Subprograms

The basic abstraction mechanism. Functions correspond to the mathematical notion of computation input output procedures affect the environment, and are called for their side-effects pure functional model possible (but awkward) hybrid model most common: functions can have (limited) side effects.

The environment of the computation

Declarations introduce names that denote entities. At execution-time, entities are bound to values or to locations: name value (functional) name location value (imperative)
Value binding takes place during function invocation Names are bound to locations on scope entry. Locations are bound to values by assignment.

Parameter passing
The rules that describe the binding of arguments to formal parameters, i.e. the meaning of a reference to a formal in the execution of the subprogram. By-value: formal is bound to value of actual by-reference: formal is bound to location of actual by name: formal is bound to expression for actual

Functional Programming
All parameter-passing by value no assignment. local declarations of constants only. consequence: functions have no side-effects. referential transparency: two occurrences of the same expression have the same meaning. awkward if need to describe computations with history, e.g. a random number generator.

Parameter passing in Ada

Separate the semantic intent from implementation. Parameter modes:
in : read-only in subprogram out : write in subprogram in out : read-write in subprogram

independent of whether binding by value, by reference, or by copy-return. Functions can only have in parameters history: can assign to global variables.

Syntactic sugar
Default values for in-parameters (Ada)
function Incr (base: integer; delt : integer :=1) return integer;

incr (a(j)) equivalent to (incr (a(j), 1);

Also available in C++

int f (int first, int second =0, char* handle = 0);

Named associations (Ada):

incr (delt => 17, base => a(I));

Parameter passing in C
C: parameter passing by value, no semantic checks. Assignment to formal is assignment to local copy If argument is pointer, effect is similar to passing designated object by reference
void incr (int* x) { (*x) ++; } incr (&counter); /* pointer to counter*/

no need to distinguish between functions and procedures: void indicates side-effects only.

Parameter-passing in C++
Default is by-value (same semantics as C) Explicit reference parameters: void incr (int& y) { y++}; incr (counter); // compiler knows profile of incr, builds reference semantic intent indicated by qualifier: void f (const double& val); // in-parameter by reference: call cannot modify it

Parameter-passing in Java
By value only. Semantics of assignment differs for primitive types and for classes: primitive types have value semantics objects have reference semantics Consequence: methods can modify objects. No way express semantic intent on primitive types: assignment allowed, affects local copy. Can express semantic intent on objects: final means that formal is read-only.

Parameter Passing in C#
As in Java: by-value is default. For class types this means the reference is immutable, but not the object Parameter can indicate intent:
out double X ; ref int X; // uninitialized, by reference // in out, by reference

Block structure
procedure outer (x : integer) is y : boolean; procedure inner (z : integer) is x : float := 3.0; -- hides outer.x function innermost (v : integer) return float is begin return x * float (v * outer.x); -- use inner.x and outer.x end innermost; begin x := innermost (z); -- assign to inner.x end inner; begin inner (x); -- outer.x, the other one is out of scope end;

Parameter passing anomalies

var global : integer := 10; another : integer := 2; procedure confuse ( var first, second : integer); begin first := first + global; second := first * global; end; begin confuse (global, another); first and global will be aliased end different result if by reference or by copy-return Semantics should not depend on implementation of parameter passing passing by value with copy-return is less error-prone

Storage outside of the block

With block structure, the lifetime of an entity coincides with the invocation of the enclosing construct. If the same entity is to be used for several invocations, it must be global to the construct Simplest: declare in the outermost context. Three storage classes:
static stack-based (automatic) heap-allocated

Bounded Nesting
C: no nested functions. Blocks are merged with activation record of enclosing function. Static storage available. Ada: arbitrary nesting of packages and subprograms. Packages provide static storage. early C++, Java: 3 levels: static objects, class members, entities local to a member. current C++, Java: nested classes provide arbitrary nesting

Run-time organization
Each subprogram invocation creates an activation record. Recursion imposes stack allocation (all languages today) Activation record hold actuals, linkage information, saved registers, local entities. caller: place actuals on stack, return address, linkage information, then transfer control to callee. Prologue: save registers, allocate space for locals Epilogue: place return value in register or stack position, update actuals, restore registers, then transfer control to caller. Binding of locations: actuals and locals are at fixed offsets from frame pointers complications: variable no. of actuals, dynamic objects.

