[LLVMdev] RE: LLVM + automake?

Chris Lattner sabre at nondot.org
Sun Nov 9 17:08:03 PST 2003


> > >      6. Have you given any thought to using automake to generate your
> > >         Makefiles? I use automake extensively and would prefer to
> > >         continue to use it with LLVM. It is simple and works well on
> > >         many platforms. If you'd like to go this route, I'd be willing
> > >         to do the conversion for you.
> >
> > I have given no thought to this, and I don't have any strong feelings
> > either way.  What would switching buy us?  Before you did this, you should
> > email John Criswell (and the LLVMdev list in general) to see what he
> > thinks, as he's the build system maintainer.
>
> It buys you a standard infrastructure that integrates cleanly with
> autoconf. It generates Makefile.in files from a very short Makefile.am
> (similar to system you're using but with many additional features). The
> Makefile.in files are fully capable of being configured by autoconf just
> as any *.in file is. The big advantage to me is that you let a bunch of
> build gurus associated with GNU/FSF handle the details of the make
> system -- less work in maintaining the Makefile system. Another big
> advantage is that it conforms to the GNU Makefile standard which
> prescribes the set of targets and implements numerous targets on your
> behalf. The resulting Makefiles will be easily understood by anyone
> working with open source, especially GNU/FSF source.  It has everything
> you have now and then some. For example, it integrates cleanly with
> DejaGNU for automated testing using "expect". This is really handy,
> especially for compilers.  It also buys you some platform independence
> since it can work with native "make" if necessary.

All of this is good, but again, I don't know how much it will buy us.  In
particular, we already require gnu make, and I am a firm believer that we
shouldn't try to support all of the old and archaic broken makes out
there.  Also, our regression suite is built on QMTest currently, not
dejagnu, so that's not a big win.

Overall, it might be worth it, it might not.  In any case, John is the one
to talk about for these issues.  As long as the build works and it is easy
to maintain, I don't care much either way. :)

-Chris

-- 
http://llvm.cs.uiuc.edu/
http://www.nondot.org/~sabre/Projects/





More information about the llvm-dev mailing list