Re: The 1980 ACM Turing Award Lecture The Emperor's Old Clothes

["Robert I. Eachus" <rieachus@attbi.com>]

Hyman Rosen wrote:

 
> Really? I find this utterly astonishing. So almost every Ada program is
> littered with error handling code interpenetrating the normal working 
> logic?
> Or do Ada programs ignore eror codes just like old C programs did?


No.  Input output routines may need to do error checking, but that is 
not an exceptional occurrence, and is usually not treated as one.  The 
'Valid attribute in Ada 95 is a great enhancement to the language.  It 
makes validity checking for input much easier to do, and to do in the 
right way.

Most other cases where return values should be checked in C and are not, 
in Ada the checking is done by constraints on parameters well before the 
call.  If your code can cause one of those constraint checks to fail, it 
may be that you need to rethink what you are doing.

This propagation of validity in Ada means that putting constraints on 
values (and running with constraint checks turned on) can be faster in 
Ada than eliminating constraint checks.

As a sort of trivial example:

function Square_Root(X: in Natural) return Natural;

..does not need to do any internal checks, and will never raise an 
exception.  (With the possible exception of Storage_Error...) 
Similarly, a procedure declared:

procedure Foo(S: in out String);  -- will never cause a buffer overrun.

Of course, sometimes you need to use Ada.Strings.Bounded or 
Ada.Strings.Unbounded to get the correct semantic behavior.  But that is 
just an instance of a general Ada rule.  Using strong typing right 
eliminates a lot of errors, whether they are coding errors, compiler 
error messages, or potential exceptions.

I hope this is all preaching to the choir, but I fell this thread 
contains too much self-flagellation as it is, and some balance is needed.