1. Louden, 20031 Chapter 8 - Control II: Procedures and Environments

2. Chapter 8K. Louden, Programming Languages2 Three major parts of a runtime environment: Static area allocated at load/startup time. Examples: global/static variables and load- time constants. Stack area for execution-time data that obeys a last-in first-out lifetime rule. Examples: method variables and temporaries. Heap or dynamically allocated area for "fully dynamic" data, e.g. data allocated with new

3. Chapter 8K. Louden, Programming Languages3 Procedure Overview When functions are “first class” data items themselves, they can be dynamically created and used like values just like any other data structure. (e.g., Haskell curried functions, eval) pass functions as arguments A procedure is called or activated. Activation record: collection of data needed to maintain a single execution of a procedure. access to local and non-local references. Static or dynamic environment (depending on scoping) must be accessible during runtime. When a procedure depends only on parameters and fixed language features – closed form. The code for a function together with its defining environment is called closure – as we can resolve all outstanding non-local environments.

4. At seats: Using what you understand about the call stack, what actions take place on call? What actions take place on return? Chapter 8K. Louden, Programming Languages4

5. Chapter 8K. Louden, Programming Languages5 Implementing “Simple” Subprograms Caller responsibilities 1. Save the execution status of the caller (calling environment) 2. Carry out the parameter-passing process by putting the parameters somewhere that the called function can access. 3. Pass the return address of the caller to the callee 4. Transfer control to the callee

6. Chapter 8K. Louden, Programming Languages6 Implementing “Simple” Subprograms Return Actions: 1. If it is a function, move the return value to a place the caller can get it 2. Restore the execution status of the caller 3. Transfer control back to the caller

7. Chapter 8K. Louden, Programming Languages7 Implementing “Simple” Subprograms Called routine must Create an activation record: –local variables –return address –points to other environments –parameters –See CSILM lesson “Call Stack” for an example csilm.usu.edu, Browse Resources, CSILM Activities, Programming Languages, Call Stack

8. Chapter 8K. Louden, Programming Languages8 Parameter Passing Aliases may be created Type checking parameters – for legality and to pick between overloaded methods a reference parameter is a nonlocal variable the same data object passed for two parameters CALL S(X,X) causes aliasing With aliasing, interesting problems in optimizations occur. x+2 y++ x+2 If x and y are aliases, x+2 isn’t a common subexpression

9. Chapter 8K. Louden, Programming Languages9 Models of Parameter Passing

10. Chapter 8K. Louden, Programming Languages10 1. Pass-by-value (in mode) –Typically we copy the value in, but can do with a constant reference pointer. –Parameters are viewed as local variables of the procedure –Disadvantages of copy: Requires more storage (duplicated space) Cost of the moves (if the parameter is large) –Disadvantages of constant reference: Must write-protect in the called subprogram or compiler check that there are no assignments. Accesses cost more (indirect addressing)

11. Chapter 8K. Louden, Programming Languages11 2. Pass-by-result (out mode) –function return value(s) –Local’s value is passed back to the caller –Physical move is usually used (copy to call stack) –Disadvantages: If value is passed, time and space costs to copy order dependence may be a problem (if output values are aliased) procedure sub1(y: int, z: int);{y=0;z=5; } sub1(x, x); –Value of x in the caller depends on order of assignments at the return

12. Chapter 8K. Louden, Programming Languages12 3. Inout mode – Pass by value-result (aka copy-in copy- out or copy-restore) –Used to save cost of indirect access. Physical move, both ways value-result (or pass by copy) Disadvantages –ordering may be a problem with a call like doit(x,x) –time/space issues –Need to know whether address is computed again before copying back. doit(i,a[i])

13. Chapter 8K. Louden, Programming Languages13 4.Pass by reference. Issues: –passing is faster (as no data copy) formal parameter is local object of type pointer If expression is passed as an in/out parameter: a temporary location may be passed (and then the copy is changed, not the original) Disadvantages: –access slower as is indirect (always follow a pointer to access), but passing is fast (only copy a pointer, not a whole structure) –may make inadvertent changes to parameters

