source: doc/rob_thesis/ctordtor.tex @ cb4d825

Last change on this file since cb4d825 was 9c14ae9, checked in by Rob Schluntz <rschlunt@…>, 7 years ago

add thesis source

  • Property mode set to 100644
File size: 77.2 KB
2\chapter{Constructors and Destructors}
5% TODO: discuss move semantics; they haven't been implemented, but could be. Currently looking at alternative models. (future work)
7% TODO: as an experiment, implement Andrei Alexandrescu's ScopeGuard
8% doesn't seem possible to do this without allowing ttype on generic structs?
10% If a Cforall constructor is in scope, C style initialization is
11% disabled by default.
12% * initialization rule: if any constructor is in scope for type T, try
13%   to find a matching constructor for the call. If there are no
14%   constructors in scope for type T, then attempt to fall back on
15%   C-style initialization.
16% + if this rule was not in place, it would be easy to accidentally
17%   use C-style initialization in certain cases, which could lead to
18%   subtle errors [2]
19% - this means we need special syntax if we want to allow users to force
20%   a C-style initialization (to give users more control)
21% - two different declarations in the same scope can be implicitly
22%   initialized differently. That is, there may be two objects of type
23%   T that are initialized differently because there is a constructor
24%   definition between them. This is not technically specific to
25%   constructors.
27% C-style initializers can be accessed with @= syntax
28% + provides a way to get around the requirement of using a constructor
29%   (for advanced programmers only)
30% - can break invariants in the type => unsafe
31% * provides a way of asserting that a variable is an instance of a
32%   C struct (i.e. a POD struct), and so will not be implicitly
33%   destructed (this can be useful at times, maybe mitigates the need
34%   for move semantics?) [3]
35% + can modernize a code base one step at a time
37% Cforall constructors can be used in expressions to initialize any
38% piece of memory.
39% + malloc() { ... } calls the appropriate constructor on the newly
40%   allocated space; the argument is moved into the constructor call
41%   without taking its address [4]
42% - with the above form, there is still no way to ensure that
43%   dynamically allocated objects are constructed. To resolve this,
44%   we might want a stronger "new" function which always calls the
45%   constructor, although how we accomplish that is currently still
46%   unresolved (compiler magic vs. better variadic functions?)
47% + This can be used as a placement syntax [5]
48% - can call the constructor on an object more than once, which could
49%   cause resource leaks and reinitialize const fields (can try to
50%   detect and prevent this in some cases)
51%   * compiler always tries to implicitly insert a ctor/dtor pair for
52%     non-@= objects.
53%     * For POD objects, this will resolve to an autogenerated or
54%       intrinsic function.
55%     * Intrinsic functions are not automatically called. Autogenerated
56%       are, because they may call a non-autogenerated function.
57%     * destructors are automatically inserted at appropriate branches
58%       (e.g. return, break, continue, goto) and at the end of the block
59%       in which they are declared.
60%   * For @= objects, the compiler never tries to interfere and insert
61%     constructor and destructor calls for that object. Copy constructor
62%     calls do not count, because the object is not the target of the copy
63%     constructor.
65% A constructor is declared with the name ?{}
66% + combines the look of C initializers with the precedent of ?() being
67%   the name for the function call operator
68% + it is possible to easily search for all constructors in a project
69%   and immediately know that a function is a constructor by seeing the
70%   name "?{}"
72% A destructor is declared with the name ^?{}
73% + name mirrors a constructor's name, with an extra symbol to
74%   distinguish it
75% - the symbol '~' cannot be used due to parsing conflicts with the
76%   unary '~' (bitwise negation) operator - this conflict exists because
77%   we want to allow users to write ^x{}; to destruct x, rather than
78%   ^?{}(&x);
80% The first argument of a constructor must be a pointer. The constructed
81% type is the base type of the pointer. E.g. void ?{}(T *) is a default
82% constructor for a T.
83% + can name the argument whatever you like, so not constrained by
84%   language keyword "this" or "self", etc.
85% - have to explicitly qualify all object members to initialize them
86%   (e.g. this->x = 0, rather than just x = 0)
88% Destructors can take arguments other than just the destructed pointer
89% * open research problem: not sure how useful this is
91% Pointer constructors
92% + can construct separately compiled objects (opaque types) [6]
93% + orthogonal design, follows directly from the definition of the first
94%   argument of a constructor
95% - may require copy constructor or move constructor (or equivalent)
96%   for correct implementation, which may not be obvious to everyone
97% + required feature for the prelude to specify as much behavior as possible
98%   (similar to pointer assignment operators in this respect)
100% Designations can only be used for C-style initialization
101% * designation for constructors is equivalent to designation for any
102%   general function call. Since a function prototype can be redeclared
103%   many times, with arguments named differently each time (or not at
104%   all!), this is considered to be an undesirable feature. We could
105%   construct some set of rules to allow this behaviour, but it is
106%   probably more trouble than it's worth, and no matter what we choose,
107%   it is not likely to be obvious to most people.
109% Constructing an anonymous member [7]
110% + same as with calling any other function on an anonymous member
111%   (implicit conversion by the compiler)
112% - may be some cases where this is ambiguous => clarify with a cast
113%   (try to design APIs to avoid sharing function signatures between
114%   composed types to avoid this)
116% Default Constructors and Destructors are called implicitly
117% + cannot forget to construct or destruct an object
118% - requires special syntax to specify that an object is not to be
119%   constructed (@=)
120% * an object will not be implicitly constructed OR destructed if
121%   explicitly initialized like a C object (@= syntax)
122% * an object will be destructed if there are no constructors in scope
123%   (even though it is initialized like a C object) [8]
125% An object which changes from POD type to non POD type will not change
126% the semantics of a type containing it by composition
127% * That is, constructors will not be regenerated at the point where
128%   an object changes from POD type to non POD type, because this could
129%   cause a cascade of constructors being regenerated for many other
130%   types. Further, there is precedence for this behaviour in other
131%   facets of Cforall's design, such as how nested functions interact.
132% * This behaviour can be simplified in a language without declaration
133%   before use, because a type can be classified as POD or non POD
134%   (rather than potentially changing between the two at some point) at
135%   at the global scope (which is likely the most common case)
136% * [9]
138% Move semantics
139% * <ongoing discussion about this. this will be filled in
140%    once we come to a consensus>
142% Changes to polymorphic type classes
143% * dtype and ftype remain the same
144% * forall(otype T) is currently essentially the same as
145%   forall(dtype T | { @size(T); void ?=?(T *, T); }).
146%   The big addition is that you can declare an object of type T, rather
147%   than just a pointer to an object of type T since you know the size,
148%   and you can assign into a T.
149%   * this definition is changed to add default constructor and
150%     destructor declarations, to remain consistent with what type meant
151%     before the introduction of constructors and destructors.
152%     * that is, forall(type T) is now essentially the same as
153%       forall(dtype T | { @size(T); void ?=?(T *, T);
154%                          void ?{}(T *); void ^?{}(T *); })
155%     + this is required to make generic types work correctly in
156%       polymorphic functions
157%     ? since declaring a constructor invalidates the autogenerated
158%       routines, it is possible for a type to have constructors, but
159%       not default constructors. That is, it might be the case that
160%       you want to write a polymorphic function for a type which has
161%       a size, but non-default constructors? Some options:
162%       * declaring a constructor as a part of the assertions list for
163%         a type declaration invalidates the default, so
164%         forall(otype T | { void ?{}(T *, int); })
165%         really means
166%         forall(dtype T | { @size(T); void ?=?(T *, T);
167%                            void ?{}(T *, int); void ^?{}(T *); })
168%       * force users to fully declare the assertions list like the
169%         above in this case (this seems very undesirable)
170%       * add another type class with the current desugaring of type
171%         (just size and assignment)
172%       * provide some way of subtracting from an existing assertions
173%         list (this might be useful to have in general)
175% Implementation issues:
176% Changes to prelude/autogen or built in defaults?
