Re: Use of constants on EFA

[dewar@merv.cs.nyu.edu (Robert Dewar)]

<<EF Software Programmer's Manual PL-J-019-V-1101 Issue 4 April 1992
which contains a section saying :

5.4.2.1 Constants
Objects shall be defined as being constant if their value is fixed
throughout the execution of the software.
>>

This is a very poor software design principle. Your coding practices
are the appropriate ones. But as often happens, when someone puts a 
rule into a formal book, even if it is completely misguided as in
this case, it tends to be taken by someone not in a position to
evaluate the value of the rule as gospel.

Note that at least in Ada 95, you do not have to declare constants
to deal with the unconstrained cased, but of course it is still better
practice to use the keyword constant wherever possible.

Undoubtedly the rule was written by someone who did not appreciate the
importance of the use of constant declarations in the manner in which
you suggest. Probably the rule was intended to force the use of constant
in appropriate situations, and it would be a surprise to the author of
the rule to find it being used to *prevent* use of constant in
appropriate situations.

The proper rule for the use of constant in Ada is roughly, use constant
on every object declaration, unless there is some reason that you cannot
do so (i.e. there is some explicit or implicit assignment to the object).

I find that most Ada programmers greatly underuse constant, partly because
the syntax is a little heavy (*). It is most certainly sad to see a style
guide insist on this poor practice. Keep up the fight to have this rule
interpreted in a sensible manner. Particularly so, since, as you note, in
Ada 83, the consequence of not being able to use a constant in the
unconstrained case is extremely painful (indeed in Ada 83, the rule is,
again probably unintentially, removing a critical piece of functionality
from the code.

It is a pity when people replace properly done quality assessment with
simple comparisons to style books, but it happens all the time.

Robert B.K. Dewar
Ada Core Technologies

(*) P.S. in Algol-68, the syntax for a constant is very light:

variable    int a;
constant    int a = 3;

and so the use of constants is much more prevalent in A68 programs (the
syntax for an initilaized variable is  int a := 3;

P.P.S. One very bad consequence of not using constant in the manner you
suggest is that it greatly increases the maintenance burdens. If you follow
this style guideline and write

   procedure x is
	y : integer := ...

instead of

   procedure x is
        y : constant integer := ...

you force the reader/maintainer to either search the entire procedure to
look for an expected assignment to y (and to be irritated when none is
found), or to read the code constantly worrying that y might be changing.

Dynamic constants are an extremely important feature of a language like
Ada. Forbidding them is seriously misguided.

I do think you should try to dig into the origins of this rule. I am almost
certain that the motiviation is to discourage people from failing to use
constant on global values that do not change, in otherwords that an expected
violation of this rule involves the failure to use the constant keyword.