On Tue, Jan 26, 2021 at 12:32 PM Bruno Haible <br...@clisp.org> wrote:
>
> Jeffrey Walton wrote:
> > > Jeffrey Walton wrote:
> > > > Placing flags in CC breaks other scripts. For example, this no longer
> > > > works:
> > > >
> > > > if [[ ! $(command -v "${CC}") ]]; then
> > > > echo "The compiler is not valid. Please install a compiler."
> > > > exit 1
> > > > fi
> > >
> > > Such scripts are not supported.
> >
> > I actually do that in my configure.ac. Quality user feedback is more
> > important
>
> No one disputes that. But you can correct your code to support a $CC value
> with arguments:
> set x $CC
> if (type $1) >/dev/null 2>/dev/null; ....
>
> > > Some flags, like '-m32' or '-m64', MUST be put in $CC. [1]
> >
> > I place them where they belong - in CFLAGS and CXXFLAGS. The downside
> > is you have to pick the triplet on Solaris and a few other platforms.
>
> Exactly that's the reason why we documented it this way in Autoconf:
> The triplet detection does not work right if a user puts '-m32' or '-m64'
> in $CFLAGS or $CPPFLAGS.
>
> > Working around the Autotools bugs just means they will never get
> > fixed. They will keep telling the world to work around their bugs.
>
> You can have a different opinion about how the Autotools should behave.
> But once we have documented how they behave and what a user needs to
> do in order to get good results, it's not a "bug" any more.
Documenting a bug does not make it any less a bug.
Jeff
