Re: Is ADA as good for graphics programming as C? (WAS: Re: Avoiding the second historic mistake)

[donh@syd.csa.com.au (Don Harrison)]

Matt Heaney wrote:

:In article <EDAsu7.MLn@syd.csa.com.au>, donh@syd.csa.com.au wrote:
:
:>Yes, through a round-about way using genericity. This is inelegant compared 
:>with multiple inheritance. However, it's simple compared to the obscure 
:>technique for acheiving multiple views in Ada95 Rationale, Section 4.6.3.
:>I'm still trying to comprehend that one!
:
:Be careful bandying about terms like "more elegant" and "less pure."  Those
:are concepts invented by humans to convey an idea to another human.  Nature
:doesn't care about what is more or less elegant, she only cares about what
:works and what doesn't.

Have to disagree here. Appreciation of elegance and purity are God-given
qualities which many either suppress or don't have much of in the first place.
Judging by God's handiwork, He possesses them in abundance. 

IMO, it's software users that don't care about elegance and purity but only 
inasmuch as it doesn't affect them. That is, they couldn't care less if 
software is inelegant under-the-covers so long as it works and is easy to use. 
Unfortunately, what's more likely is that it won't work and won't be easy to 
use if it's design is complex - a poor designer doesn't radically change the 
way they do things when they come to the GUI and complexity tends to compromise
reliability. 

:As system builders, especially as builders of software systems, we always
:have to be concerned with keeping things simple.  

I agree.

:The language designer
:must balance expressiveness of the language against inclusion of yet more
:features.

I tend to think there are good and bad ways of delivering functionality -
see below... 


:Does the putative "elegance" of the MI solution outweigh the language
:complexity that would result had MI been included in Ada?  Doesn't
:simplicity count for something?  

Yes, it counts for a lot. However, I think it's helpful to recognise the 
relative complexity engendered by the two primary means of delivering 
functionality - language mechanisms and libraries. The former increases 
overall complexity more than the latter because language features interact 
with each other *non-orthogonality* and libraries are *standalone*. Also, I 
think we need to distinguish between *actual* and *perceived* orthogonality. 
A naive user, for example, may perceive a language such as Ada to be more 
complex than it actually is simply because they don't know what is orthogonal 
and what isn't. 

As an example of the standalone complexity of libraries (strictly in Ada it's
optional language and libraries), Ada is not made more complex for a 
realtime user due to the existence of an Information Systems Annexe
because it's immediately obvious from the fact that it's optional that it 
doesn't concern them. The same can't be said of language mechanisms whose 
orthogonality is recognised more through use rather than immediate 
comprehension. Most humans can only grasp a handful of concepts at a time 
so it's asking too much of them to immediately comprehend which of a multitude
of language mechanisms are independent of each other. Obviously, the fewer 
there are, the less there is to learn and the quicker you will master the 
language.

I doubt anyone would disagree that Annexes are a good idea. Minimalist 
languages (such as Eiffel) take this principle further by placing as much 
functionality as possible in "Annexes" (called libraries in Eiffel). In this 
way, the basic functionality of the language is reused heavily to provide 
other functionality. The analogy on the back of "Eiffel The Language" (ETL) 
is very apt:

  "Eiffel has been called a 'RISC language'".

Some examples of this reuse which contrast with Ada are:

  - Implementation of basic types as classes.
  - Implementation of parts of the exception facilities as a class.
  - Implementation of various aspects of concurrency (timed interrupts, 
    coroutines etc.) as classes.

The more functionality you can implement in standalone libraries, the simpler
the language itself becomes. A rough measure of that complexity is language 
validity rules. Eiffel only has about 81 including SCOOP. How many does Ada95 
have? (I mean the number of distinct RM references which an Ada compiler can
spit out) I've roughly estimated about 2400 - I can't tell for sure because the 
Librarian has annexed my copy of the LRM. This difference translates fairly 
directly into how complex the language is perceived to be, especially to a 
beginner.


In the specific case of MI, I think it's such a useful modelling tool as to 
warrant its inclusion in the language itself if that will make it easier to use
(which I think it does).


:Should one reject Ada out of hand, because
:mixin inheritance in Ada using generics is "inelegant" compared to using
:multiple inheritence in Eiffel?

No, but it may mean that you're not able to model things as directly as you 
might otherwise.


:Isn't it possible that there are other features of Ada, that allow some
:things to be done more easily or more elegantly than in Eiffel?  

Sure. Low-level facilities may be one such area. But, IME, it's typically the 
other way round.

:Is the
:mechanism used to implement mixin inheritance so important that those other
:features aren't significant?

I think they're likely to be orthogonal wrt MI.


:The multiple view technique is an admittedly difficult idiom to understand
:- at first.  But once apprehended, it makes perfect sense, and isn't hard
:to do.  You get multiple views for free, because it builds on existing
:language mechanisms, and it very nearly gives you what you get in MI,
:without requiring that MI be part of the language.
:
:You may want to study my discussion of the Adapter pattern in the SIGAda
:patterns WG archives, where I specifically discuss and give code examples
:of the multiple views technique.
:
:<http://www.acm.org/sigada>

Thanks for the reference. If I get a chance, I'll take a look.


Don.
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Don Harrison             donh@syd.csa.com.au