On 3/22/07, C. Michael Pilato <cmpilato@collab.net> wrote:
> Daniel Rall wrote:
> > On Thu, 22 Mar 2007, Garrett Rooney wrote:
> >
> >> On 3/22/07, Paul Burba <pburba@collab.net> wrote:
> >>> Recently when Erik added the new private header file svn_wc_private.h I
> >>> noticed he didn't include the extern "C" logic:
> >>>
> >>> #ifdef __cplusplus
> >>> extern "C" {
> >>> #endif /* __cplusplus */
> >>>
> >>> #ifdef __cplusplus
> >>> }
> >>> #endif /* __cplusplus */
> >>>
> >>> Which seems the correct thing to do since these functions are not part
> >>> of Subversion's public API and won't be called by any C++ code. We do
> >>> have the extern logic in all the other include/private header files
> >>> however. Is this necessary for any reason? I don't really care about
> >>> changing this since it seems harmless enough, just want to make sure I
> >>> understand it correctly.
> >> The extern "C" statements are needed for any C header that might be
> >> included by C++ code. If you don't wrap the code in extern "C" then
> >> the C++ code that tries to use it will use C++ name mangling on the
> >> symbols and you'll get a link error. Wrapping everything in extern
> >> "C" blocks tells the C++ compiler to not mangle the symbol names so
> >> everything works out at link time.
> >
> > svn_wc_private.h should have the extern "C" markup.
>
> Dan, Garrett:
>
> It might just be me, but I feel like I'm not seeing the real answer to
> Paul's question here.
>
> I think he understands what the extern "C" is all about. His question is
> about a specific file which has the specific feature that it is *not* part
> of our public API and therefore should *not* be directly included in an C++
> code.
>
> Both your answers look like recitations of a guideline, but (to me) leave
> room to wonder if the guideline applies in this specific situation. I can
> read your answers as ways of saying, "It doesn't matter if the file is
> included by C++ code directly or not, you need the extern "C" wrapper
> regardless". But this opposes my understanding of way C/C++ interact, so
> I'm left to wonder if, in fact, you both noticed that tiny little detail
> about the fact svn_wc_private.h is not meant to be called by C++ code.
The only case I can think of is if we expect this private header to be
included by C++ code in our tree, like JavaHL. Other than that it's
not necessary, but then again it's not like it really hurts anything
to put it there.
-garrett
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Fri Mar 23 05:45:41 2007