Compiler Construction
It is commonly suggested that Algol-60 call-by-name can be modelled by passing a
function as a call-by-value parameter. Show how a program containing a definition
int f(int x:name) { … x … x … }
of f (where x occurs only in Rvalue context) and a call f(e) to f can be replaced
by an equivalent definition and call using only call-by-value. [6 marks]
Most such explanations assume that the uses of x within f occur only in Rvalue
context. However, Algol-60 also permits the equivalent of
int g(int x:name) { if (p) { … x := x+1; x := -x; … }
return x;
}
and calls like g(a[k()]) which, when p is true, would have the effect of calling
k() five times and consequent access to five (possibly different) subscripts of array
a[]. Develop your explanation for the first part of this question to cover also the
case of a call-by-name parameter being used in both Lvalue and Rvalue contexts.
[Hint: note that when p is false then the actual parameter to g need not be an
Lvalue, so you may need two parameterless procedure arguments (“thunks”).]
[8 marks]
Using the previous part or otherwise, give a translation of a definition and call h(e)
using call-by-value-result (Ada in out mode) with no uses of the address-of (&)
operator other than those involved in call-by-name. Your explanation is allowed to
deviate from call-by-value-result by allowing side-effects in e to take place twice.
[6 marks]