[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Autoconf / Automake for generating Cons scripts

From: Tom Tromey
Subject: Re: Autoconf / Automake for generating Cons scripts
Date: 21 Jul 2001 20:02:43 -0600

>>>>> "Eric" == Eric Siegerman <address@hidden> writes:

Eric>   - On the other hand, it probably DOES make sense to use
Eric>   autoconf with cons, and you can probably start doing so in
Eric>   small ways without a lot of effort.  Simply identify a few
Eric>   system-dependent facts that autoconf can figure out more
Eric>   easily than cons can, and figure out what it takes to encode
Eric>   the results as macros in the Cons* files.

If I were to redesign auto* from scratch, I wouldn't have any
separation between autoconf and automake.  Instead I would have the
build tool run configuration tests for me.  Caching would be done by
the tool in a global way for an entire project (you need a cache of
some kind in order to take a build tool beyond `make' anyway).

One benefit of this kind of merge is that reconfiguring, like
rebuilding, becomes a minimal operation: you only rebuild the bits you
need to.

Another more important benefit is that the configuration tool can
express interdependencies between tests to the build tool.  This means
you can run configuration checks in parallel.  This is important
because, these days, `configure' actually takes a noticeable chunk of
a large (e.g., gcc) build, at least if you're running with a parallel
make on a relatively powerful box (say, 4 processors).

Finally, in my ultimately plan, libtool simply wouldn't exist at all
as an independent entity.  Instead it would be integrated into the
build tool (as a variant of the "compiler object").  This would make
it much more efficient, one would hope.

Unfortunately, a redesign like this requires a lot of work: technical,
political, and cultural.  It's hard to guess how long it would take.

So, in the meantime, we continue on our present course.  For automake
at least, we haven't quite exhausted the capabilities of make.
However, we're getting there.  We can implement single-Makefile
builds.  However we probably can't implement "object winking", or
checksumming, or command dependencies in a portable way.  (These are
all important features for any successor to make.)

As far as Cons goes, I skimmed the manual a few weeks back but then
forgot what I read.  Integrating it with autoconf is probably possible
and desirable.  However, this direction has limits if you intend to
use Cons (or make, or anything) on projects beyond a certain size.


reply via email to

[Prev in Thread] Current Thread [Next in Thread]