177% * pointer ctors/dtors [prelude]
178%   * other pointer type routines are declared in the prelude, and this
179%     doesn't seem like it should be any different
180% * basic type ctors/dtors [prelude]
181%   * other basic type routines are declared in the prelude, and this
182%     doesn't seem like it should be any different
183% ? aggregate types [undecided, but leaning towards autogenerate]
184%   * prelude
185%     * routines specific to aggregate types cannot be predeclared in
186%       the prelude because we don't know the name of every
187%       aggregate type in the entire program
188%   * autogenerate
189%     + default assignment operator is already autogenerated for
190%       aggregate types
191%       * this seems to lead us in the direction of autogenerating,
192%         because we may have a struct which contains other objects
193%         that require construction [10]. If we choose not to
194%         autogenerate in this case, then objects which are part of
195%         other objects by composition will not be constructed unless
196%         a constructor for the outer type is explicitly defined
197%       * in this case, we would always autogenerate the appropriate
198%         constructor(s) for an aggregate type, but just like with
199%         basic types, pointer types, and enum types, the constructor
200%         call can be elided when when it is not necessary.
201%     + constructors will have to be explicitly autogenerated
202%       in the case where they are required for a polymorphic function,
203%       when no user defined constructor is in scope, which may make it
204%       easiest to always autogenerate all appropriate constructors
205%     - n+2 constructors would have to be generated for a POD type
206%       * one constructor for each number of valid arguments [0, n],
207%         plus the copy constructor
208%         * this is taking a simplified approach: in C, it is possible
209%           to omit the enclosing braces in a declaration, which would
210%           lead to a combinatorial explosion of generated constructors.
211%           In the interest of keeping things tractable, Cforall may be
212%           incompatible with C in this case. [11]
213%       * for non-POD types, only autogenerate the default and copy
214%         constructors
215%       * alternative: generate only the default constructor and
216%         special case initialization for any other constructor when
217%         only the autogenerated one exists
218%         - this is not very sensible, as by the previous point, these
219%           constructors may be needed for polymorphic functions
220%           anyway.
221%     - must somehow distinguish in resolver between autogenerated and
222%       user defined constructors (autogenerated should never be chosen
223%       when a user defined option exists [check first parameter], even
224%       if full signature differs) (this may also have applications
225%       to other autogenerated routines?)
226%     - this scheme does not naturally support designation (i.e. general
227%       functions calls do not support designation), thus these cases
228%       will have to be treated specially in either case
229%   * defaults
230%     * i.e. hardcode a new set of rules for some "appropriate" default
231%       behaviour for
232%     + when resolving an initialization expression, explicitly check to
233%       see if any constructors are in scope. If yes, attempt to resolve
234%       to a constructor, and produce an error message if a match is not
235%       found. If there are no constructors in scope, resolve to
236%       initializing each field individually (C-style)
237%     + does not attempt to autogenerate constructors for POD types,
238%       which can be seen as a space optimization for the program
239%       binary
240%     - as stated previously, a polymorphic routine may require these
241%       autogenerated constructors, so this doesn't seem like a big win,
242%       because this leads to more complicated logic and tracking of
243%       which constructors have already been generated
244%     - even though a constructor is not explicitly declared or used
245%       polymorphically, we might still need one for all uses of a
246%       struct (e.g. in the case of composition).
247%   * the biggest tradeoff in autogenerating vs. defaulting appears to
248%     be in where and how the special code to check if constructors are
249%     present is handled. It appears that there are more reasons to
250%     autogenerate than not.
252% --- examples
253% [1] As an example of using constructors polymorphically, consider a
254% slight modification on the foldl example I put on the mailing list a
255% few months ago:
257% context iterable(type collection, type element, type iterator) {
258%   void ?{}(iterator *, collection); // used to be makeIterator, but can
259%                             // idiomatically use constructor
260%   int hasNext(iterator);
261%   iterator ++?(iterator *);
262%   lvalue element *?(iterator);
263% };
266% forall(type collection, type element, type result, type iterator
267%   | iterable(collection, element, iterator))
268% result foldl(collection c, result acc,
269%     result (*reduce)(result, element)) {
270%   iterator it = { c };
271%   while (hasNext(it)) {
272%     acc = reduce(acc, *it);
273%     ++it;
274%   }
275%   return acc;
276% }
278% Now foldl makes use of the knowledge that the iterator type has a
279% single argument constructor which takes the collection to iterate
280% over. This pattern allows polymorphic code to look more natural
281% (constructors are generally preferred to named initializer/creation
282% routines, e.g. "makeIterator")
284% [2] An example of some potentially dangerous code that we don't want
285% to let easily slip through the cracks - if this is really what you
286% want, then use @= syntax for the second declaration to quiet the
287% compiler.
289% struct A { int x, y, z; }
290% ?{}(A *, int);
291% ?{}(A *, int, int, int);
293% A a1 = { 1 };         // uses ?{}(A *, int);
294% A a2 = { 2, 3 };      // C-style initialization -> no invariants!
295% A a3 = { 4, 5, 6 };   // uses ?{}(A *, int, int, int);
297% [3] Since @= syntax creates a C object (essentially a POD, as far as
298% the compiler is concerned), the object will not be destructed
299% implicitly when it leaves scope, nor will it be copy constructed when
300% it is returned. In this case, a memcpy should be equivalent to a move.
302% // Box.h
303% struct Box;
304% void ?{}(Box **, int};
305% void ^?{}(Box **);
306% Box * make_fortytwo();
308% //
309% Box * make_fortytwo() {
310%   Box *b @= {};
311%   (&b){ 42 }; // construct explicitly
312%   return b; // no destruction, essentially a move?
313% }
315% [4] Cforall's typesafe malloc can be composed with constructor
316% expressions. It is possible for a user to define their own functions
317% similar to malloc and achieve the same effects (e.g. Aaron's example
318% of an arena allocator)
320% // CFA malloc
321% forall(type T)
322% T * malloc() { return (T *)malloc(sizeof(T)); }
324% struct A { int x, y, z; };
325% void ?{}(A *, int);
327% int foo(){
328%   ...
329%   // desugars to:
330%   // A * a = ?{}(malloc(), 123);
331%   A * a = malloc() { 123 };
332%   ...
333% }
335% [5] Aaron's example of combining function calls with constructor
336% syntax to perform an operation similar to C++'s std::vector::emplace
337% (i.e. to construct a new element in place, without the need to
338% copy)
340% forall(type T)
341% struct vector {
342%   T * elem;
343%   int len;
344%   ...
345% };
347% ...
348% forall(type T)
349% T * vector_new(vector(T) * v) {
350%   // reallocate if needed
351%   return &v->elem[len++];
352% }
354% int main() {
355%   vector(int) * v = ...
356%   vector_new(v){ 42 };  // add element to the end of vector
357% }
359% [6] Pointer Constructors. It could be useful to use the existing
360% constructor syntax even more uniformly for ADTs. With this, ADTs can
361% be initialized in the same manor as any other object in a polymorphic
362% function.
364% // vector.h
365% forall(type T) struct vector;
366% forall(type T) void ?{}(vector(T) **);
367% // adds an element to the end
368% forall(type T) vector(T) * ?+?(vector(T) *, T);
370% //
371% // don't want to expose the implementation to the user and/or don't
372% // want to recompile the entire program if the struct definition
373% // changes
375% forall(type T) struct vector {
376%   T * elem;
377%   int len;
378%   int capacity;
379% };
381% forall(type T) void resize(vector(T) ** v) { ... }
383% forall(type T) void ?{}(vector(T) ** v) {
384%   vector(T) * vect = *v = malloc();
385%   vect->capacity = 10;
386%   vect->len = 0;
387%   vect->elem = malloc(vect->capacity);
388% }
390% forall(type T) vector(T) * ?+?(vector(T) *v, T elem) {
391%   if (v->len == v->capacity) resize(&v);
392%   v->elem[v->len++] = elem;
393% }
395% //
396% #include "adt.h"
397% forall(type T | { T ?+?(T, int); }
398% T sumRange(int lower, int upper) {
399%   T x;    // default construct
400%   for (int i = lower; i <= upper; i++) {
401%     x = x + i;
402%   }
403%   return x;
404% }
406% int main() {
407%   vector(int) * numbers = sumRange(1, 10);
408%   // numbers is now a vector containing [1..10]
410%   int sum = sumRange(1, 10);
411%   // sum is now an int containing the value 55
412% }
414% [7] The current proposal is to use the plan 9 model of inheritance.
415% Under this model, all of the members of an unnamed struct instance
416% become members of the containing struct. In addition, an object
417% can be passed as an argument to a function expecting one of its
418% base structs.
