Re: [Mingw-users] problem with inline C++ member functions in DLLs

[Ruediger W. <rue...@la...>]

Hello Andrew!

it is not an Bug in gcc its an feature ;-) .
gcc always emitts an out of line copy of on inline function even when all
functions had been inlined.
To turn this off use following commandswitch when compiling.

-fno-keep-inline-functions


> I compiled this into its .o, ar'd it into a .a, and then used Paul's
> a2dll utility (extremely handy, thanks!) to turn it into a DLL.
>
> Then, I create another file (george.cc) that #include's this class,
> along with the inline function definition. I compile and link that, but
> when I used a2dll to try to turn that into a DLL, I get a multiple
> definition error: hello() is defined twice, once in the import library
> for the DLL that a2dll created, and once in the george.cc file.
>
> Looking in the import library for my first DLL, I see three definitions
> re: hello:
>
> I __imp___hello__C3Foo...
> I __nm__hello__C3Foo...
> T _hello__C3Foo...
>
> This last definition is the problem I think. Should this really be in
> the import library? I think that this definition causes the a2dll'ing of
> george.cc to fail, since I'm sure that its .o contains the inline'd
> definition of hello() as well.
>
> Is this a GCC bug (i'm using gcc 2.95.2-3 from the mingw.org web page)
> where inline functions that are inlined after the class definition are
> marked to be included in the import library, when they really shouldn't?
>