We define the following function, whose behavior depends on a boolean parameter Increasing:
f := proc(x)
begin
for i from 1 to 10 do
if Increasing then
x := 2*x
else
x := x/2
end_if
end_for;
x;
end_proc:
partialSpecialization specializes the code of f for a specific value of the parameter:
g := prog::partialSpecialization(f, table(Increasing = TRUE)):
expose(g)
proc(x)
name f;
begin
for i from 1 to 10 do
x := 2*x
end_for;
x
end_proc
As desired, the if statement has been optimized away.
The parameters may be local variables in the calling context instead of global variables. Here, we replay the previous example, but now Increasing is a parameter of an enclosing procedure:
makeProc :=
proc(Increasing : DOM_BOOL) : DOM_PROC
begin
prog::partialSpecialization(
proc(x)
begin
for i from 1 to 10 do
if Increasing then
x := 2*x
else
x := x/2
end_if
end_for;
x
end_proc,
table(hold(Increasing) = val(Increasing)))
end_proc:
expose(makeProc(TRUE))
proc(x)
begin
for i from 1 to 10 do
x := 2*x
end_for;
x
end_proc
Note that in that case makeProc does not need the option escape; indeed, after the specialization the returned procedure does not refer anymore to any variable of makeProc.
Without using the partial specialization, calling makeProc several times with the same parameter produces procedure whose code is physically shared in memory. Only the context of the procedure differ. Partial specialization breaks this: even if the produced procedures have a syntactically identical code, it will not be physically shared in memory.