Re: Whirlwind - composing analyses automatically

From: Craig Chambers (chambers@cs.washington.edu)
Date: Tue Feb 11 2003 - 14:30:38 PST

  • Next message: Craig Chambers: "Re: A few suggestions for the V/W environment"

    Yes, this is one way to try to modularly organize analyses. But with
    precedence, if two unrelated operators are used together, there's a static error
    about unspecified precedence. This helps to make the modular approach robust to
    unanticipated combinations. I'm not sure the same property would be true about
    unrelated optimizations. It seems that you're assuming by default that two
    optimizations have no interactions by default, which doesn't seem true.

    An alternative is to write a separate script to order the optimizations, which,
    which being non-modular, is possibly easier to change than the Cecil source.

    -- Craig
     

    Andrei Alexandrescu wrote:
    >
    > I was just thinking of an interesting analogy between combining analyses in
    > whirlwind and the way operator precedence is defined in Cecil (the best
    > approach I've seen btw).
    >
    > The idea is that maybe new analyses can be positioned relatively to the
    > existing ones. So someone who introduces a new analysis can specify some
    > precendence requirements, and then let the framework combine freely the
    > analyses of the same "strength".
    >
    > For example, cfg_constant_prop can specify that it must always run before
    > constant_prop. This does introduce a little dependency between different
    > analyses, but only between those of which order is important anyway. Then,
    > analyses that can be combined arbitrarily would not depend on each other at
    > all.
    >
    > Would this work?
    >
    > Andrei
    >
    > _______________________________________________
    > Cecil mailing list
    > Cecil@cs.washington.edu
    > http://mailman.cs.washington.edu/mailman/listinfo/cecil
    _______________________________________________
    Cecil mailing list
    Cecil@cs.washington.edu
    http://mailman.cs.washington.edu/mailman/listinfo/cecil



    This archive was generated by hypermail 2.1.5 : Tue Feb 11 2003 - 14:30:43 PST