420% struct Point {
421%   double x;
422%   double y;
423% };
425% struct ColoredPoint {
426%   Point;        // anonymous member (no identifier)
427%                 // => a ColoredPoint has an x and y of type double
428%   int color;
429% };
431% ColoredPoint cp = ...;
432% cp.x = 10.3;    // x from Point is accessed directly
433% cp.color = 0x33aaff; // color is accessed normally
434% foo(cp);        // cp can be used directly as a Point
436% void ?{}(Point *p, double x, double y) {
437%   p->x = x;
438%   p->y = y;
439% }
441% void ?{}(ColoredPoint *cp, double x, double y, int color) {
442%   (&cp){ x, y };  // unambiguous, no ?{}(ColoredPoint*,double,double)
443%   cp->color = color;
444% }
446% struct Size {
447%   double width;
448%   double height;
449% };
451% void ?{}(Size *s, double w, double h) {
452%   p->width = w;
453%   p->height = h;
454% }
456% struct Foo {
457%   Point;
458%   Size;
459% }
461% ?{}(Foo &f, double x, double y, double w, double h) {
462%   // (&F,x,y) is ambiguous => is it ?{}(Point*,double,double) or
463%   // ?{}(Size*,double,double)? Solve with a cast:
464%   ((Point*)&F){ x, y };
465%   ((Size*)&F){ w, h };
466% }
468% [8] Destructors will be called on objects that were not constructed.
470% struct A { ... };
471% ^?{}(A *);
472% {
473%   A x;
474%   A y @= {};
475% } // x is destructed, even though it wasn't constructed
476%   // y is not destructed, because it is explicitly a C object
479% [9] A type's constructor is generated at declaration time using
480% current information about an object's members. This is analogous to
481% the treatment of other operators. For example, an object's assignment
482% operator will not change to call the override of a member's assignment
483% operator unless the object's assignment is also explicitly overridden.
484% This problem can potentially be treated differently in Do, since each
485% compilation unit is passed over at least twice (once to gather
486% symbol information, once to generate code - this is necessary to
487% achieve the "No declarations" goal)
489% struct A { ... };
490% struct B { A x; };
491% ...
492% void ?{}(A *);  // from this point on, A objects will be constructed
493% B b1;           // b1 and b1.x are both NOT constructed, because B
494%                 // objects are not constructed
495% void ?{}(B *);  // from this point on, B objects will be constructed
496% B b2;           // b2 and b2.x are both constructed
498% struct C { A x; };
499% // implicit definition of ?{}(C*), because C is not a POD type since
500% // it contains a non-POD type by composition
501% C c;            // c and c.x are both constructed
503% [10] Requiring construction by composition
505% struct A {
506%   ...
507% };
509% // declared ctor disables default c-style initialization of
510% // A objects; A is no longer a POD type
511% void ?{}(A *);
513% struct B {
514%   A x;
515% };
517% // B objects can not be C-style initialized, because A objects
518% // must be constructed => B objects are transitively not POD types
519% B b; // b.x must be constructed, but B is not constructible
520%      // => must autogenerate ?{}(B *) after struct B definition,
521%      // which calls ?{}(&b.x)
523% [11] Explosion in the number of generated constructors, due to strange
524% C semantics.
526% struct A { int x, y; };
527% struct B { A u, v, w; };
529% A a = { 0, 0 };
531% // in C, you are allowed to do this
532% B b1 = { 1, 2, 3, 4, 5, 6 };
533% B b2 = { 1, 2, 3 };
534% B b3 = { a, a, a };
535% B b4 = { a, 5, 4, a };
536% B b5 = { 1, 2, a, 3 };
538% // we want to disallow b1, b2, b4, and b5 in Cforall.
539% // In particular, we will autogenerate these constructors:
540% void ?{}(A *);             // default/0 parameters
541% void ?{}(A *, int);        // 1 parameter
542% void ?{}(A *, int, int);   // 2 parameters
543% void ?{}(A *, const A *);  // copy constructor
545% void ?{}(B *);             // default/0 parameters
546% void ?{}(B *, A);          // 1 parameter
547% void ?{}(B *, A, A);       // 2 parameters
548% void ?{}(B *, A, A, A);    // 3 parameters
549% void ?{}(B *, const B *);  // copy constructor
551% // we will not generate constructors for every valid combination
552% // of members in C. For example, we will not generate
553% void ?{}(B *, int, int, int, int, int, int);   // b1 would need this
554% void ?{}(B *, int, int, int);                  // b2 would need this
555% void ?{}(B *, A, int, int, A);                 // b4 would need this
556% void ?{}(B *, int, int, A, int);               // b5 would need this
557% // and so on
561% TODO: talk somewhere about compound literals?
563Since \CFA is a true systems language, it does not provide a garbage collector.
564As well, \CFA is not an object-oriented programming language, i.e. structures cannot have routine members.
565Nevertheless, one important goal is to reduce programming complexity and increase safety.
566To that end, \CFA provides support for implicit pre/post-execution of routines for objects, via constructors and destructors.
568% TODO: this is old. remove or refactor
569% Manual resource management is difficult.
570% Part of the difficulty results from not having any guarantees about the current state of an object.
571% Objects can be internally composed of pointers that may reference resources which may or may not need to be manually released, and keeping track of that state for each object can be difficult for the end user.
573% Constructors and destructors provide a mechanism to bookend the lifetime of an object, allowing the designer of a type to establish invariants for objects of that type.
574% Constructors guarantee that object initialization code is run before the object can be used, while destructors provide a mechanism that is guaranteed to be run immediately before an object's lifetime ends.
575% Constructors and destructors can help to simplify resource management when used in a disciplined way.
576% In particular, when all resources are acquired in a constructor, and all resources are released in a destructor, no resource leaks are possible.
577% This pattern is a popular idiom in several languages, such as \CC, known as RAII (Resource Acquisition Is Initialization).
579This chapter details the design of constructors and destructors in \CFA, along with their current implementation in the translator.
580Generated code samples have been edited to provide comments for clarity and to save on space.
582\section{Design Criteria}
584In designing constructors and destructors for \CFA, the primary goals were ease of use and maintaining backwards compatibility.
586In C, when a variable is defined, its value is initially undefined unless it is explicitly initialized or allocated in the static area.
588int main() {
589  int x;        // uninitialized
590  int y = 5;    // initialized to 5
591  x = y;        // assigned 5
592  static int z; // initialized to 0
595In the example above, @x@ is defined and left uninitialized, while @y@ is defined and initialized to 5.
596Next, @x@ is assigned the value of @y@.
597In the last line, @z@ is implicitly initialized to 0 since it is marked @static@.
598The key difference between assignment and initialization being that assignment occurs on a live object (i.e. an object that contains data).
599It is important to note that this means @x@ could have been used uninitialized prior to being assigned, while @y@ could not be used uninitialized.
600Use of uninitialized variables yields undefined behaviour, which is a common source of errors in C programs. % TODO: *citation*
602Declaration initialization is insufficient, because it permits uninitialized variables to exist and because it does not allow for the insertion of arbitrary code before the variable is live.
603Many C compilers give good warnings most of the time, but they cannot in all cases.
605int f(int *);  // never reads the parameter, only writes
606int g(int *);  // reads the parameter - expects an initialized variable
608int x, y;
609f(&x);  // okay - only writes to x
610g(&y);  // will use y uninitialized
612Other languages are able to give errors in the case of uninitialized variable use, but due to backwards compatibility concerns, this cannot be the case in \CFA.
614In C, constructors and destructors are often mimicked by providing routines that create and teardown objects, where the teardown function is typically only necessary if the type modifies the execution environment.
616struct array_int {
617  int * x;
619struct array_int create_array(int sz) {
620  return (struct array_int) { malloc(sizeof(int)*sz) };
622void destroy_rh(struct resource_holder * rh) {
623  free(rh->x);
626This idiom does not provide any guarantees unless the structure is opaque, which then requires that all objects are heap allocated.
628struct opqaue_array_int;
629struct opqaue_array_int * create_opqaue_array(int sz);
630void destroy_opaque_array(opaque_array_int *);
631int opaque_get(opaque_array_int *);  // subscript
633opaque_array_int * x = create_opaque_array(10);
634int x2 = opaque_get(x, 2);
636This pattern is cumbersome to use since every access becomes a function call.
