Ch. 7: Generic abstraction. Generic units and instantiation. Generic units and instantiation. Generic units and instantiation

Transcription

1 Ch. 7: Generic abstraction Generic units and instantiation Reusable program units are applicable in a variety of applications stack queue list set such program units are generic wrt the data the process A generic unit is a program unit parameterized wrt entities on which it depends Instantiation of a generic unit generates an ordinary program unit Generic units can be instantiated more than once prevents code duplication / Faculteit Wiskunde en Informatica PAGE 0 / Faculteit Wiskunde en Informatica PAGE 1 Generic units and instantiation Generic units and instantiation A procedure is an abstraction over an expression A generic unit is an abstraction over a declaration Generic units are supported by: C++ Ada Java (since 2004 and only for types) C++ generic class templates template <int capacity> class Queue { private: char elems[capacity]; int front, rear, length; public: Queue(); void add (char e); char remove(); / Faculteit Wiskunde en Informatica PAGE 2 / Faculteit Wiskunde en Informatica PAGE 3

2 Generic units and instantiation Generic units and instantiation C++ generic function templates template <int capacity> Queue<capacity>::Queue() { front = rear = length template <int capacity> void Queue<capacity>::add(char e) { elems[rear] = e; rear = (rear+1) % capacity; length++; template <int capacity> void Queue<capacity:: remove() { Every constructor and method definition must be prefixed by template <int capacity> Instantiation of generic classes via: Typedef Queue<80> Input_Buffer; Typedef Queue<120> Line_Buffer; Declaration of variables of type Input_Buffer and Line_Buffer: Input_Buffer inbuf; Line_ Buffer outbuf; Line_Buffer errbuf; Alternatively: Queue<80> inbuf; Queue<120>outbuf; e<120>o Queue<120>errbuf; / Faculteit Wiskunde en Informatica PAGE 4 / Faculteit Wiskunde en Informatica PAGE 5 Generic units and instantiation Type and class parameters A conceptual problem with C++ generic classes: Type equivalence: outbuf and errbuf are equivalent because both are derived from Queue<120> Queue<m> and Queue<n-1> are not equivalent arguments must be evaluated at compile-time A program unit that uses a type can also become a generic program unit, with a type parameter Ada, C++, and Java have generic units with type parameters A pragmatic problem with C++ generic classes: if Queue<120> occurs multiple times, multiple instance may be generated by a simple-minded i d compiler / Faculteit Wiskunde en Informatica PAGE 6 / Faculteit Wiskunde en Informatica PAGE 7

3 Type parameters in C++ Type parameters in C++ A C++ generic unit may be parameterized wrt any type or calls on which it depends Encapsulation of homogeneous lists template <class Element> class List is private: const int capacity = ; int length; Element elems[capacity]; public: List(); void append(element e); //other methods <class Element> states that Element is a formal parameter of the generic class and denotes an unknown type generic class constructor and methods: template <class Element> List<Element>::List () { length = 0; template <class Element> void List<Element>::append(Element e) { elems[length++] = e; instantiated via typedef List<char> Phrase or struct Trans { ; typedef List<Trans> Trans_List; / Faculteit Wiskunde en Informatica PAGE 8 / Faculteit Wiskunde en Informatica PAGE 9 Type parameters in C++ Type parameters in C++ Encapsulation of sequences, i.e., sortable lists template t <class Element> class List is private: const int capacity = ; int length; Element elems[capacity]; public: sequence(); void append(element e); void sort(); //other methods Generic class constructor and methods: template <class Element> void Sequence<Element>::sort() { Element e; if (e < elems[i]) Note the use of the < operator in the sort method / Faculteit Wiskunde en Informatica PAGE 10 / Faculteit Wiskunde en Informatica PAGE 11

