[Mingw-users] Unwanted warning: inline dllimport

[Greg C. <chi...@co...>]

This report seems similar to
  http://gcc.gnu.org/bugzilla/show_bug.cgi?id=12477
yet I'm also reporting a gcc-3.4.5 regression.

12477 is the only report I find with keywords 'inline dllimport'.

* To reproduce:

C:/tmp[0]$cat inline_dllimport.cpp
inline int __declspec(dllimport) foo(int i)
{ return i+i; }

C:/tmp[0]$/MinGW-20050120/bin/g++ -dumpversion
3.4.2
C:/tmp[0]$/MinGW-20050120/bin/g++ -c inline_dllimport.cpp

C:/tmp[0]$/MinGW-20050827/bin/g++ -dumpversion
3.4.4
C:/tmp[0]$/MinGW-20050827/bin/g++ -c inline_dllimport.cpp

C:/tmp[0]$/MinGW-20060119/bin/g++ -dumpversion
3.4.5
C:/tmp[0]$/MinGW-20060119/bin/g++ -c inline_dllimport.cpp
inline_dllimport.cpp:2: warning: inline function `int foo(int)' \
declared as dllimport: attribute ignored

Only 3.4.5 complains. I can make 3.4.2 complain by adding '-W':

C:/tmp[0]$/MinGW-20050120/bin/g++ -c -W inline_dllimport.cpp
inline_dllimport.cpp:2: warning: inline function 'int foo(int)' \
is declared as dllimport: attribute ignored.

But 3.4.5 complains with no explicit options.

* Why this warning seems very aggressive, with or without '-W'

Here is a real-world case from wxWidgets:

  wxWidgets/include/wx/dnd.h

  48  inline WXDLLEXPORT bool wxIsDragResultOk(wxDragResult res)
  49  {
  50      return res == wxDragCopy || res == wxDragMove || res == wxDragLink;
  51  }

I asked one of the wx developers to consider removing their
dll[im|ex]port macro. He replied that we can't know for sure
whether the compiler actually inlines that function, and, if
it doesn't, then we really do need dllimport to use it.

Is there any way to write that function correctly without
getting this diagnostic? (Please don't say '-w'.)

* Guidance requested

Is this sufficiently distinct from 12477 to warrant a separate
report? Or is it worth adding a comment to 12477 indicating
that 3.4.5 gives this warning by default?

Or is it vain to report it, and should I just wait and hope
that something like '-Wno-attributes' lets me turn off the
warning? The 4.1.0 manual says "This will not stop errors for
incorrect use of supported attributes", though I think a case
can be made that the testcase usage is not incorrect.