From: Bob Leif
To: Mark Lundquist et al.

For Ada 'Next, it would help to generalize the string libraries as generics
similar to Ada.Strings.Bounded. All of the present subprograms would be
included in the generic versions of the present string libraries.
The present positive type could be replaced in the generic packages as Type
Generic_Positive_Type or Gen_Pos_T. Thus it would be possible to have a
positive_8, Positive_16, Positive_32 length.

Generic
Type Generic_Character_Type is(<>);
type Character_Set_Type is private;
-- Representation for a set of character values:
package Ada.Generic_Strings.Maps is

with Ada.Generic_Strings.Maps;
Generic
 Type Generic Positive_Type is range <>;
 Max   : Generic_Positive_Type;    -- Maximum length of a Bounded_String
package Ada.Generic_Strings.Bounded is

--The names used above are only for purposes of illustration.

This would permit the use of any type of character including 4 bit
characters, Char_4_Type). Two uses for Char_4_Type would be used for binary
coded decimal arithmetic and to represent nucleic acid bases. I hope that it
is not heretical in these days of 64 bit processors with multimillions of
transistors, to suggest that instructions be included for 4 bit operations.

In short, we should make maximum use of Ada's generic to generalize the Ada
libraries.


-----Original Message-----
From: Team Ada: Ada Advocacy Issues (83 & 95) [mailto:[log in to unmask]]On
Behalf Of Mark Lundquist
Sent: Sunday, October 31, 1999 5:34 PM
To: [log in to unmask]
Subject: Re: Low level String abstraction (was: Java for Real Time?)


From: Christoph & Ursula Grein

> Mark Lundquist schrieb:
>
> > ... This is precisely my answer to "Ada isn't portable because
> > you don't even know how big an Integer is" (for instance). Portability
> > doesn't mean identical behavior, it means that you can deploy under a
> > different implementation of the language system without changes to the
> > source text and the requirements of the system continue to be met. If I
> > say (to continue the example)
> >
> > X: Integer;
> >
> > instead of using an implementation-independent integer type, this is an
> > implicit statement that "I do not care about the range of values of this
> > object"; that is, it has no bearing on the system requirements. You
> > should be able to say that, because there are a lot of times when you
> > really don't care (so for instance, the fact that you run out of
> > integers sooner on some platform is no more of a "portability" concern
> > than when you run out of memory or disk space! :-)
> >
> >
> ...
> > > It's all but impossible to avoid the use of Standard types; for
example,
> > > Integer is the index type for String (a bad decision made in Ada 83).
> >
> > Good point.
> >
> > To stray from the topic for a bit -- the low level of abstraction of
> > String does seem to cause some problems, beginning in Ada95.
> >
>
> I see a contradiction in these two statements. I do not care for the range
of
> Integer when I use Strings. Who would use indices such that a program runs
> with 32 bit integers but doesn't for 16 bit? If such ranges are needed, it
> seems to me that one would also run out of memory when using a machine
> where one doesn't have 32 bits.

Note that I wasn't arguing in the first statement that you shouldn't care
about representations, nor in the second that you should. Sometimes you
do, and sometimes you don't! It depends on the situation.

Actually, what I wrote wasn't very clear, because what I had in mind was
really not the index type issue Ben brought up. If you don't care about
the index base type of Standard.String, that's fine. If you do care,
you can define a string type with the index type you want. Literals,
and all the predefined string type operations from RM 3.6.3, will be
defined for it, and it will be type-convertible to Standard.String...

> And where in Ada95 does the 'low level abstraction' lead to problems?
> What would be a 'high level abstraction'? We have Ada.Strings.Fixed,
> -Bounded, -Unbounded.

Exactly! The operations defined in these packages are virtually the
same, suggesting that they represent different implementations of
the same abstraction -- But there's no type that expresses the
abstraction itself. Instead, there are four different types expressing

4 different implementations of the abstraction (times two counting the
Wide_ variants, which are all replicated again). Ada really has two
families of string types: the predefined string type family of RM 3.6.3,
and the Fixed/Bounded/Unbounded family (including the Wide_ variants).
The two families are unrelated, except that they overlap on
Standard.String/Fixed (and Standard.Wide_String/Wide_Fixed).

One family defines the types in terms of their representations
(low-level) and the other defines them in terms of their operations
(high-level, but without any factoring).

Ada95 was constrained by the Ada83 definition of String. But even aside
from that, isn't it nice that String is publicly an array type, so you
can index and slice it? Would that have been worth trading away to
unify the string type families? I don't know. It's tempting to say
Ada, like C++, should provide user-defined indexability (and
slicability), maybe using an attribute-definition-clause. But then, you
would have to invent new syntax for declaring the public view of the
type to have array-like properties, so that the compiler knows an
indexed component or slice when it sees one (in C++ it's not a problem
because the '[]' operator is its own syntax). Worse yet, Ada would have
to somehow follow C++ into the morass of function calls being lvalues,
i.e. functions returning variables instead of values.

I don't know if the current state of affairs is pragmatically that big
of a deal, other than just cluttering up Annex A. You can always write
statically polymorphic string methods using generics...

> [To me, some people seem to be very paranoid about predefined types.
> Don't you dare write "type Something is array (1..10);" - that's bad and
> not portable; "type Something is array (Integer_8 range 1..10);" however
> is very good and portable.]
>

Right, I totally agree with you! Dogmatism of that kind is often a
workaround for cluelessness. The point is, know what you care about and
why, and then know how to express that in language you're programming
in.

> --------- __o __o /\_ _ \\o (_)\__/o (_)
> ------- _`\(_) (_)/ ------ (_)/ (_) (_)/ (_) (_) (_) (_) (_)' _\o_
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> Mein Radunfall
> My bike accident

Yikes!

> Christoph Grein


--

Mark Lundquist
Senior Software Engineer
Rational Software
Development Solutions Business Unit
UNIX Suites Group
Aloha, OR, USA