Re: use of reals

ma_friedrich@gmx.de (Martina) writes:

> I am working with a program based on SML'97. As I just learned it is
> not possible to use "real" to define a datatype, e.g.
> datatype <name> = real;
> Maybe anyone can explain to me how I can instruct SML'97 to define a
> float number.

I think you are seriously confused:

1. A "datatype" declaration creates a brand-new type. But the type
"real" already exists, so that's probably not what you want anyway.

You could make an alias for "real" by declaring

type foo = real

Doing so, however, is not necessary for being able to work with
real numbers in SML programs.

2. A datatype declaration consists of one or more "cases", each case
introducing a value constructor. Your attempt above misses the
constructor, so this is not even valid SML (regardless of whether
or not you write "real" on the right-hand side).

You could have written something like:

datatype foo = BAR of real

This makes a new type "foo" and gives you a constructor named "BAR"
which turns reals into foos:

BAR 1.0 : foo

Conversely, by pattern-matching against BAR you can extract reals
from foos:

(fn BAR r => r) : foo -> real

3. I don't know what you mean by "define a float number". Maybe you
want to "declare a variable of floating-point type"? If so, you
could just add a type constraint of the form ": real" wherever it
pleases you. Example:

fun float_inc (x : real) = x + 1.0

Notice that in this example the compiler would have figured out all
by itself that x must have type real. Because of the way
overloading resolution works in SML'97, this is not always so. In
some cases you must add a type constraint of some sort to make the
compiler realize you meant "real" and not, e.g., "int". Example:

fun float_add (x : real, y : real) = x + y

(BTW, a constraint on either x or y would have been enough.)

Matthias