14. Chapter 8K. Louden, Programming Languages14 5. Pass-by-name (Delayed evaluated parameters) –By textual substitution –parameter is evaluated everytime it is used, but it is evaluated in the caller’s environment –Purpose: flexibility of late binding –costly Thunks: a pass by name argument is implemented by a little procedure (called a thunk) which evaluates the argument. A thunk is a method to evaluate an expression that is yet to be evaluated.

15. Chapter 8K. Louden, Programming Languages15 Pass-by-name How is it like other methods? If actual is a scalar variable, it is pass-by-reference If actual is a constant expression, it is pass-by-value If actual is an array element, it is like nothing else e.g. procedure sub1(x: int; y: int); begin x := 1; Seems like nothing is happening y := 2; with first assignments but it is x := 2; y := 3; end; sub1(i, a[i]);

16. Chapter 8K. Louden, Programming Languages16 Example: procedure R(var i,j: integer); begin m:boolean; m := true; i := i + 1; j := j + 1; end; m := 2; for(i=0;i<10;i++) c[i]=10*i; R(m,c[m]); pass by reference: adds 1 to m and c[2] Pass by name: adds 1 to m and c[3]

17. Chapter 8K. Louden, Programming Languages17 Parameter Passing Methods Design Considerations for Parameter Passing 1. Efficiency 2. One-way or two-way –These two are in conflict with one another! –Good programming => limited access to variables, which means one-way whenever possible –Efficiency => pass by reference is fastest way to pass structures of significant size –Also, functions should not allow reference parameters

18. Chapter 8K. Louden, Programming Languages18 Languages and Environments Languages differ on where activation records must go in the environment: –Fortran is static: all data, including activation records, are statically allocated. (Each function has only one activation record—no recursion!) –Functional languages (Scheme,ML) and some OO languages (Smalltalk) are heap-oriented: all (or almost all) data, including activation records, are allocated dynamically. –Most languages are in between: data can go anywhere (depending on its properties); activation records go on the stack.

19. Chapter 8K. Louden, Programming Languages19 Simple stack-based allocation Described in Chapter 5. Nested declarations are added to the stack as their code blocks are entered, and removed as their code blocks are exited. Example:Stack at Point 1: { int x; int y; { int z; } { int w; // Point 1 } } Note,z has been removed at point 1 as have exited scope w y x

20. Chapter 8K. Louden, Programming Languages20 Example (C): main →q →p int x; void p( int y) { int i = x; char c;... } void q ( int a) { int x; p(1); } main() { q(2); return 0; }

21. Chapter 8K. Louden, Programming Languages21 Local variable access using the ep In a typical language with a stack-based runtime environment, the local declarations in a procedure are fixed at compile-time, both in size and in sequence. This information can be used to speed up accesses to local variables, by precomputing these locations as offsets from the ep. Then the local frame need not have a name-based lookup operation (unlike the symbol table). In fact, names can be dispensed with altogether.

22. Chapter 8K. Louden, Programming Languages22 Non-local variable access Requires that the environment be able to identify frames representing enclosing scopes. Using the control link results in dynamic scope (and also kills the fixed-offset property as you are not sure which method will contain the x. Thus, you can’t depend on a fixed location). If procedures can't be nested (C, C++, Java), the enclosing scope is always locatable by other means: it is either global (accessed directly) or belongs to the current object. If procedures can be nested, to maintain lexical scope a new link must be added to each frame: the access link, pointing to the activation of the defining environment of each procedure.

23. Chapter 8K. Louden, Programming Languages23 Nested Subprograms The process of locating a nonlocal reference: 1. Find the correct activation record instance 2. Determine the correct offset within that activation record instance May need to follow several links (access chaining) The number of links is known from compile time. If used stack of symbol tables, can count how many tables you had to search to find it. If used individual stacks for each value, you can record the nesting depth of each variable.

