Type Compatibility (was:failure due to compiler?)

> DEC
> 'extended' their pascal compiler to permit records with the same structure to
> be type compatible. This caused us several bugs, one of which got so deeply
> embeded in the system before it was found that we were never able to
> completely
> fix it.
>
This was not an "extension". The original CDC6600 compiler did exactly the same
thing.


This was "structure" equivalence rather than "name" equivalence. At the
time, there was a degree of debate about which way type equivalence should
be done. The ideas of data abstraction were not yet worked out, or, at
least, if they were, were not widely known. Remmember, Pascal was designed
in 1968, and that strong typing was a new idea.


Strictly one should also mention "identifier" equivalence to distinguish
it from name equivalence. This is what C and C++ have. These languages
will treat structures with the same identifier as equivalent even though
there is no guarantee that the two structures have the same definition. I
think a good software engineering environment could fix this problem.


"Name" equivalence should mean that there is only one definition, just as
the name of a variable is not its identifier, but rather its address in
memory.


In the original ETH Pascal compiler, the following would loop the compiler
in the type equivalence checker:


        A = RECORD
                  x:^A;
                  END;
        B = RECORD
                  x:^B; {actually y:^B would fail too, I think}
                  END;
        VAR
                aa:A;
                bb:B;
        BEGIN
                aa := bb;


Eventually, you ran out of stack as the compiler recursively tried to
determine if A was structurally equivalent to B.
--

Return to the comp.compilers page.
Search the comp.compilers archives again.