637While useful in some situations, this compromise is too restrictive.
638Furthermore, even with this idiom it is easy to make mistakes, such as forgetting to destroy an object or destroying it multiple times.
640A constructor provides a way of ensuring that the necessary aspects of object initialization is performed, from setting up invariants to providing compile-time checks for appropriate initialization parameters.
641This goal is achieved through a guarantee that a constructor is called implicitly after every object is allocated from a type with associated constructors, as part of an object's definition.
642Since a constructor is called on every object of a managed type, it is impossible to forget to initialize such objects, as long as all constructors perform some sensible form of initialization.
644In \CFA, a constructor is a function with the name @?{}@.
645Every constructor must have a return type of @void@ and at least one parameter, the first of which is colloquially referred to as the \emph{this} parameter, as in many object-oriented programming-languages (however, a programmer can give it an arbitrary name).
646The @this@ parameter must have a pointer type, whose base type is the type of object that the function constructs.
647There is precedence for enforcing the first parameter to be the @this@ parameter in other operators, such as the assignment operator, where in both cases, the left-hand side of the equals is the first parameter.
648There is currently a proposal to add reference types to \CFA.
649Once this proposal has been implemented, the @this@ parameter will become a reference type with the same restrictions.
651Consider the definition of a simple type encapsulating a dynamic array of @int@s.
654struct Array {
655  int * data;
656  int len;
660In C, if the user creates an @Array@ object, the fields @data@ and @len@ are uninitialized, unless an explicit initializer list is present.
661It is the user's responsibility to remember to initialize both of the fields to sensible values.
662In \CFA, the user can define a constructor to handle initialization of @Array@ objects.
665void ?{}(Array * arr){
666  arr->len = 10;    // default size
667  arr->data = malloc(sizeof(int)*arr->len);
668  for (int i = 0; i < arr->len; ++i) {
669    arr->data[i] = 0;
670  }
672Array x;  // allocates storage for Array and calls ?{}(&x)
675This constructor initializes @x@ so that its @length@ field has the value 10, and its @data@ field holds a pointer to a block of memory large enough to hold 10 @int@s, and sets the value of each element of the array to 0.
676This particular form of constructor is called the \emph{default constructor}, because it is called on an object defined without an initializer.
677In other words, a default constructor is a constructor that takes a single argument, the @this@ parameter.
679In \CFA, a destructor is a function much like a constructor, except that its name is \lstinline!^?{}!.
680A destructor for the @Array@ type can be defined as such.
682void ^?{}(Array * arr) {
683  free(arr->data);
686Since the destructor is automatically called at deallocation for all objects of type @Array@, the memory associated with an @Array@ is automatically freed when the object's lifetime ends.
687The exact guarantees made by \CFA with respect to the calling of destructors are discussed in section \ref{sub:implicit_dtor}.
689As discussed previously, the distinction between initialization and assignment is important.
690Consider the following example.
692Array x, y;
693Array z = x;  // initialization
694y = x;        // assignment
696By the previous definition of the default constructor for @Array@, @x@ and @y@ are initialized to valid arrays of length 10 after their respective definitions.
697On line 3, @z@ is initialized with the value of @x@, while on line @4@, @y@ is assigned the value of @x@.
698The key distinction between initialization and assignment is that a value to be initialized does not hold any meaningful values, whereas an object to be assigned might.
699In particular, these cases cannot be handled the same way because in the former case @z@ does not currently own an array, while @y@ does.
701\begin{cfacode}[emph={other}, emphstyle=\color{red}]
702void ?{}(Array * arr, Array other) {  // copy constructor
703  arr->len = other.len;               // initialization
704  arr->data = malloc(sizeof(int)*arr->len)
705  for (int i = 0; i < arr->len; ++i) {
706    arr->data[i] =[i];     // copy from other object
707  }
709Array ?=?(Array * arr, Array other) { // assignment
710  ^?{}(arr);                          // explicitly call destructor
711  ?{}(arr, other);                    // explicitly call constructor
712  return *arr;
715The two functions above handle these cases.
716The first function is called a \emph{copy constructor}, because it constructs its argument by copying the values from another object of the same type.
717The second function is the standard copy-assignment operator.
718These four functions are special in that they control the state of most objects.
720It is possible to define a constructor that takes any combination of parameters to provide additional initialization options.
721For example, a reasonable extension to the array type would be a constructor that allocates the array to a given initial capacity and initializes the array to a given @fill@ value.
723void ?{}(Array * arr, int capacity, int fill) {
724  arr->len = capacity;
725  arr->data = malloc(sizeof(int)*arr->len);
726  for (int i = 0; i < arr->len; ++i) {
727    arr->data[i] = fill;
728  }
731In \CFA, constructors are called implicitly in initialization contexts.
733Array x, y = { 20, 0xdeadbeef }, z = y;
735In \CFA, constructor calls look just like C initializers, which allows them to be inserted into legacy C code with minimal code changes, and also provides a very simple syntax that veteran C programmers are familiar with.
736One downside of reusing C initialization syntax is that it isn't possible to determine whether an object is constructed just by looking at its declaration, since that requires knowledge of whether the type is managed at that point.
738This example generates the following code
740Array x;
741?{}(&x);                  // implicit default construct
742Array y;
743?{}(&y, 20, 0xdeadbeef);  // explicit fill construct
744Array z;
745?{}(&z, y);               // copy construct
746^?{}(&z);                 // implicit destruct
747^?{}(&y);                 // implicit destruct
748^?{}(&x);                 // implicit destruct
750Due to the way that constructor calls are interleaved, it is impossible for @y@ to be referenced before it is initialized, except in its own constructor.
751This loophole is minor and exists in \CC as well.
752Destructors are implicitly called in reverse declaration-order so that objects with dependencies are destructed before the objects they are dependent on.
755\label{sub:syntax} % TODO: finish this section
756There are several ways to construct an object in \CFA.
757As previously introduced, every variable is automatically constructed at its definition, which is the most natural way to construct an object.
759struct A { ... };
760void ?{}(A *);
761void ?{}(A *, A);
762void ?{}(A *, int, int);
764A a1;             // default constructed
765A a2 = { 0, 0 };  // constructed with 2 ints
766A a3 = a1;        // copy constructed
767// implicitly destruct a3, a2, a1, in that order
769Since constructors and destructors are just functions, the second way is to call the function directly.
771struct A { int a; };
772void ?{}(A *);
773void ?{}(A *, A);
774void ^?{}(A *);
776A x;               // implicitly default constructed: ?{}(&x)
777A * y = malloc();  // copy construct: ?{}(&y, malloc())
779?{}(&x);    // explicit construct x
780?{}(y, x);  // explit construct y from x
781^?{}(&x);   // explicit destroy x
782^?{}(y);    // explicit destroy y
784// implicit ^?{}(&y);
785// implicit ^?{}(&x);
787Calling a constructor or destructor directly is a flexible feature that allows complete control over the management of a piece of storage.
788In particular, constructors double as a placement syntax.
790struct A { ... };
791struct memory_pool { ... };
792void ?{}(memory_pool *, size_t);
794memory_pool pool = { 1024 };  // create an arena of size 1024
796A * a = allocate(&pool);      // allocate from memory pool
797?{}(a);                       // construct an A in place
799for (int i = 0; i < 10; i++) {
800  // reuse storage rather than reallocating
801  ^?{}(a);
802  ?{}(a);
803  // use a ...
806deallocate(&pool, a);         // return to memory pool
808Finally, constructors and destructors support \emph{operator syntax}.
809Like other operators in \CFA, the function name mirrors the use-case, in that the first $N$ arguments fill in the place of the question mark.
811struct A { ... };
812struct B { A a; };
814A x, y, * z = &x;
815(&x){}          // default construct
816(&x){ y }       // copy construct
817(&x){ 1, 2, 3 } // construct with 3 arguments
818z{ y };         // copy construct x through a pointer
819^(&x){}         // destruct
821void ?{}(B * b) {
822  (&b->a){ 11, 17, 13 };  // construct a member
825Constructor operator syntax has relatively high precedence, requiring parentheses around an address-of expression.
