Re: Using C++ conditionnally

[Ralf Wildenhues]

Hi Timothée,

* Timothée Lecomte wrote on Tue, May 02, 2006 at 07:06:09AM CEST:
> >
> > This the bug that the second AC_PROG_C* invocation won't fail:
> 
> And in which version does this bug occur ? Currently the gnuplot
> configure.in asks for 2.58 at least.

AFAIK this issue is present in 2.58 up to current CVS (untested).

> >> so that automake will finally think that the project contains some
> >> C++ files, and use g++ to link the object files. .... And there it
> >> fails.
> >
> > Wait; I don't see why this happens.

> You are right, the C++ files are listed in EXTRA_gnuplot_SOURCES, and the
> corresponding object files are placed in gnuplot_DEPENDENCIES when the
> feature is enabled in the configure script.

That's not so great.  Automake treats object file names as an internal
detail, to facilitate compiling the same source file with different
options (and IIRC also to stand a chance on 8+3 file systems).

> I chose this scheme from the
> autoconf or automake manual, I don't remember precisely ;-)

I don't think the Automake manual recommends that.

If you can avoid it, don't touch *_DEPENDENCIES, but let Automake
compute them for you.  Do this instead:

gnuplot_SOURCES = ...normal sources..
if WANT_WXWIDGETS
gnuplot_LINK = $(CCLD) $(AM_CFLAGS) $(CFLAGS) $(AM_LDFLAGS) $(LDFLAGS) -o $@
gnuplot_SOURCES += gp_cairo.c ...
else
gnuplot_LINK = $(CXXLD) $(AM_CXXFLAGS) $(CXXFLAGS) $(AM_LDFLAGS) $(LDFLAGS) -o 
$@
endif

If gnuplot is the only program to be linked in this Makefile.am, you can
get away with 
  CXXLD = @CC@

instead of setting gnuplot_LINK, I guess.

For the Automake conditional to work, put
  AM_CONDITIONAL([WANT_WXWIDGETS], [test "${enable_wxwidgets_ok}" = yes])

in configure.ac

> > It looks ok for Autoconf-2.59 and CVS as of now, if the AC_PROG_CXX
> > comes after AC_PROG_CC, but depending on how above issues will be fixed,
> > this may need to be revisited.
> 
> So, to make something more waterproof, I could try :
> 
> m4_pushdef([AC_MSG_ERROR], [cxx_error=yes])
> AC_PROG_CXX
> AC_LANG_PUSH([C++])
> AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[const char hw[] = "Hello, World\n";]],
>                 [[cout << hw;]])],
>       [AC_PROG_CXXCPP
>       cxx_error=no],
>       [cxx_error=yes])
> AC_LANG_POP([C++])
> m4_popdef([AC_MSG_ERROR])

> So that if any error occur during with the c++ compiler, cxx_error is
> defined to "no" and I can do tests on it. Is it correct ?

Well, yes, for the code at the moment that looks ok.  I don't want to
give any guarantees here though since you are munging with internal
Autoconf details.  Using `cout' without namespace is going to get you in
trouble though, if you want to be portable to ISO C++ compilers.  (Using
it with namespace will not work with older compilers, to which I don't
know whether you want to be portable to them; if yes, look at the
AC_CXX_NAMESPACES macro provided by the Autoconf Macro Archive.)

Cheers,
Ralf