Re: Saving extension semantic values from collection

[Mark Mitchell <mark at codesourcery dot com>]

>>>>> "Richard" == Richard Kenner <kenner@vlsi1.ultra.nyu.edu> writes:

    Richard>     Speaking as an ex-mathematician, prove is a very
    Richard> strong word.

    Richard> I purposely used such a loaded word.  But, in some sense,
    Richard> that's the issue.  If you depend on an "obscure" property
    Richard> and you can't even *begin* to try to prove it's true,
    Richard> it's a very dubious thing to be depending on.

Well, I was a theoretician.  Alex, for example, is an empiricist.
What he calls a proof is very different from what I call a proof --
proofs for him are often statistical observations of experiments.

As I said before, I think you'd be hard-pressed to prove most
significant properties of GCC.  Try, for example, to prove that `int
main () { return 3; }' will actually generate code that will return
three.

I can, and have, *begun* to prove that the parse nodes are all
reachable; partly by inspection, and partly by empirical testing.

--
Mark Mitchell                   mark@codesourcery.com
CodeSourcery, LLC               http://www.codesourcery.com