826Destructor operator syntax is actually an statement, and requires parentheses for symmetry with constructor syntax.
828\subsection{Function Generation}
829In \CFA, every type is defined to have the core set of four functions described previously.
830Having these functions exist for every type greatly simplifies the semantics of the language, since most operations can simply be defined directly in terms of function calls.
831In addition to simplifying the definition of the language, it also simplifies the analysis that the translator must perform.
832If the translator can expect these functions to exist, then it can unconditionally attempt to resolve them.
833Moreover, the existence of a standard interface allows polymorphic code to interoperate with new types seamlessly.
835To mimic the behaviour of standard C, the default constructor and destructor for all of the basic types and for all pointer types are defined to do nothing, while the copy constructor and assignment operator perform a bitwise copy of the source parameter (as in \CC).
837There are several options for user-defined types: structures, unions, and enumerations.
838To aid in ease of use, the standard set of four functions is automatically generated for a user-defined type after its definition is completed.
839By auto-generating these functions, it is ensured that legacy C code will continue to work correctly in every context where \CFA expects these functions to exist, since they are generated for every complete type.
841The generated functions for enumerations are the simplest.
842Since enumerations in C are essentially just another integral type, the generated functions behave in the same way that the builtin functions for the basic types work.
843% TODO: examples for enums
844For example, given the enumeration
846enum Colour {
847  R, G, B
850The following functions are automatically generated.
852void ?{}(enum Colour *_dst){
853  // default constructor does nothing
855void ?{}(enum Colour *_dst, enum Colour _src){
856  (*_dst)=_src;  // bitwise copy
858void ^?{}(enum Colour *_dst){
859  // destructor does nothing
861enum Colour ?=?(enum Colour *_dst, enum Colour _src){
862  return (*_dst)=_src; // bitwise copy
865In the future, \CFA will introduce strongly-typed enumerations, like those in \CC.
866The existing generated routines will be sufficient to express this restriction, since they are currently set up to take in values of that enumeration type.
867Changes related to this feature only need to affect the expression resolution phase, where more strict rules will be applied to prevent implicit conversions from integral types to enumeration types, but should continue to permit conversions from enumeration types to @int@.
868In this way, it will still be possible to add an @int@ to an enumeration, but the resulting value will be an @int@, meaning that it won't be possible to reassign the value into an enumeration without a cast.
870For structures, the situation is more complicated.
871For a structure @S@ with members @M$_0$@, @M$_1$@, ... @M$_{N-1}$@, each function @f@ in the standard set calls \lstinline{f(s->M$_i$, ...)} for each @$i$@.
872That is, a default constructor for @S@ default constructs the members of @S@, the copy constructor with copy construct them, and so on.
873For example given the struct definition
875struct A {
876  B b;
877  C c;
880The following functions are implicitly generated.
882void ?{}(A * this) {
883  ?{}(&this->b);  // default construct each field
884  ?{}(&this->c);
886void ?{}(A * this, A other) {
887  ?{}(&this->b, other.b);  // copy construct each field
888  ?{}(&this->c, other.c);
890A ?=?(A * this, A other) {
891  ?=?(&this->b, other.b);  // assign each field
892  ?=?(&this->c, other.c);
894void ^?{}(A * this) {
895  ^?{}(&this->c);  // destruct each field
896  ^?{}(&this->b);
899It is important to note that the destructors are called in reverse declaration order to resolve conflicts in the event there are dependencies among members.
901In addition to the standard set, a set of \emph{field constructors} is also generated for structures.
902The field constructors are constructors that consume a prefix of the struct's member list.
903That is, $N$ constructors are built of the form @void ?{}(S *, T$_{\text{M}_0}$)@, @void ?{}(S *, T$_{\text{M}_0}$, T$_{\text{M}_1}$)@, ..., @void ?{}(S *, T$_{\text{M}_0}$, T$_{\text{M}_1}$, ..., T$_{\text{M}_{N-1}}$)@, where members are copy constructed if they have a corresponding positional argument and are default constructed otherwise.
904The addition of field constructors allows structs in \CFA to be used naturally in the same ways that they could be used in C (i.e. to initialize any prefix of the struct), e.g., @A a0 = { b }, a1 = { b, c }@.
905Extending the previous example, the following constructors are implicitly generated for @A@.
907void ?{}(A * this, B b) {
908  ?{}(&this->b, b);
909  ?{}(&this->c);
911void ?{}(A * this, B b, C c) {
912  ?{}(&this->b, b);
913  ?{}(&this->c, c);
917For unions, the default constructor and destructor do nothing, as it is not obvious which member if any should be constructed.
918For copy constructor and assignment operations, a bitwise @memcpy@ is applied.
919In standard C, a union can also be initialized using a value of the same type as its first member, and so a corresponding field constructor is generated to perform a bitwise @memcpy@ of the object.
920An alterantive to this design is to always construct and destruct the first member of a union, to match with the C semantics of initializing the first member of the union.
921This approach ultimately feels subtle and unsafe.
922Another option is to, like \CC, disallow unions from containing members that are themselves managed types.
923This restriction is a reasonable approach from a safety standpoint, but is not very C-like.
924Since the primary purpose of a union is to provide low-level memory optimization, it is assumed that the user has a certain level of maturity.
925It is therefore the responsibility of the user to define the special functions explicitly if they are appropriate, since it is impossible to accurately predict the ways that a union is intended to be used at compile-time.
927For example, given the union
929union X {
930  Y y;
931  Z z;
934The following functions are automatically generated.
936void ?{}(union X *_dst){  // default constructor
938void ?{}(union X *_dst, union X _src){  // copy constructor
939  __builtin_memcpy(_dst, &_src, sizeof(union X ));
941void ^?{}(union X *_dst){  // destructor
943union X ?=?(union X *_dst, union X _src){  // assignment
944  __builtin_memcpy(_dst, &_src, sizeof(union X));
945  return _src;
947void ?{}(union X *_dst, struct Y src){  // construct first field
948  __builtin_memcpy(_dst, &src, sizeof(struct Y));
952% This feature works in the \CFA model, since constructors are simply special functions and can be called explicitly, unlike in \CC. % this sentence isn't really true => placement new
953In \CCeleven, this restriction has been loosened to allow unions with managed members, with the caveat that any if there are any members with a user-defined operation, then that operation is not implicitly defined, forcing the user to define the operation if necessary.
954This restriction could easily be added into \CFA once \emph{deleted} functions are added.
956\subsection{Using Constructors and Destructors}
957Implicitly generated constructor and destructor calls ignore the outermost type qualifiers, e.g. @const@ and @volatile@, on a type by way of a cast on the first argument to the function.
958For example,
960struct S { int i; };
961void ?{}(S *, int);
962void ?{}(S *, S);
964const S s = { 11 };
965volatile S s2 = s;
967Generates the following code
969const struct S s;
970?{}((struct S *)&s, 11);
971volatile struct S s2;
972?{}((struct S *)&s2, s);
974Here, @&s@ and @&s2@ are cast to unqualified pointer types.
975This mechanism allows the same constructors and destructors to be used for qualified objects as for unqualified objects.
976Since this applies only to implicitly generated constructor calls, the language does not allow qualified objects to be re-initialized with a constructor without an explicit cast.
978Unlike \CC, \CFA provides an escape hatch that allows a user to decide at an object's definition whether it should be managed or not.
979An object initialized with \ateq is guaranteed to be initialized like a C object, and has no implicit destructor call.
980This feature provides all of the freedom that C programmers are used to having to optimize a program, while maintaining safety as a sensible default.
982struct A { int * x; };
983// RAII
984void ?{}(A * a) { a->x = malloc(sizeof(int)); }
985void ^?{}(A * a) { free(a->x); }
987A a1;           // managed
988A a2 @= { 0 };  // unmanaged
990In this example, @a1@ is a managed object, and thus is default constructed and destructed at the end of @a1@'s lifetime, while @a2@ is an unmanaged object and is not implicitly constructed or destructed.
991Instead, @a2->x@ is initialized to @0@ as if it were a C object, due to the explicit initializer.
992Existing constructors are ignored when \ateq is used, so that any valid C initializer is able to initialize the object.
994In addition to freedom, \ateq provides a simple path to migrating legacy C code to Cforall, in that objects can be moved from C-style initialization to \CFA gradually and individually.