24. Chapter 8K. Louden, Programming Languages24 Procedure values as pointer pairs Each procedure becomes a pair of pointers: a code pointer (called the instruction pointer or ip in the text), and an environment pointer (ep in the text) pointing to the definition environment of the procedure (which will become the access link during a call). Such an pair is sometimes called a closure.

25. Chapter 8K. Louden, Programming Languages25 The Process of Locating a Nonlocal Reference Finding the offset is easy Finding the correct activation record instance: –Static semantic rules guarantee that all nonlocal variables that can be referenced have been allocated in some activation record instance that is on the stack when the reference is made

26. Chapter 8K. Louden, Programming Languages26 Nested Subprograms Technique 1 - Static Chains A static chain is a chain of static links that connects certain activation record instances The static link in an activation record instance for subprogram A points to one of the activation record instances of A's static parent The static chain from an activation record instance connects it to all of its static ancestors

27. Chapter 8K. Louden, Programming Languages27 Static Chains Show the static/dynamic chains when main →C →A →B →C main ----- static_depth = 0 A ----- static_depth = 1 B ----- static_depth = 2 C ----- static_depth = 1

28. Chapter 8K. Louden, Programming Languages28 Nested Subprograms Static Chain Maintenance –At the call : The activation record instance must be built The dynamic link is just the old stack top pointer The static link must point to the most recent ari of the static parent (in most situations) –Two Methods to set static chain: 1. Search the dynamic chain until the first ari for the static parent is found--easy, but slow

29. Chapter 8K. Louden, Programming Languages29 Nested Subprograms 2. Treat procedure calls and definitions like variable references and definitions (have the compiler compute the nesting depth, or number of enclosing scopes between the caller and the procedure that declared the called procedure; store this nesting depth and send it with the call) –e.g. Look at MAIN_2 and the stack contents. At the call to SUB1 in SUB3, this nesting depth is 1, which is sent to SUB1 with the call. The static link in the new ari for SUB1 is set to point to the ari that is pointed to by the second static link in the static chain from the ari for SUB3

30. Chapter 8K. Louden, Programming Languages30 Nested Subprograms Evaluation of the Static Chain Method –Problems: 1. A nonlocal reference is slow if the number of scopes between the reference and the declaration of the referenced variable is large 2. Time-critical code is difficult, because the costs of nonlocal references are not equal, and can change with code upgrades and fixes

31. Chapter 8K. Louden, Programming Languages31 Nested Subprograms Technique 2 (for locating non-local variables with static scope) - Displays –The idea: Put the static links in a separate stack called a display. The entries in the display are pointers to the ari's that have the variables in the referencing environment. –Represent references as (display_offset, local_offset) where display_offset is the same as chain_offset –Can access via computation. display offset of 10 is one lookup (not a chain of length 10)

32. Chapter 8K. Louden, Programming Languages32 Main – level 0 q level 3 p p level1 s level 2 r level 2 t level 1

33. Chapter 8K. Louden, Programming Languages33 main t s q 100Level 0 200Level 1 300Level 2 400Level 3 main-> t -> s-> q When s calls q, a single element is added to the table. 100 200 300 400 Stack is shown growing downwards. Display contains pointers to each activation record at each reachable level

34. At your seats… Why do we do this? This seems well and good – but how do we keep it current? Chapter 8K. Louden, Programming Languages34

35. Chapter 8K. Louden, Programming Languages35 main t s q 100Level 0 500Level 1 300Level 2 400Level 3 main-> t -> s-> q-> p When q calls p, a new level 1 entry is needed. Store the old one, so you can get it back. Level 2 and level 3 are unused (but unchanged) 100 200 300 400 p old level 1 is 200 500