4 Type parameters in C++ Type parameters in C++ The argument type <class Element> should be equipped with a < operation A proper instantiation is: typedef Sequence<float> Number_Sequence; Number_Sequence readings; readings.sort(); A seemingly proper instantiation is: typedef char* String; typedef Sequence<String> String_Sequence; however < < operation compares pointers instead of lexicographically Operations used for T in the generic unit operations with which the argument type is equipped The C++ compiler cannot completely type-check the definition of a generic unit An incorrect instantiation is: struct Trans { ; typedef Sequence<Trans> Trans_ Sequence; Trans type is not equipped with < operation / Faculteit Wiskunde en Informatica PAGE 12 / Faculteit Wiskunde en Informatica PAGE 13 Class parameters in Java Class parameters in Java Since 2004 Java supports generic abstraction in the form of generic classes, parameterized with other classes encapsulation of homogeneous lists class List <Element> { private length; private Element[] elems; public List() { public void append(element e) { heading states that Element is a formal parameter of the generic class List Generic class can be instantiated with: List<Character> sentence; List<Transaction> sentence; Argument must be a class, not a primitive type List<char> sentence; // illegal! / Faculteit Wiskunde en Informatica PAGE 14 / Faculteit Wiskunde en Informatica PAGE 15

5 Class parameters in Java Class parameters in Java If the generic Java class must be equipped with particular operations, the class parameter must be specified as an interface The class parameter is bounded by the interface Java generic class with a bounded class parameter interface Comparable <Item> { public abstract int compareto(item that); Following Java generic class encapsulates sequences equipped with a sort operation class Sequence <Element implements Comparable<Element>> { private length; private Element[] elems; public Sequence() { public void append() { public void sort() { Element e; if (e.compareto(elems[i]) < 0 ) / Faculteit Wiskunde en Informatica PAGE 16 / Faculteit Wiskunde en Informatica PAGE 17 Class parameters in Java Ch.8: Type systems A Java generic unit is to have a class parameter C of the form class C implements Interface The compiler checks the generic unit to ensure operations used for C in the generic unit operations declared for Interface operations declared in Interface operations with which the argument class is equipped Java generic units are based on type theory: compiler can type-check the declarations of each generic unit separately type-check every instantiation Weakness of Java generic classes: only classes as parameter Older languages g had very simple type systems C s type system is too weak Pascal s type system is too rigid Modern languages more powerful type systems: subtypes polymorphism overloading / Faculteit Wiskunde en Informatica PAGE 18 / Faculteit Wiskunde en Informatica PAGE 19

6 Inclusion polymorphism Inclusion polymorphism Inclusion polymorphism is a type system where types may have subtypes A type T is a set of values equipped with some operations A subtype of T is a subset of values equipped with the same operations as T If we know that a variable only ranges over a subset of values of T, it is better to allow a subtype declaration C has rudimentary subtypes: char and int are subtypes of long float is a subtype of double General properties of subtypes S is subtype of T if every value of S is also a value of T: S T subtype S inherits the operations of type T T 1 is compatible with T 2 T 1 is a subtype of T 2 T 2 is a subtype of T 1 T 1 and T 2 are subtypes of some other type / Faculteit Wiskunde en Informatica PAGE 20 / Faculteit Wiskunde en Informatica PAGE 21 Inclusion polymorphism Classes and subclasses If T 1 is compatible with T 2 T 1 is a subtype of T 2, no run-time check is necessary T 1 is not a subtype of T 2 some of fthe values of ft 1 are values of T 2 A run-time check is needed to check whether a value of T 1 is also a value of T 2 This kind of run-time check is cheaper the full run- time checking as for dynamic languages Class S is a subclass of class C a class C is a set of objects with variable components and operations each object of class S inherits all variable components and operations, but may add extra ones Subclasses are not exactly analogous to subtypes objects of class S may be used wherever objects of class C are expected S may have additional components, so the objects of S are not a subset of the objects of C / Faculteit Wiskunde en Informatica PAGE 22 / Faculteit Wiskunde en Informatica PAGE 23

7 Parametric polymorphism Parametric polymorphism Monomorphic procedures can only operate on arguments of a fixed type Polymorphic procedures can operate uniformly on arguments of a family of types Parametric polymorphism is a type system in which we can write polymorphic procedures functional languages, ML, Haskell, etc., provide parametric ti polymorphism A type variable is an identifier that stands for a family of types monomorphic second(x : Int, y : Int) = y polymorphic second(x : δ, y : τ) = y A polytype derives a family of similar types δ τ τ includes Integer Boolean Boolean String String String but not Boolean Integer Boolean Integer Integer / Faculteit Wiskunde en Informatica PAGE 24 / Faculteit Wiskunde en Informatica PAGE 25 Parameterized types Type inference A parameterized type takes other type(s) as parameter(s) In C array typesτ[]are parameterized types: char[] float[] All programming languages have built-in parameterized types ML and Haskell allow programmer to define their own parameterized types type Pair τ = (τ, τ) data List τ = Nil Cons(τ, List τ) plus polymorphic functions head, tail, length on these lists Statically type programming languages insist on explicit declaration of the type of entity in programs integer I := E In Haskell we can write a definition I = E where type of I is not explicitly stated, but inferred from E Type inference is a process where the type of a declared entity is inferred instead of explicitly stated monotype, if strong clues, e.g., if a monomorphic operator is used polytype in case of polymorphic operators / Faculteit Wiskunde en Informatica PAGE 26 / Faculteit Wiskunde en Informatica PAGE 27

