[PATCH] Warn on duplicate function specifier

[Reid Kleckner]

================
Comment at: lib/Parse/ParseDecl.cpp:2770
@@ -2769,3 +2769,3 @@
     case tok::kw___forceinline: {
-      isInvalid = DS.setFunctionSpecInline(Loc);
+      isInvalid = DS.setFunctionSpecInline(Loc, PrevSpec, DiagID);
       IdentifierInfo *AttrName = Tok.getIdentifierInfo();
----------------
Richard Smith wrote:
> Do we really want a diagnostic for combining `inline` with `__forceinline`? That combination doesn't seem unreasonable to me.
MSVC does something unreasonable like this:

  void __forceinline inline foo() {} // expected-warning {{C4141: 'inline' : used more than once}}
  void inline __forceinline bar() {} // nothing

Clang **requires** users to apply inline in conjunction with always_inline, which leads users to write code like this to make all compilers happy:

  #ifdef _MSC_VER
  # define PORT_ALWAS_INLINE __forceline
  #else
  # define PORT_ALWAYS_INLINE __attribute__((always_inline))
  #endif
  void inline PORT_ALWAYS_INLINE foo() {}

We *definitely* shouldn't warn on that regardless of the ifdef taken.


http://llvm-reviews.chandlerc.com/D2025