"Subtract C, add Ada"

[bill@valiant (R.A.L Williams)]

[Our news system broke down earlier this week so I don't know
 whether this message got out. If it did I apologise for the
 wasted bandwidth.

 Bill Williams]

Robert Dewar wrote..
> Bill comments that for his points 3. and 4.
> 
>   3. uninitialised pointer access
>   4. pointer references to local variables in defunct procedures
> 
> "The syntax of Ada .. does not really address 3. or 4."
> 
> Taking point 3, it is impossible to have an uninitialized pointer in Ada,
> all pointers are initialized to null, and any attempt to use a null pointer,
> i.e. to dereference it, will cause a constraint error. Seems like pretty
> complete protection to me.

Yes, you're right, Ada *guarantees* a run time crash, in C/C++ it's just very
likely. It's still not a compile time check so you're relying on the
quality of your run time environment to tie down where the problem
occurs. Of course, in embedded systems you don't want *any* exceptions,
no matter how cleanly you handle them! I'm not saying that we necessarily
want to go to a language which either doesn't allow pointers (yuck!) or
constrains their use to such an extent that compile time checks are
feasible; I was just trying to point out that using Ada is no substitute
for good engineering practice in design and coding.

> With regard to point 4, ther is no way to get pointer references to local
> variables in Ada 83 (except use of 'Address, and non-portable maybe-works,
> maybe-doesn't unchecked conversion of the address to an access type, which
> really doesn't count). Certainly students (and most other Ada programmers)
> should be forbidden from using such constructs (the use of 'Address, and
> indeed of unchecked conversion of addresses to access types is permissible
> in some situations, but taking the 'Address of locals in a subprogram is
> pretty dubious).

Again, you're right BUT, 'ADDRESS is something you need to do fairly
often when interfacing to foreign languages. Sorry, don't want to start a
religious argument about libraries etc. it's just a fact of life.

I agree whole heartedly about restricting the constructs used by 'novice'
programmers or those coding for `critical' environments.

> In Ada 95, it is indeed possible to take the 'Access of local variables,
> provided they are marked aliased, but the accessibility rules make it
> impossible to have dangling pointers. Dangling pointers can be created
> by the use of Unchecked_Access, but this is in the same category as
> Unchecked_Conversion, to be used with great care in rare cases.

I'm looking forward to trying out Ada95. Got GNAT 2.0 up and going on
a SPARC station that I have occasional access to. I'm really looking
forward to the HP700 port because that's what our local network is.

Thanks for the feedback.

Bill Williams