Activation record layout

actual actual Frame pointer

Handled by caller

Return addr
Save area local local

Handled by callee

Stack pointer

Functions with variable number of parameters

printf (this is %d a format %d string, x, y); within body of printf, need to locate as many actuals as placeholders in the format string. Solution: place parameters on stack in reverse order. Actuals at positive offset from FP, locals at negative offset from FP.
actual n actual n-1 actual 1 (format string) return address

Objects of dynamic size

declare x : string (1..N); -- N global, non-constant y : string (1..N); begin... where is the start of y in the activation record? Solution 1: use indirection: activation record hold pointers. Simpler implementation, costly dynamic allocation., deallocation. solution 2: local indirection: activation record holds offset into stack. Faster allocation/deallocation, complex implementation.

Run-time access to globals

procedure outer is global : integer; procedure inner is local : integer; begin ... if global = local then -- recursive
-- recursive

-- how do we locate global?

Need run-time structure to locate activation record of statically enclosing scopes. Environment includes current activation record AND activation records of parent scopes.

Global linkage
Static chain: pointer to activation record of statically enclosing scope display: array of pointers to activation records does not work for function values functional languages allocate activation records on heap may not work for pointers to functions simpler if there is no nesting (C, C++, Java) can check static legality in many cases (Ada)

Static Links
Activation record hold pointer to activation record of enclosing scope. Setup as part of call prologue.
To enclosing scope To retrieve entity 3 frames out: 3 dereference operations

outer outer outer

inner inner inner inner

Display
Global array of pointers to current activation records
outermost

display

...
outer outer outer
inner inner inner inner

To retrieve entity 3 frames out: one indexing operation

Subprogram parameters
type proc is access procedure (X : Integer); procedure Perform (Helper : proc) is begin proc (42); end; procedure Action (X : Integer) is procedure Proxy is begin Perform (ActionAccess); end; Proxy can see (and call) Action, therefore environment of Action ( e.g static link) Is known to Proxy. Proxy transmits to Perform both a pointer to the code for Action, and the proper static link for it. -- Access creates pair (ptr to Action , environment of Action)

simplest implementation if Env is pointer (static link) but can also be display: more efficient to retrieve non-local entities, less efficient for subprogram parameters, because array of variable size.

Subprogram parameters in C/C++

void (*pf) (string); // pf is a pointer to a void function that takes a string argument typedef void (*PROC)(int); // type abbreviation clarifies syntax void use_it (PROC); PROC ptr = &do_it; use_it (ptr); use_it (&do_it);

Subprogram parameters in Java

No notion of pointer, so no immediate translation Dynamic dispatching on a virtual method is indirect call, so can be obtained using interfaces, classes, and extensions (to be continued!)

The limits of stack allocation

type ptr is access function (x : integer) return integer; function make_incr (x : integer) return ptr is function new_incr (base : integer) return integer is begin return base + x; -- reference to formal of make_incr end; begin return new_incraccess; -- will it work? end; add_five: ptr := make_incr (5); total : integer := add_five (10); -- where does add_five find x ?

Functions as first-class values forces heap allocation of activation record

The environment of definition of the function must be preserved until the point of call: activation record cannot be reclaimed if it creates functions. Functional languages require more complex run-time management. Higher-order functions: functions that return (build) functions, are powerful but complex mechanisms. Imperative languages restrict their use. A function that returns a pointer to a function is a higher-order function.

Higher-order functions
Both arguments and result can be (pointers to) subprograms:
type Func is access function (x : integer) return integer; function compose (first, second : Func) return Func is begin function result (x : integer) return integer is begin return (second (first (x)); -- implicit dereference on call end; -- in C++ as well. begin return result access; -- but first and second wont exist at the point end; -- of call, so illegal in Ada.

Restricting higher-order functions

C: no nested definitions, so environment is always global. C++: ditto, except for nested classes. Ada: static checks to reject possible dangling references Modula: pointers to function illegal if function not declared at top-level. LISP: special syntax to indicate capture of environment ML, Haskell: no restriction: compose is a primitive

Returning composite values

Intermediate problem: functions that return values of non-static sizes:
function conc3 (x, y, z : string) return string is begin return x & : & y & : & z; end; example : string := conc3 (this, that, theother);

best not to use heap, but still need indirection. Simplest : forbid it (Pascal, C) or use heap automatically (Java)