8 Type inference Type inference even n = (n mod 2 = 0) type of even is: Integer Integer Integer id x = x type of id is: τ τ f. g = \x -> f(g(x)) types of f and g are β γ and α β, respectively variable x must be of type α subexpression f(g(x)) is of type γ expression \x -> f(g(x)) is of type α γ. is of type (β γ) (α β) (α γ) Type inference may lead to programs that are difficult to understand Explicitly declaring (redundant) types is good programming practice / Faculteit Wiskunde en Informatica PAGE 28 / Faculteit Wiskunde en Informatica PAGE 29 Overloading Overloading An identifier is overloaded of it denotes two or more distinct procedures in the same scope In older programming languages (e.g. C) identifiers and operators for certain built-in in functions are overloaded C - operator: integer negation (Integer Integer) floating-point negation (Float Float) integer subtraction (Integer Integer t I t Integer) ) floating-point subtraction (Float Float Float) C++, Java allow programmers to define overloaded identifiers and operators Identifier F denotes function (f 1 ) of type S 1 T 1 function (f 2 ) of type S 2 T 2 context-independent overloading requires S 1 and S 2 are non- equivalent if actual parameter E of F(E) is of type S 1 then F denotes f 1 if E is of type S 2 then F denotes f 2 with context-independent overloading the function can be uniquely identified by the type of the actual parameter / Faculteit Wiskunde en Informatica PAGE 30 / Faculteit Wiskunde en Informatica PAGE 31

9 Overloading Type conversions Identifier F denotes function (f 1 )of type S 1 T 1 function (f 2 ) of type S 2 T 2 context-dependent overloading requires S 1 and S 2 are non- equivalent or T 1 and T 2 are non-equivalent if S 1 and S 2 are non-equivalent, see previous slide if S 1 and S 2 are equivalent, but T 1 and T 2 are non-equivalent the context must be taken into consideration if the context F(E) is of type T 1 then F denotes f 1 if the context is of type T 2 then F denotes f 2 with context-dependent overloading, it is possible to formulate expressions which cannot be uniquely identified the programming language should prohibit this A type conversion is a mapping from the values of one type to corresponding values of a different type integers to real numbers real numbers to integers involve rounding and truncation A cast is an explicit type conversion In C, C++, Java: (T)E A coercion is an implicit type conversion C allows very restricted coercions: from integers to real number, from narrow-range to wide-range integers modern programming languages minimize or eliminate coercions altogether because of conflicts with polymorphism and overloading / Faculteit Wiskunde en Informatica PAGE 32 / Faculteit Wiskunde en Informatica PAGE 33 Ch 9: Control flow Sequencers We will study a number of alternatives traditional sequencers: sequential conditional iterative jumps, low-level e sequencers s to transfer control o escapes, sequencers to transfer control out of commands and procedures exceptions, sequencers to signal abnormal situations ti A sequencer is a language g construct to transfer control to some other point in a program, destination A sequencer implements a flow of control / Faculteit Wiskunde en Informatica PAGE 34 / Faculteit Wiskunde en Informatica PAGE 35

