Author: None (llvmbot)

| | | | --- | --- | | Bugzilla Link | [14375](https://llvm.org/bz14375) | | Version | trunk | | OS | All | | Reporter | LLVM Bugzilla Contributor | | CC | @DougGregor,@zygoloid |

Extended Description

http://gcc.gnu.org/ml/gcc/2012-11/msg00210.html and followups: GCC produces an error with diagnostic for the invalid code snippets posted below, while clang 3.0 and ToT (tested with r168286) don't, even with e.g. -Weverything.

According to the followups, compilers are not required to produce diagnostics for that situation, but it would nevertheless be nice for users to have them :)
The followups mention 3.3.6/1 and [dcl.typedef]/6.

===== begin snippet 1 (original testcase) =====
class A {};
class B : public A {

typedef A super;

public:

class X {};

};

class C : public B {

typedef B super;

class X : public super::X {

   typedef super::X super;
};

};
===== end snippet 1 =====

===== begin snippet 2 (reduced testcase in first followup) =====
typedef struct { typedef int type; } s1;
struct S2 { s1::type s1; };
===== end snippet 2 =====

[wheatman]

confirming that this is still the case in post 17 trunk(5a56f00)
https://godbolt.org/z/zcMej1veK
code

struct S1 { };
struct S2 { S1 S1; };

gcc error

<source>:2:16: error: declaration of 'S1 S2::S1' changes meaning of 'S1' [-Wchanges-meaning]
    2 | struct S2 { S1 S1; };
      |                ^~
<source>:2:13: note: used here to mean 'struct S1'
    2 | struct S2 { S1 S1; };
      |             ^~
<source>:1:8: note: declared here
    1 | struct S1 { };
      |        ^~
Compiler returned: 1

[shafik]

gcc is the only one to reject this: https://godbolt.org/z/4qxnz7ETe

I think this is ill-formed no diagnostic required from class.member.looku p6:

If it differs from the result of a search in T for M in a complete-class context ([class.mem]) of T, the program is ill-formed, no diagnostic required.

Which means everyone is correct here.

CC @zygoloid