Re: Delphi & Ada; Ada to C++

[dewar@merv.cs.nyu.edu (Robert Dewar)]

In article <01bd3e24$280e1c00$a8fc82c1@xhv46.dial.pipex.com> "Nick Roberts" <Ni
k.Roberts@dial.pipex.com> writes:
>Now to throw the cat among the pigeons: how difficult would it be to
>retro-compile the tree into C (or even C++)?


(tree here = GNAT tree)

It would conceivably be possible to compile it into VERY low level C, where
you had a few variables corresponding to registers, say called r0, r1, r2
and you generated C that for example, in unoptimized mode, might
compile

   A := B + C;

where A,B,C are all local variables, into something like

   r0 = *(fp+100);
   r1 = *(fp+104);
   r2 = r0 + r1;
   *(fp+108) = r2;

This is *really* using C as a low level assembler :-)

The way you could do this is to write a config file for gcc that treated
C in this way as a kind of pseudo-machine. We once thought about this as
a way to make a version of GNAT that could be ported in native mode, rather
than requiring cross-compilers, but it was never more than a passing
interesting idea.

You could try to reverse compile the tree at a much higher level into
C, but this would be a huge task, comparable to that of writing a
compiler, and in some ways much more difficult (it would be like trying
to do a complete house renovation, rather than building a new house --
as those who have gone through it know, the renovation can often be the
harder task).

Furthermore, some of the constructs would still generate rubbish, since
C just does not have the required semantic richness.
(e.g. what do you do with overflow checking, and what do you do with
 nested procedures ....)