36. Chapter 8K. Louden, Programming Languages36 main t s q 100Level 1 600Level 2 300Level 3 400Level 4 main-> t -> s-> q-> p->t When p calls t, a new level 2 entry is needed Level 3 and level 4 are unused (but unchanged) 100 200 300 400 p old level 2 is 200 500 old level 2 is 500 t 600

37. Chapter 8K. Louden, Programming Languages37 main t s q 100Level 1 600Level 2 300Level 3 400Level 4 main-> t -> s-> q-> p->t When p calls t, a new level 2 entry is needed Level 3 and level 4 are unused (but unchanged) 100 200 300 400 p old level 2 is 200 500 old level 2 is 500 t 600

38. Chapter 8K. Louden, Programming Languages38 Blocks Two Methods: 1. Treat blocks as parameterless subprograms –Use activation records 2. Allocate locals on top of the ari of the subprogram –Must use a different method to access locals –A little more work for the compiler writer

39. Chapter 8K. Louden, Programming Languages39 Implementing Dynamic Scoping 1. Deep Access (search) - nonlocal references are found by searching the activation record instances on the dynamic chain –Length of chain cannot be statically determined –Every activation record instance must have variable names recorded

40. Chapter 8K. Louden, Programming Languages40 Implementing Dynamic Scoping 2. Shallow Access - put locals in a central place How implemented? a. One stack for each variable name b. Central referencing table with an entry for each variable name At subprogram entry, add location for each variable. At subprogram exit, remove location for each variable.

41. Chapter 8K. Louden, Programming Languages41 Using Shallow Access to Implement Dynamic Scoping

42. Chapter 8K. Louden, Programming Languages42 Parameter Passing Conventions Actual/Formal Parameter Correspondence: 1. Positional (this is what we are used to) 2. Keyword –e.g. SORT(LIST => A, LENGTH => N); –Advantage: order is irrelevant –Disadvantage: user must know the formal parameter’s names 3. Default Values: e.g. procedure SORT(LIST : LIST_TYPE; LENGTH : INTEGER := 100);... SORT(LIST => A);

43. Chapter 8K. Louden, Programming Languages43 Overloaded Subprograms Def: An overloaded subprogram is one that has the same name as another subprogram in the same referencing environment C++ and Ada have overloaded subprograms built-in, and users can write their own overloaded subprograms Overloaded subprograms provide ad hoc (non-generalizable) polymorphism

44. Chapter 8K. Louden, Programming Languages44 Generic Subprograms Analogy: Would you rather have: – Vacuum that can adjust to any carpet height versus –A different vacuum for each type of carpet. A generic or polymorphic subprogram is one (not many) subprogram that takes parameters of different types on different activations A subprogram that takes a generic parameter needs to figure out what type was passed (Haskell)

45. Chapter 8K. Louden, Programming Languages45 Separate & Independent Compilation Def: Independent compilation is compilation of some of the units of a program separately from the rest of the program, without the benefit of interface information Def: Separate compilation is compilation of some of the units of a program separately from the rest of the program, using interface information to check the correctness of the interface between the two parts.

46. Chapter 8K. Louden, Programming Languages46 Memory Management Benefit (of system controlled storage management): ability to delay the binding of a storage segment's size and/or location reuse of a storage segment for different jobs (from system supervisor point of view) reuse of storage for different data structures increased generality, not have to specify maximum data structure size dynamic data structures recursive procedures - garbage collection is automatic Adv/Disadv of programmer controlled storage management Disadvantage: burden on programmer & may interfere with necessary system control May lead to subtle errors May interfere with system-controlled storage management Advantage: difficult for system to determine when storage may be most effectively allocated and freed

47. Chapter 8K. Louden, Programming Languages47 Heap management –Single-size cells vs. variable-size cells –Reference counters (eager approach) vs. garbage collection (lazy approach) 1. Reference counters: maintain a counter in every cell that store the number of pointers currently pointing at the cell –Disadvantages: space required, complications for cells connected circularly Expensive - when making a pointer assignment p=q 1.decrement count for old value of p 2.if 0, return to free storage. Check if contains references to other blocks. Could be recursive 3.do pointer assignment 4.Increment reference count for q

