r180603 - C++1y: support simple variable assignments in constexpr functions.

[Abramo Bagnara]

Il 05/05/2013 17:05, Richard Smith ha scritto:
> On Sun, May 5, 2013 at 2:53 AM, Abramo Bagnara
> <abramo.bagnara at bugseng.com <mailto:abramo.bagnara at bugseng.com>> wrote:
> 
>     Il 26/04/2013 16:36, Richard Smith ha scritto:
>     > Modified: cfe/trunk/lib/Sema/SemaInit.cpp
>     > URL:
>     http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Sema/SemaInit.cpp?rev=180603&r1=180602&r2=180603&view=diff
>     >
>     ==============================================================================
>     > --- cfe/trunk/lib/Sema/SemaInit.cpp (original)
>     > +++ cfe/trunk/lib/Sema/SemaInit.cpp Fri Apr 26 09:36:30 2013
>     > @@ -97,6 +97,7 @@ static void CheckStringInit(Expr *Str, Q
>     >      DeclT = S.Context.getConstantArrayType(IAT->getElementType(),
>     >                                             ConstVal,
>     >                                             ArrayType::Normal, 0);
>     > +    Str->setType(DeclT);
>     >      return;
>     >    }
> 
>     Is this change deliberate? It seems to introduce a regression:
> 
>     $ cat z.c
> 
>     void f() {
>       signed char s[] = "a";
>       unsigned char u[] = "a";
>     }
>     $ _clang -cc1 -ast-dump z.c
>     TranslationUnitDecl 0x67d96e0 <<invalid sloc>>
>     |-TypedefDecl 0x67d9bc0 <<invalid sloc>> __int128_t '__int128'
>     |-TypedefDecl 0x67d9c20 <<invalid sloc>> __uint128_t 'unsigned __int128'
>     |-TypedefDecl 0x67d9f70 <<invalid sloc>> __builtin_va_list
>     '__va_list_tag [1]'
>     `-FunctionDecl 0x67da010 <z.c:2:1, line:5:1> f 'void ()'
>       `-CompoundStmt 0x67da350 <line:2:10, line:5:1>
>         |-DeclStmt 0x67da208 <line:3:3, col:24>
>         | `-VarDecl 0x67da100 <col:3, col:21> s 'signed char [2]'
>         |   `-StringLiteral 0x67da198 <col:21> 'signed char [2]' lvalue "a"
>         `-DeclStmt 0x67da338 <line:4:3, col:26>
>           `-VarDecl 0x67da270 <col:3, col:23> u 'unsigned char [2]'
>             `-StringLiteral 0x67da2c8 <col:23> 'unsigned char [2]'
>     lvalue "a"
> 
>     Type of string literal should be plain char.
> 
> 
> Yes, this is deliberate; we intended to set the string literal's type to
> the type of the initialized variable (otherwise we would be initializing
> an array of 'unsigned char' from an array of 'char'), but accidentally
> only updated it in either the array bound or the type, but not both.
> 
>   unsigned char a[] = "foo", b[4] = "bar";
> 
> ... used to produce ...
> 
> |-VarDecl 0x6222ad0 <<stdin>:1:1, col:21> a 'unsigned char [4]'
> | `-StringLiteral 0x6222ba8 <col:21> 'const char [4]' lvalue "foo"
> `-VarDecl 0x6222c60 <col:1, col:35> b 'unsigned char [4]'
>   `-StringLiteral 0x6222cb8 <col:35> 'unsigned char [4]' lvalue "bar"

Of course this was bad and should be fixed for uniformity, but the
alternative chosen is the right one?

How is this compatible with C99 6.4.5p4 and C11 6.4.5p6?

"For character string literals, the array elements have
type char, and are initialized with the individual bytes of the
multibyte character sequence."


-- 
Abramo Bagnara

BUGSENG srl - http://bugseng.com
mailto:abramo.bagnara at bugseng.com