bcw (11/01/82)
From: Bruce C. Wrigt @ Duke University
Re: Problems in C and Ada
The overloading capabilities of Ada are definitely not the *worst* feature
of the language; in fact there are very few languages which do *not* have
some amount of overloading (usually small). Forth, Bliss, and BCPL are the
only languages which come to my mind immediately which have *no* overloading -
most languages implicitly overload operators like + so that they work with
both integer and floating point arithmetic. Ada and a couple of other
languages allow the user to define additional overloaded operations. I
don't agree with the way it's done (I don't think it should be legal for
the programmer to hide the system-defined instance of the "+" operator for
integers, for example; probably at least the operators should be defined
in some kind of encapsulation where the data type itself is defined and
modification of the operators outside of the encapsulation should probably
be illegal), but it does not strike me as an unreasonable idea.
A much worse problem with Ada (unless the revised report has fixed the
problem) is with the definition of fixed-point data. This is an attempt
to allow a "poor man's floating point" by defining a number with a radix
point (similar to PL/I's fixed point data). The problem is that the old
version of the language did not define the radix in which the fixed-point
data was to be stored; it just allowed the programmer to specify a delta
value which is guaranteed to be greater than or equal to the smallest possible
difference between two numbers. For example, if you specify a delta value
of 0.1, the compiler could comply both if it compiled your data type as a
fixed decimal number with one digit to the right of the decimal point, or
a fixed binary number with four bits to the right of the decimal point
(hence the "actual" delta would be 0.0625). This will cause obscure problems
with portability and verification of code which uses the feature, since
there will be unpredictable rounding effects depending on the implementation.
There were several other problems with the initial report, which I think
have been fixed in the final report, so this may have been fixed as well;
there are several other things which they did which look much more
questionable than overloading.
Bruce C. Wright @ Duke University
PS -- I have noted that many people on the net misspell the name of the
PL/I language. The name is not PL/1, PL-1 or whatever; the ANSI
committee specifically states that the correct spelling is PL/I.
-- BCW