48. Chapter 8K. Louden, Programming Languages48 One-bit reference counting Another variation on reference counting, called one-bit reference counting, uses a single bit flag to indicate whether each object has either "one" or "many" references. If a reference to an object with "one" reference is removed, then the object can be recycled. If an object has "many" references, then removing references does not change this, and that object will never be recycled. It is possible to store the flag as part of the pointer to the object, so no additional space is required in each object to store the count. One-bit reference counting is effective in practice because most actual objects have a reference count of one.

49. Chapter 8K. Louden, Programming Languages49 2. Garbage collection: allocate until all available cells allocated; then begin gathering all garbage –Every heap cell has an extra bit used by collection algorithm –All cells initially set to garbage –All pointers traced into heap, and reachable cells marked as not garbage –All garbage cells returned to list of available cells Disadvantages: when you need it most, it works worst (takes most time when program needs most of cells in heap)

50. Chapter 8K. Louden, Programming Languages50 Mark-Sweep - Java uses In a mark-sweep collection, the collector first examines the program variables; any blocks of memory pointed to are added to a list of blocks to be examined. For each block on that list, it sets a flag (the mark) on the block to show that it is still required, and also that it has been processed. It also adds to the list any blocks pointed to by that block that have not yet been marked. In this way, all blocks that can be reached by the program are marked. In the second phase, the collector sweeps all allocated memory, searching for blocks that have not been marked. If it finds any, it returns them to the allocator for reuse Can find circular references. Easy if regular use of pointers (Like in LISP) All elements must be reachable by a chain of pointers which begins outside the heap Have to be able to know where all pointers are - both inside the heap and outside. How can a chain be followed from a pointer if there is no predefined location for that pointer in the pointed-to cell?

51. Chapter 8K. Louden, Programming Languages51 Conservative garbage collection Although garbage collection was first invented in 1958, many languages have been designed and implemented without the possibility of garbage collection in mind. It is usually difficult to add normal garbage collection to such a system, but there is a technique, known as conservative garbage collection, that can be used. The usual problem with such a language is that it doesn't provide the collector with information about the data types, and the collector cannot therefore determine what is a pointer and what isn't. A conservative collector assumes that anything might be a pointer. It regards any data value that looks like a pointer to or into a block of allocated memory as preventing the recycling of that block. You might think that conservative garbage collection could easily perform quite poorly, leaving a lot of garbage uncollected. In practice, it does quite well, and there are refinements that improve matters further. Because references are ambiguous, some objects may be retained despite being actually unreachable. In practice, this happens rarely, and refinements such as black-listing can further reduce the odds.

52. Chapter 8K. Louden, Programming Languages52 Incremental Garbage Collection Older garbage collection algorithms relied on being able to start collection and continue working until the collection was complete, without interruption. This makes many interactive systems pause during collection, and makes the presence of garbage collection obtrusive. Fortunately, there are modern techniques (known as incremental collection) to allow garbage collection to be performed in a series of small steps while the program is never stopped for long. In this context, the program that uses and modifies the blocks is sometimes known as the mutator. While the collector is trying to determine which blocks of memory are reachable by the mutator, the mutator is busily allocating new blocks, modifying old blocks, and changing the set of blocks it is actually looking at.

53. Chapter 8K. Louden, Programming Languages53 Heap Storage Management - Variable Sized Elements Memory operations Initially one large block Free space list, as space is recovered allocate from free list: first fit, best fit, worst fit compact: maintain a length of each block recover via explicit, garbage collection or reference counts Need length of each to locate pieces and coalesce fragmentation partial compaction (coalescing of adjacent free blocks) full compaction (move blocks) Show how