10 Jumps Jumps A jump transfers control to a specified program point A jump has typical the form goto L; and transfer the control to program point L, which is a label if (E 1 ) C 1 else { C 2 goto X; C 3 ; while (E 2 ) { C 4 ; X: C 5 Unrestricted jumps lead to spaghetti code, thus hard to understand, see edu/users/ewd/ewd02xx/ewd215 Most programming languages support gotos Java refuses jumps, because it supports higher-level escapes and exceptions Some languages have restrictions: the jump goto L; is legal only within the scope of L in C the scope of each label is the smallest enclosing block: jumps within a block jumps from a block to an enclosing one jumps into a block from outside is not allowed / Faculteit Wiskunde en Informatica PAGE 36 / Faculteit Wiskunde en Informatica PAGE 37 Jumps Escapes C restricts jumps because a jump out of a block must destroy the local variables Jumps out of a procedure should lead to destroying local variables and termination of procedure activation Jumps out of a recursive procedure is even more complicated Jumps introduce unwanted complexity in the semantics of high-level programming languages An escape is a sequencer that terminates the execution of a textually t enclosing command or procedure re In C, C++, Java: break sequencer to break loops In C, C++, Java: return sequencer to break loops and end procedures multiple return sequencers in a procedure is hard to maintain Escapes are restricted so that control cannot be transfer out of procedures A halt sequencer stops the entire program exit() in C / Faculteit Wiskunde en Informatica PAGE 38 / Faculteit Wiskunde en Informatica PAGE 39

11 Exceptions Exceptions Exceptions are a mechanism to deal with abnormal situations: arithmetic operation overflows uncompleted input/output operations Exceptions take of two things: error handling Controlled termination of flow of control Code that detects an abnormal situation throws an exception the exception can be caught in another part of the program programmer have control over exceptions and handling of it C++ and Java, being object-oriented languages, treat exceptions as objects / Faculteit Wiskunde en Informatica PAGE 40 / Faculteit Wiskunde en Informatica PAGE 41 Exceptions Exceptions Java has a built-in Exception class every exception is a subclass of Exception every exception represents a different abnormal situation try C 0 catch (T 1 I 1 ) C 1 catch (T n I n ) C n finally C f Able to catch any exception from class T 1 or or T n if C 0 throws an exception of type T i handler C i will be executed with identifier I i bound to that exception Exceptions can be caught at various levels: static float readfloat(bufferedreader input) throws IOException, NumberFormatException { if () throw new IOException( end of input ); String literal = ; float f = Float.parseFloat(literal); return f; Float.parseFloat throws the NumberFormatException / Faculteit Wiskunde en Informatica PAGE 42 / Faculteit Wiskunde en Informatica PAGE 43

12 Exceptions Exceptions Method annual rainfall data may catch the NumberFormatException static float[] readannual(bufferedreader input) throws IOException { try { float r = readfloat(input); catch (NumberFormatException e) { The main program calls readannual and deals with the IOException static void main() { float[] rainfall; try { rainfall = readannual(); catch (IOException e) { / Faculteit Wiskunde en Informatica PAGE 44 / Faculteit Wiskunde en Informatica PAGE 45 Ch 10: Concurrency Programs and processes Why concurrency? program design becomes more complex testing becomes less effective historically, i improvement of efficiency i via multiprogramming systems, usage of idle resources currently, end of Moore s law, efficiency e cy gain via multi-core processors A sequential process is a totally ordered set of steps each step is a change of state A sequential program specifies the state changes of a sequential process A concurrent program specifies the possible state changes of 2 or more sequential processes / Faculteit Wiskunde en Informatica PAGE 46 / Faculteit Wiskunde en Informatica PAGE 47

13 Problems with concurrency Problems with concurrency Nondeterminism: sequential programs are deterministic collateral and nondeterministic conditional commands may introduce unpredictability in sequential programs compiler is free to determine the order of execution different compilers may lead to different behavior A concurrent program is genuinely nondeterministic even for a specific compiler incorrect concurrent programs may behave correctly in general, but have sometimes unpredictable behavior Speed dependence sequential programs are speed-independent because correctness does not depend on execution speed concurrent programs are speed-dependentdependent behavior depends on the relative speed at which its constituent sequential processes run if absolute speeds are considered, outside world, we have real-time behavior if the outcome is speed-dependent, dependent, there is a race condition / Faculteit Wiskunde en Informatica PAGE 48 / Faculteit Wiskunde en Informatica PAGE 49 Problems with concurrency Problems with concurrency Deadlock means that processes are unable to make progress because of mutually incompatible demands for resources mutual exclusion: a process may be given exclusive access to resources incremental acquisition: a process holds previously acquired resource while waiting for new resources no preemption: resources cannot be removed from a process until it voluntarily releases them circular waiting: a cycle of resources or processes in which each process is waiting for resources that are held by the next process in the cycle Solutions for deadlocks: Ignore them and restart system if it occurs Recovery by detecting and kill involved processes Prevention by remove some of the preconditions Avoid by scheduling of resources / Faculteit Wiskunde en Informatica PAGE 50 / Faculteit Wiskunde en Informatica PAGE 51

