Re: the Ada mandate, and why it collapsed and died (was): 64 bit addressing and OOP

["Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de>]

Robert A Duff wrote:

> "Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de> writes:
> 
>> Robert A Duff wrote:
>> 
>> > Yes, by "L-value", I meant on the left-hand side of an assignment, or
>> > an
>> > 'out' parameter, or similar things.  If X(1..80) is just syntactic
>> > sugar for a function call Slice(X, 1, 80) or Slice(X, First => 1, Last
>> > => 80), then it would be little trouble for the Ada compiler.
>> 
>> Do you mean that Slice (...) would always have a copy semantics?
> 
> Yes.

Ooch

>> > And I would think it would be but a minor convenience for the
>> > programmer.
>> 
>> > But when slices are used as L-values, there are various interactions
>> > with packing and whatnot that cause compiler complexity.
>> 
>> Isn't it already there because array elements can be also L-values? Or
>> that objects of some types have to be constructed in-place? I mean, with
>> or without slices, one cannot get rid of this problem anyway.
> 
> No, I think slices-as-L-values cause compiler complexity beyond what
> components-as-L-values do.

I suppose it is not a big problem to generate a loop, so it must be 
something else. Maybe you mean things like optimization of an assignment 
into a memory copy if no element finalization required? But we do need 
that! Randy Brukardt have already mentioned a buffer extension technique 
based on slices. I am using it all the time in generic containers in hope 
that for non-controlled types it will become just a memory copy. Now let we 
remove slices. Then to achive the same level of optimization, the compiler 
has to be able to redice a hand-written loop (+unpack/pack rubbish 
generated) to a memory copy. Is it easier?

From a global perspective. To me slices is a good example of how things 
*should* be done. I mean the issues like:

- unconstrained string vs. constrained string;
- unconstrained discriminated object vs. a dynamically constrained one;
- class-wide object vs. specific one

Only strings and slices have it right. The dope is used only when necessary. 
On contrary discriminants and tags are always there. Which makes, for 
instance, the problem of measurement units unsolvable. From this point of 
view removing slices would be a step back.

-- 
Regards,
Dmitry A. Kazakov
www.dmitry-kazakov.de