TEAM-ADA Archives

Team Ada: Ada Programming Language Advocacy


Options: Use Forum View

Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
"Robert I. Eachus" <[log in to unmask]>
Reply To:
Robert I. Eachus
Tue, 24 Nov 1998 17:04:16 -0500
text/plain (153 lines)
(I said) >    This is starting to get silly.

At 08:19 PM 11/20/98 -0600, Samuel Mize wrote:

>Sorry, I thought this list was (among other things) a place to toss
>out and develop ideas that might help make Ada more saleable to others.

   No problem with that.  I was just saying that I thought too many
restrictions were getting rolled into one pragma.

(Me again:)
>>Ada is a very rich language, and while I
>> can understand having a tool to detect some departures from local coding
>> standards and software engineering practices, putting on a straight-jacket
>> is not all that good an idea.

>I don't see why marking some packages, and putting limits on those
>marked packages, would be "putting on a straight-jacket."  If a
>given package needs to exceed those limits, you don't use the
>marker, just as you don't use "positive" for counting from zero.

   I agree.

>But I was just tossing out a thought about "class" packages.  If
>you're saying I defined too tight a limit on parameter lists, I'd
>like to understand what you mean.  Someone else (I think it was Simon)
>pointed out that we need selector functions.

>Did you perhaps think I was saying that all packages should be
>classes?  Good grief, no.

   Definitely no, but I think it should be possible to write "pure"
OO programs with all class packages conforming to the pragma. If the
pragma requires deviating from pure OO to conform to the pragma, it is
too tight.

>Perhaps a "class" pragma would have a parameter that states whether
>the package's parameter lists are limited to object-first order.
>Perhaps the pragma's parameters would let you specify several types
>of object-oriented coding limits -- another advantage of marking
>packages as "classes" with a pragma.

    I would go along with 1) procedures must have a parameter of the
type first in the parameter list, and 2) constructor functions which
return an object of the type are also allowed, whether or not they have
a parameter in the class.

>To some extent I was reacting to the typical comment that you can
>do classes in Ada, it's just a tagged type in a package.  No.  That
>isn't a language-defined class.  There are semantic constraints
>behind the concept "class" that are not checked by putting a tagged
>type in a package.

>If you do a tagged-type-in-a-package combo right, you wind up with
>something equivalent to a class.

    Ah, now we get to where I disagree with your premise.  I think that
Ada allows and encourages a much richer vocabulary than "traditional"
object oriented programming.  I also content from experience that where
Ada 95 and OO part ways for sound reasons, using the richer vocabulary
makes for much more maintainable software, and allows you to build much
larger systems without getting burned by the complexity of the class

    Many years ago, Lori Clarke, et. al. wrote a wonderful paper titled
"Nesting in Ada is for the Birds."  The points they made in that paper about
the disadvantages of deep nesting map directly to the disadvantages of
multiple derivation/sub-classing in Ada 95.  For that matter the same
applies to all OO languages, but in most of them you are stuck with the
problem.  There is a style of Ada 95 OO programming where you have long
mix-in chains in the private part of a package, but that is not part of the
visible interface, where usually you want at most two levels of visible
subclassing.  The only reason I say two is to cover the trees rooted at
Controlled and Limited_Controlled.

>I think it's interesting and exciting that such methodology checking
>could be integrated into an Ada compilation system with pragmas, if
>we want to, without some kludgy thing like a pre-processor.

     Now you are talking.  Build a tool using ASIS to note style
deviations.  But you shouldn't say, "Don't do this."  Instead the warning
messages should
enumerate reasons for violating the guideline.

>I don't know much about ASIS -- might it allow class-checking to be
>integrated into a tool suite WITHOUT being integrated into a specific
>compiler?  Kewl, d00d!

   Possibly, but that isn't a problem.  Any code that won't be analyzed
correctly by any compiler front-end is not code you want to put through
such a tool.

>>     function "+"(L,R: Set) return Set;
>>     function "-"(L: Set, R: Element) return Set;
>>     -- and so on.
>Can you explain what you are trying to communicate with your example?
>I don't understand your point.

    The first operation is a class operation with two parameters and a
result of the class type, the second was to point out that it doesn't
always make sense to order parameters.  Just examples of good practice you
don't want to outlaw if your pragma is used.

>Yep, lots of places except for tagged types, which is how Ada
>implements object-oriented programming, which is the context in
>which people are carping about not having it.

>So as long as you don't want to define a class hierarchy, or do
>programming by extension, or have polymorphism, and you have a
>very limited number of instances, you can use that technique.  I
>don't think you'll win a lot of converts from C++ or Smalltalk
>with that sales pitch.

     What sales pitch?  There are many ways to model classes in Ada 95, for
each potential class you need to choose the right model.  The package as
object model is not right for all cases, but you would be surprised at how
often you use it once you understand it.  And by the way, your first three
restrictions do not apply.  You can have a hierarchy

>I'm not arguing for or against object.method notation.  Some people
>think it's important for clarity, a desideratum I would think we can
>all respect.  I was trying to think up something that might help
>such people transition to Ada.
>"Look," we could say, "you can require that a particular tagged
>type's operations start their parameter list with the controlling
>object.  It makes the receiving object just as obvious as
>object.method notation does.  But if you're just doing programming
>by extension, not object-oriented programming, you don't have to
>live with that limit."
>By the way, perhaps such a "class" package's type should only be
>extensible through other, similarly-limited "class" packages, to
>ensure that all its operations follow the parameter-ordering rule.
>Sam Mize
>Samuel Mize -- [log in to unmask] (home email) -- Team Ada
>Fight Spam: see \\\ Smert Spamonam

                                        Robert I. Eachus

with Standard_Disclaimer;
use  Standard_Disclaimer;
function Message (Text: in Clever_Ideas) return Better_Ideas is...