> dn> I'll remain skeptical about this until I see something
 > dn> with my own eyes which actually does what it advertises.
 > dn> So far I haven't.

 > I've often wondered why it would be difficult to produce a
 > compiler and library specific disassembler that encapsulated
 > the standard library calls and just got down to the basics
 > :).  For instance, the disassembler could recognise Borland's
 > large model startup code, and each of the library functions,
 > and reduce the output to 'user-written' code only.

Umm, yeah, but which particular version of which Borland compiler ?

'course you could hardwire lots of stuff, recognise library functions, but
you'd have to throw it all out when a new version was released, or somehow
produce a library that recognised lots of different things and could be
updated. An AI approach shows more promise.

Personally, I don't thing it's worth the effort.  By the time you trained
your AI system to recognise various constructs, you could have written the
original code and you'd own the copyright to your very own guaranteeed
fully 'legal' source code without any possibility of breaching anyone's
license agreement by reverse engineering published object code (he says,
hoping that this statement doesn't degenerate into Yet Another Legal Thread
:-().

  cheers,
  david

---