995It is worth noting that the use of unmanaged objects can be tricky to get right, since there is no guarantee that the proper invariants are established on an unmanaged object.
996It is recommended that most objects be managed by sensible constructors and destructors, except where absolutely necessary.
998When the user declares any constructor or destructor, the corresponding intrinsic/generated function and all field constructors for that type are hidden, so that they will not be found during expression resolution unless the user-defined function goes out of scope.
999Furthermore, if the user declares any constructor, then the intrinsic/generated default constructor is also hidden, making it so that objects of a type may not be default constructable.
1000This closely mirrors the rule for implicit declaration of constructors in \CC, wherein the default constructor is implicitly declared if there is no user-declared constructor. % TODO: cite C++98 page 186??
1002struct S { int x, y; };
1004void f() {
1005  S s0, s1 = { 0 }, s2 = { 0, 2 }, s3 = s2;  // okay
1006  {
1007    void ?{}(S * s, int i) { s->x = i*2; }
1008    S s4;  // error
1009    S s5 = { 3 };  // okay
1010    S s6 = { 4, 5 };  // error
1011    S s7 = s5; // okay
1012  }
1013  S s8, s9 = { 6 }, s10 = { 7, 8 }, s11 = s10;  // okay
1016In this example, the inner scope declares a constructor from @int@ to @S@, which hides the default constructor and field constructors until the end of the scope.
1018When defining a constructor or destructor for a struct @S@, any members that are not explicitly constructed or destructed are implicitly constructed or destructed automatically.
1019If an explicit call is present, then that call is taken in preference to any implicitly generated call.
1020A consequence of this rule is that it is possible, unlike \CC, to precisely control the order of construction and destruction of subobjects on a per-constructor basis, whereas in \CC subobject initialization and destruction is always performed based on the declaration order.
1022struct A {
1023  B w, x, y, z;
1025void ?{}(A * a, int i) {
1026  (&a->x){ i };
1027  (&a->z){ a->y };
1030Generates the following
1032void ?{}(A * a, int i) {
1033  (&a->w){};   // implicit default ctor
1034  (&a->y){};   // implicit default ctor
1035  (&a->x){ i };
1036  (&a->z){ a->y };
1039Finally, it is illegal for a subobject to be explicitly constructed for the first time after it is used for the first time.
1040If the translator cannot be reasonably sure that an object is constructed prior to its first use, but is constructed afterward, an error is emitted.
1041More specifically, the translator searches the body of a constructor to ensure that every subobject is initialized.
1043void ?{}(A * a, double x) {
1044  f(a->x);
1045  (&a->x){ (int)x }; // error, used uninitialized on previous line
1048However, if the translator sees a subobject used within the body of a constructor, but does not see a constructor call that uses the subobject as the target of a constructor, then the translator assumes the object is to be implicitly constructed (copy constructed in a copy constructor and default constructed in any other constructor).
1050void ?{}(A * a) {
1051  // default constructs all members
1052  f(a->x);
1055void ?{}(A * a, A other) {
1056  // copy constructs all members
1057  f(a->y);
1060void ^?{}(A * a) {
1061  ^(&a->x){}; // explicit destructor call
1062} // z, y, w implicitly destructed, in this order
1064If at any point, the @this@ parameter is passed directly as the target of another constructor, then it is assumed that constructor handles the initialization of all of the object's members and no implicit constructor calls are added. % TODO: confirm that this is correct. It might be possible to get subtle errors if you initialize some members then call another constructor... -- in fact, this is basically always wrong. if anything, I should check that such a constructor does not initialize any members, otherwise it'll always initialize the member twice (once locally, once by the called constructor).
1065To override this rule, \ateq can be used to force the translator to trust the programmer's discretion.
1066This form of \ateq is not yet implemented.
1068Despite great effort, some forms of C syntax do not work well with constructors in \CFA.
1069In particular, constructor calls cannot contain designations (see \ref{sub:c_background}), since this is equivalent to allowing designations on the arguments to arbitrary function calls.
1070In C, function prototypes are permitted to have arbitrary parameter names, including no names at all, which may have no connection to the actual names used at function definition.
1071Furthermore, a function prototype can be repeated an arbitrary number of times, each time using different names.
1073// all legal forward declarations in C
1074void f(int, int, int);
1075void f(int a, int b, int c);
1076void f(int b, int c, int a);
1077void f(int c, int a, int b);
1078void f(int x, int y, int z);
1080f(b:10, a:20, c:30);  // which parameter is which?
1082As a result, it was decided that any attempt to resolve designated function calls with C's function prototype rules would be brittle, and thus it is not sensible to allow designations in constructor calls.
1083% Many other languages do allow named arguments, such as Python and Scala, but they do not allow multiple arbitrarily named forward declarations of a function.
1085In addition, constructor calls cannot have a nesting depth greater than the number of array components in the type of the initialized object, plus one.
1086For example,
1088struct A;
1089void ?{}(A *, int);
1090void ?{}(A *, A, A);
1092A a1[3] = { { 3 }, { 4 }, { 5 } };
1093A a2[2][2] = {
1094  { { 9 }, { 10 } },  // a2[0]
1095  { {14 }, { 15 } }   // a2[1]
1097A a3[4] = {
1098  { { 11 }, { 12 } },  // error
1099  { 80 }, { 90 }, { 100 }
1102% TODO: in CFA if the array dimension is empty, no object constructors are added -- need to fix this.
1103The body of @A@ has been omitted, since only the constructor interfaces are important.
1104In C, having a greater nesting depth means that the programmer intends to initialize subobjects with the nested initializer.
1105The reason for this omission is to both simplify the mental model for using constructors, and to make initialization simpler for the expression resolver.
1106If this were allowed, it would be necessary for the expression resolver to decide whether each argument to the constructor call could initialize to some argument in one of the available constructors, making the problem highly recursive and potentially much more expensive.
1107That is, in the previous example the line marked as an error could mean construct using @?{}(A *, A, A)@, since the inner initializer @{ 11 }@ could be taken as an intermediate object of type @A@ constructed with @?{}(A *, int)@.
1108In practice, however, there could be many objects that can be constructed from a given @int@ (or, indeed, any arbitrary parameter list), and thus a complete solution to this problem would require fully exploring all possibilities.
1109It should be noted that unmanaged objects can still make use of designations and nested initializers in \CFA.
1111\subsection{Implicit Destructors}
1113Destructors are automatically called at the end of the block in which the object is declared.
1114In addition to this, destructors are automatically called when statements manipulate control flow to leave a block in which the object is declared, e.g., with return, break, continue, and goto statements.
1115The example below demonstrates a simple routine with multiple return statements.
1117struct A;
1118void ^?{}(A *);
1120void f(int i) {
1121  A x;  // construct x
1122  {
1123    A y; // construct y
1124    {
1125      A z; // construct z
1126      {
1127        if (i == 0) return; // destruct x, y, z
1128      }
1129      if (i == 1) return; // destruct x, y, z
1130    } // destruct z
1131    if (i == 2) return; // destruct x, y
1132  } // destruct y
1136%% having this feels excessive, but it's here if necessary
1137% This procedure generates the following code.
1138% \begin{cfacode}
1139% void f(int i){
1140%   struct A x;
1141%   ?{}(&x);
1142%   {
1143%     struct A y;
1144%     ?{}(&y);
1145%     {
1146%       struct A z;
1147%       ?{}(&z);
1148%       {
1149%         if ((i==0)!=0) {
1150%           ^?{}(&z);
1151%           ^?{}(&y);
1152%           ^?{}(&x);
1153%           return;
1154%         }
1155%       }
1156%       if (((i==1)!=0) {
1157%           ^?{}(&z);
1158%           ^?{}(&y);
1159%           ^?{}(&x);
1160%           return ;
1161%       }
1162%       ^?{}(&z);
1163%     }
1165%     if ((i==2)!=0) {
1166%       ^?{}(&y);
1167%       ^?{}(&x);
1168%       return;
1169%     }
1170%     ^?{}(&y);
1171%   }
1173%   ^?{}(&x);
1174% }
1175% \end{cfacode}
1177The next example illustrates the use of simple continue and break statements and the manner that they interact with implicit destructors.
