Re: to conditionally test, or not to conditionally test?

[Ralf Wildenhues]

Hi Ed,

* Ed Hartnett wrote on Fri, May 20, 2005 at 04:51:41PM CEST:
> 
> I read somewhere in my autoconf researches on the web that tests
> should not be run conditionally, based on earlier tests - they should
> always run. In other words, don't try and optimize configure.ac.
> 
> Is this really good advice?

In a way, yes.  Autoconf cannot see the condition, and thus when a later
test depends on an earlier test (but is not run because of some shell
condition), things break.  There is AS_IF, but it's rather a kludge than
a solution.

> For example, in my library, there are optionally-built C++, F77, and
> F90 APIs. If you want the C++ API, I need to run a bunch of C++ tests.
> 
> If you don't want C++, I would prefer not to run them. 

I believe that should be ok.  F77 tests are not going to depend on C++
tests, generally.

> I wonder if anyone would like to comment on the configure.ac fragment
> below. This code seems to work fine, but am I missing something? Note
> that checking the C++ compiler occurs only if $nc_build_cxx = yes.

> >From configure.ac:
> 
> # On some systems, a discovered compiler nevertheless won't work
> # (because it's a script to a non-existent executable, for example);
> # thus, we check the compiler with a test program.  We also test for
> # "iostream" and the standard C++ library because we need these to
> # work.
> if test "x${nc_build_cxx}" = xyes; then
>    AC_MSG_CHECKING([if C++ compiler works])
>    AC_LANG_SAVE()
>    AC_LANG_CPLUSPLUS()
>    AC_RUN_IFELSE([
>       AC_LANG_SOURCE([[
>                       #include <iostream>

I believe that should be
        AC_LANG_SOURCE([[#include <iostream>
#include <whatever_else_is_necessary>

At least portable C needs the hash in the first column, I'm unsure about
C++.

>               int main() {
>                   std::cout << "";
>                   return 0;
>               }
>               ]])],

I believe there are useful C++ macros in the Autoconf macro archive[1]
which you might want to check out.

>    [nc_cxx_worked=yes],
>    [nc_cxx_worked=no],
>    [])
>    AC_LANG_RESTORE()
>    AC_MSG_RESULT([$nc_cxx_worked])
> 
> # If the CXX compiler doesn't work, exit if --disable-compiler-recover
> # was used, otherwise, turn off the C++ API build.
>    AC_MSG_CHECKING([whether CXX API can be built])
>    if test "x$nc_cxx_worked" = xno ; then
>       nc_build_cxx=no && CXX=''
>    fi
>    AC_MSG_RESULT([$nc_build_cxx])
>    if test "x${nc_cxx_worked}" = xno ; then
>       if test "x${nc_compiler_recover}" = xno ; then
>          AC_MSG_WARN([exiting because of broken CXX compiler and 
> --disable-compiler-recover])
>          exit 1
>       else
>          AC_MSG_WARN([Can't get C++ compiler to work. C++ API won't be 
> built.])
>       fi
>    fi
> fi
> AM_CONDITIONAL(BUILD_CXX, [test "x$nc_build_cxx" = xyes])

I believe conditionalizing on this will in practice work quite well.

Regards,
Ralf

[1] http://autoconf-archive.cryp.to/