[cfe-dev] error for narrowing type in switch case with c++11?

[Richard Smith]

On Mon, Feb 13, 2012 at 9:43 PM, Bob Wilson <bob.wilson at apple.com> wrote:

> When you compile with -std=c++11 for 32-bit targets, clang complains about
> narrowing types in switch cases:
>
> $ cat t1.cpp
> int foo(unsigned long x) {
>  switch (x) {
>  case 1: return 2;
>  case -1: return 0;
>  }
>  return 1;
> }
> $ clang++ -arch i386 -c t1.cpp -std=c++0x
> t1.cpp:4:8: error: case value evaluates to -1, which cannot be narrowed to
> type
>      'unsigned long'
>  case -1: return 0;
>       ^
> 1 error generated.
>
> The corresponding code written with if-statements does not get errors, and
> clang only warns if you enable -Wsign-compare:
>
> $ cat t2.cpp
> int foo(unsigned long x) {
>  if (x == 1) return 2;
>  else if (x == -1) return 0;
>  else return 1;
> }
> $ clang++ -arch i386 -c t2.cpp -std=c++0x
> $
>
> There is no error when compiling for x86_64 or if you don't specify
> -std=c++11.  I think this error was introduced recently -- Howard just
> changed libc++ to avoid it in svn 150082.  Is this error mandated by the
> c++11 standard or is this something we could relax?  We're seeing
> regressions building some code due to this error
> (<rdar://problem/10820369>).
>

This error is mandated by the C++11 standard; this is one of the four
places where you are not permitted narrowing conversions in C++11 (these
are: braced initializers, case values, enumerators with fixed underlying
types, and non-type template arguments). We saw some compile failures due
to this change too, but far fewer than were introduced by the narrowing
restrictions for braced initializers.

- Richard
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20120213/29f5f3b1/attachment.html>