1179for (int i = 0; i < 10; i++) {
1180  A x;
1181  if (i == 2) {
1182    continue;  // destruct x
1183  } else if (i == 3) {
1184    break;     // destruct x
1185  }
1186} // destruct x
1188Since a destructor call is automatically inserted at the end of the block, nothing special needs to happen to destruct @x@ in the case where control reaches the end of the loop.
1189In the case where @i@ is @2@, the continue statement runs the loop update expression and attemps to begin the next iteration of the loop.
1190Since continue is a C statement, which does not understand destructors, a destructor call is added just before the continue statement to ensure that @x@ is destructed.
1191When @i@ is @3@, the break statement moves control to just past the end of the loop.
1192Like the previous case, a destructor call for @x@ is inserted just before the break statement.
1194\CFA also supports labelled break and continue statements, which allow more precise manipulation of control flow.
1195Labelled break and continue allow the programmer to specify which control structure to target by using a label attached to a control structure.
1196\begin{cfacode}[emph={L1,L2}, emphstyle=\color{red}]
1197L1: for (int i = 0; i < 10; i++) {
1198  A x;
1199  L2: for (int j = 0; j < 10; j++) {
1200    A y;
1201    if (j == 0) {
1202      continue;    // destruct y
1203    } else if (j == 1) {
1204      break;       // destruct y
1205    } else if (i == 1) {
1206      continue L1; // destruct y
1207    } else if (i == 2) {
1208      break L1;    // destruct x,y
1209    }
1210  } // destruct y
1211} // destruct X
1213The statement @continue L1@ begins the next iteration of the outer for-loop.
1214Since the semantics of continue require the loop update expression to execute, control branches to the \emph{end} of the outer for loop, meaning that the block destructor for @x@ can be reused, and it is only necessary to generate the destructor for @y@.
1215Break, on the other hand, requires jumping out of the loop, so the destructors for both @x@ and @y@ are generated and inserted before the @break L1@ statement.
1217Finally, an example which demonstrates goto.
1218Since goto is a general mechanism for jumping to different locations in the program, a more comprehensive approach is required.
1219For each goto statement $G$ and each target label $L$, let $S_G$ be the set of all managed variables alive at $G$, and let $S_L$ be the set of all managed variables alive at $L$.
1220If at any $G$, $S_L \setminus S_G = \emptyset$, then the translator emits an error, because control flow branches from a point where the object is not yet live to a point where it is live, skipping the object's constructor.
1221Then, for every $G$, the destructors for each variable in the set $S_G \setminus S_L$ is inserted directly before $G$, which ensures each object that is currently live at $G$, but not at $L$, is destructed before control branches.
1223int i = 0;
1225  L0: ;     // S_L0 = { x }
1226    A y;
1227  L1: ;     // S_L1 = { x }
1228    A x;
1229  L2: ;     // S_L2 = { y, x }
1230    if (i == 0) {
1231      ++i;
1232      goto L1;    // S_G = { y, x }
1233      // S_G-S_L1 = { x } => destruct x
1234    } else if (i == 1) {
1235      ++i;
1236      goto L2;    // S_G = { y, x }
1237      // S_G-S_L2 = {} => destruct nothing
1238    } else if (i == 2) {
1239      ++i;
1240      goto L3;    // S_G = { y, x }
1241      // S_G-S_L3 = {}
1242    } else if (false) {
1243      ++i;
1244      A z;
1245      goto L3;    // S_G = { z, y, x }
1246      // S_G-S_L3 = { z } => destruct z
1247    } else {
1248      ++i;
1249      goto L4;    // S_G = { y, x }
1250      // S_G-S_L4 = { y, x } => destruct y, x
1251    }
1252  L3: ;    // S_L3 = { y, x }
1253    goto L2;      // S_G = { y, x }
1254    // S_G-S_L2 = {}
1256L4: ;  // S_L4 = {}
1257if (i == 4) {
1258  goto L0;        // S_G = {}
1259  // S_G-S_L0 = {}
1262Labelled break and continue are implemented in \CFA in terms of goto statements, so the more constrained forms are precisely goverened by these rules.
1264The next example demonstrates the error case.
1267    goto L1; // S_G = {}
1268    // S_L1-S_G = { y } => error
1269    A y;
1270  L1: ; // S_L1 = { y }
1271    A x;
1272  L2: ; // S_L2 = { y, x }
1274goto L2; // S_G = {}
1275// S_L2-S_G = { y, x } => error
1278\subsection{Implicit Copy Construction}
1279When a function is called, the arguments supplied to the call are subject to implicit copy construction (and destruction of the generated temporary), and the return value is subject to destruction.
1280When a value is returned from a function, the copy constructor is called to pass the value back to the call site.
1281Exempt from these rules are intrinsic and builtin functions.
1282It should be noted that unmanaged objects are subject to copy constructor calls when passed as arguments to a function or when returned from a function, since they are not the \emph{target} of the copy constructor call.
1283This is an important detail to bear in mind when using unmanaged objects, and could produce unexpected results when mixed with objects that are explicitly constructed.
1285struct A;
1286void ?{}(A *);
1287void ?{}(A *, A);
1288void ^?{}(A *);
1290A f(A x) {
1291  return x;
1294A y, z @= {};
1298Note that @z@ is copy constructed into a temporary variable to be passed as an argument, which is also destructed after the call.
1299A special syntactic form, such as a variant of \ateq, could be implemented to specify at the call site that an argument should not be copy constructed, to regain some control for the C programmer.
1301This generates the following
1303struct A f(struct A x){
1304  struct A _retval_f;
1305  ?{}((&_retval_f), x);
1306  return _retval_f;
1309struct A y;
1311struct A z = { 0 };
1313struct A _tmp_cp1;     // argument 1
1314struct A _tmp_cp_ret0; // return value
1315_tmp_cp_ret0=f((?{}(&_tmp_cp1, y) , _tmp_cp1)), _tmp_cp_ret0;
1316^?{}(&_tmp_cp_ret0);   // return value
1317^?{}(&_tmp_cp1);       // argument 1
1319struct A _tmp_cp2;     // argument 1
1320struct A _tmp_cp_ret1; // return value
1321_tmp_cp_ret1=f((?{}(&_tmp_cp2, z), _tmp_cp2)), _tmp_cp_ret1;
1322^?{}(&_tmp_cp_ret1);   // return value
1323^?{}(&_tmp_cp2);       // argument 1
1327A known issue with this implementation is that the return value of a function is not guaranteed to have the same address for its entire lifetime.
1328Specifically, since @_retval_f@ is allocated and constructed in @f@ then returned by value, the internal data is bitwise copied into the caller's stack frame.
1329This approach works out most of the time, because typically destructors need to only access the fields of the object and recursively destroy.
1330It is currently the case that constructors and destructors which use the @this@ pointer as a unique identifier to store data externally will not work correctly for return value objects.
1331Thus is it not safe to rely on an object's @this@ pointer to remain constant throughout execution of the program.
1333A * external_data[32];
1334int ext_count;
1335struct A;
1336void ?{}(A * a) {
1337  // ...
1338  external_data[ext_count++] = a;
1340void ^?{}(A * a) {
1341  for (int i = 0; i < ext_count) {
1342    if (a == external_data[i]) { // may never be true
1343      // ...
1344    }
1345  }
1348In the above example, a global array of pointers is used to keep track of all of the allocated @A@ objects.
1349Due to copying on return, the current object being destructed will not exist in the array if an @A@ object is ever returned by value from a function.
1351This problem could be solved in the translator by mutating the function signatures so that the return value is moved into the parameter list.
1352For example, the translator could restructure the code like so
1354void f(struct A x, struct A * _retval_f){
1355  ?{}(_retval_f, x);  // construct directly into caller's stack frame
1358struct A y;
1360struct A z = { 0 };
1362struct A _tmp_cp1;     // argument 1
1363struct A _tmp_cp_ret0; // return value
1364f((?{}(&_tmp_cp1, y) , _tmp_cp1), &_tmp_cp_ret0), _tmp_cp_ret0;
1365^?{}(&_tmp_cp_ret0);   // return value
1366^?{}(&_tmp_cp1);       // argument 1
1368This transformation provides @f@ with the address of the return variable so that it can be constructed into directly.
1369It is worth pointing out that this kind of signature rewriting already occurs in polymorphic functions which return by value, as discussed in \cite{Bilson03}.