14 Problems with concurrency Process interactions Starvation A concurrent program has the liveness property if it is guaranteed that every process will make some progress over a sufficiently long period of time Free of deadlock Fair scheduling Fair scheduling means that no process needing a resource is indefinitely prevented from obtaining it Sequential and collateral commands do not allow simultaneously execution of commands The parallel command B C indicates that B and C may executed simultaneously or arbitrarily interleaved / Faculteit Wiskunde en Informatica PAGE 52 / Faculteit Wiskunde en Informatica PAGE 53 Process interactions Process interactions Independent processes: commands B and C are independent if the execution of B has no effect on the execution of C, and vice versa concurrent composition of independent processes is deterministic Competing processes Commands B and C are competing if they need exclusive access to the same resource r Let B be the sequence B 1 ; B 2 ; B 3 Let C be the sequence C 1 ; C 2 ; C 3 B 1, C 1, B 3, C 3 are independent, none need r B 2 and C 2 both need r, so they cannot take place simultaneously, they are critical section wrt r B C may be executed as: ; B 2 ; ; C 2 ; ; C 2 ; ; B 2 ; but not as ; B 2 C 2 ; / Faculteit Wiskunde en Informatica PAGE 54 / Faculteit Wiskunde en Informatica PAGE 55

15 Process interactions Concurrency primitives Communicating processes let B be the sequence B 1 ; B 2 ; B 3 let C be the sequence C 1 ; C 2 ; C 3 there is communication from B to C if B 2 produces data that C 2 consumes, so B 2 must end before C 2 starts Thus B C has the same behavior as B; C Processes B and C intercommunicate if there is communication in both directions: B C yields numerous outcomes Process until now was a flow of control through a program a conventional, heavyweight, ht process is a program, which h involves: an address space allocation of main storage, share of CPU time access to files, devices, etc. context switching from one process to another involves a lot of time a thread is a lightweight alternative flow of control through a program without computational resources switching of threads involves swapping of content of working registers / Faculteit Wiskunde en Informatica PAGE 56 / Faculteit Wiskunde en Informatica PAGE 57 Concurrency primitives Concurrency primitives Process creation and control create a new child process load program code to be executed by a process start execution of a process suspend execution of a process resume execution of a (suspended) process let a process stop at the end of its execution let the creator wait for the process to stop destroy a stopped process, freeing resources create, load, start are combined into fork wait and destroy into join Process creation and control abstract operations for competition in order to make the critical sections disjoint in time acquire(r) to gain exclusive access to resource r relinquish(r) to give up exclusive access if resource r is already allocated, acquire(r) blocks the process it calls it if resource r is freed, processes waiting for access are unblocked and rescheduled abstract operations for (synchronous) communication transmit(m) called by a sender to send m receive(m) called by a receiver to wait for m asynchronous communication via broadcasting / Faculteit Wiskunde en Informatica PAGE 58 / Faculteit Wiskunde en Informatica PAGE 59

16 Concurrency primitives Concurrency primitives Interrupts ending a concurrent input/output operation is a infrequent operation to which the CPU should respond quickly ending input/output operations causes interrupts Spin locks and wait-free algorithms On a multiprocessor several processes may be executed simultaneously Most algorithms, Dekker s algorithm (presented by Dijkstra), Petterson s s algorithm, Simpson s s algorithm are complex and their behavior maybe corrupted by compiler optimizations Primitives should be built into the language events semaphores messages remote procedure calls / Faculteit Wiskunde en Informatica PAGE 60 / Faculteit Wiskunde en Informatica PAGE 61 Languages with concurrency primitives Next on GLT Object-oriented languages, g e.g. Java with threads Specification languages mcrl2 CHI POOSL Scripting languages ToolBus Meta-modeling: meta-model design xtext model transformations ti xpand GLT technology: Scanners Parsers Parser generators / Faculteit Wiskunde en Informatica PAGE 62 / Faculteit Wiskunde en Informatica PAGE 63