1370A key difference in this case is that every function would need to be rewritten like this, since types can switch between managed and unmanaged at different scope levels, e.g.
1372struct A { int v; };
1373A x; // unmanaged
1375  void ?{}(A * a) { ... }
1376  void ^?{}(A * a) { ... }
1377  A y; // managed
1379A z; // unmanaged
1381Hence there is not enough information to determine at function declaration to determine whether a type is managed or not, and thus it is the case that all signatures have to be rewritten to account for possible copy constructor and destructor calls.
1382Even with this change, it would still be possible to declare backwards compatible function prototypes with an @extern "C"@ block, which allows for the definition of C-compatible functions within \CFA code, however this would require actual changes to the way code inside of an @extern "C"@ function is generated as compared with normal code generation.
1383Furthermore, it isn't possible to overload C functions, so using @extern "C"@ to declare functions is of limited use.
1385It would be possible to regain some control by adding an attribute to structs which specifies whether they can be managed or not (perhaps \emph{manageable} or \emph{unmanageable}), and to emit an error in the case that a constructor or destructor is declared for an unmanageable type.
1386Ideally, structs should be manageable by default, since otherwise the default case becomes more verbose.
1387This means that in general, function signatures would have to be rewritten, and in a select few cases the signatures would not be rewritten.
1389__attribute__((manageable)) struct A { ... };   // can declare constructors
1390__attribute__((unmanageable)) struct B { ... }; // cannot declare constructors
1391struct C { ... };                               // can declare constructors
1393A f();  // rewritten void f(A *);
1394B g();  // not rewritten
1395C h();  // rewritten void h(C *);
1397An alternative is to instead make the attribute \emph{identifiable}, which states that objects of this type use the @this@ parameter as an identity.
1398This strikes more closely to the visibile problem, in that only types marked as identifiable would need to have the return value moved into the parameter list, and every other type could remain the same.
1399Furthermore, no restrictions would need to be placed on whether objects can be constructed.
1401__attribute__((identifiable)) struct A { ... };  // can declare constructors
1402struct B { ... };                                // can declare constructors
1404A f();  // rewritten void f(A *);
1405B g();  // not rewritten
1408Ultimately, this is the type of transformation that a real compiler would make when generating assembly code.
1409Since a compiler has full control over its calling conventions, it can seamlessly allow passing the return parameter without outwardly changing the signature of a routine.
1410As such, it has been decided that this issue is not currently a priority.
1413\subsection{Array Initialization}
1414Arrays are a special case in the C type system.
1415C arrays do not carry around their size, making it impossible to write a standalone \CFA function that constructs or destructs an array while maintaining the standard interface for constructors and destructors.
1416Instead, \CFA defines the initialization and destruction of an array recursively.
1417That is, when an array is defined, each of its elements is constructed in order from element 0 up to element $n-1$.
1418When an array is to be implicitly destructed, each of its elements is destructed in reverse order from element $n-1$ down to element 0.
1419As in C, it is possible to explicitly provide different initializers for each element of the array through array initialization syntax.
1420In this case, each of the initializers is taken in turn to construct a subsequent element of the array.
1421If too many initializers are provided, only the initializers up to N are actually used.
1422If too few initializers are provided, then the remaining elements are default constructed.
1424For example, given the following code.
1426struct X {
1427  int x, y, z;
1429void f() {
1430  X x[10] = { { 1, 2, 3 }, { 4 }, { 7, 8 } };
1433The following code is generated for @f@.
1435void f(){
1436  struct X x[((long unsigned int )10)];
1437  // construct x
1438  {
1439    int _index0 = 0;
1440    // construct with explicit initializers
1441    {
1442      if (_index0<10) ?{}(&x[_index0], 1, 2, 3);
1443      ++_index0;
1444      if (_index0<10) ?{}(&x[_index0], 4);
1445      ++_index0;
1446      if (_index0<10) ?{}(&x[_index0], 7, 8);
1447      ++_index0;
1448    }
1450    // default construct remaining elements
1451    for (;_index0<10;++_index0) {
1452      ?{}(&x[_index0]);
1453    }
1454  }
1455  // destruct x
1456  {
1457    int _index1 = 10-1;
1458    for (;_index1>=0;--_index1) {
1459      ^?{}(&x[_index1]);
1460    }
1461  }
1464Multidimensional arrays require more complexity.
1465For example, a two dimensional array
1467void g() {
1468  X x[10][10] = {
1469    { { 1, 2, 3 }, { 4 } }, // x[0]
1470    { { 7, 8 } }            // x[1]
1471  };
1473Generates the following
1475void g(){
1476  struct X x[10][10];
1477  // construct x
1478  {
1479    int _index0 = 0;
1480    for (;_index0<10;++_index0) {
1481      {
1482        int _index1 = 0;
1483        // construct with explicit initializers
1484        {
1485          switch ( _index0 ) {
1486            case 0:
1487              // construct first array
1488              if ( _index1<10 ) ?{}(&x[_index0][_index1], 1, 2, 3);
1489              ++_index1;
1490              if ( _index1<10 ) ?{}(&x[_index0][_index1], 4);
1491              ++_index1;
1492              break;
1493            case 1:
1494              // construct second array
1495              if ( _index1<10 ) ?{}(&x[_index0][_index1], 7, 8);
1496              ++_index1;
1497              break;
1498          }
1499        }
1500        // default construct remaining elements
1501        for (;_index1<10;++_index1) {
1502            ?{}(&x[_index0][_index1]);
1503        }
1504      }
1505    }
1506  }
1507  // destruct x
1508  {
1509    int _index2 = 10-1;
1510    for (;_index2>=0;--_index2) {
1511      {
1512        int _index3 = 10-1;
1513        for (;_index3>=0;--_index3) {
1514            ^?{}(&x[_index2][_index3]);
1515        }
1516      }
1517    }
1518  }
1521% It is possible to generate slightly simpler code for the switch cases, since the value of @_index1@ is known at compile-time within each case, however the procedure for generating constructor calls is complicated.
1522% It is simple to remove the increment statements for @_index1@, but it is not simple to remove the
1523%% technically, it's not hard either. I could easily downcast and change the second argument to ?[?], but is it really necessary/worth it??
1525\subsection{Global Initialization}
1526In standard C, global variables can only be initialized to compile-time constant expressions.
1527This places strict limitations on the programmer's ability to control the default values of objects.
1528In \CFA, constructors and destructors are guaranteed to be run on global objects, allowing arbitrary code to be run before and after the execution of the main routine.
1529By default, objects within a translation unit are constructed in declaration order, and destructed in the reverse order.
1530The default order of construction of objects amongst translation units is unspecified.
1531% TODO: not yet implemented, but g++ provides attribute init_priority, which allows specifying the order of global construction on a per object basis
1533% suggestion: implement this in CFA by picking objects with a specified priority and pulling them into their own init functions (could even group them by priority level -> map<int, list<ObjectDecl*>>) and pull init_priority forward into constructor and destructor attributes with the same priority level
1534It is, however, guaranteed that any global objects in the standard library are initialized prior to the initialization of any object in the user program.
1536This feature is implemented in the \CFA translator by grouping every global constructor call into a function with the GCC attribute \emph{constructor}, which performs most of the heavy lifting. % CITE:
1537A similar function is generated with the \emph{destructor} attribute, which handles all global destructor calls.
1538At the time of writing, initialization routines in the library are specified with priority \emph{101}, which is the highest priority level that GCC allows, whereas initialization routines in the user's code are implicitly given the default priority level, which ensures they have a lower priority than any code with a specified priority level.
1539This mechanism allows arbitrarily complicated initialization to occur before any user code runs, making it possible for library designers to initialize their modules without requiring the user to call specific startup or teardown routines.
1541For example, given the following global declarations.
1543struct X {
1544  int y, z;
1546void ?{}(X *);
1547void ?{}(X *, int, int);
1548void ^?{}(X *);
1550X a;
1551X b = { 10, 3 };
1553The following code is generated.
1555__attribute__ ((constructor)) static void _init_global_ctor(void){
1556  ?{}(&a);
1557  ?{}(&b, 10, 3);
1559__attribute__ ((destructor)) static void _destroy_global_ctor(void){
1560  ^?{}(&b);
1561  ^?{}(&a);
1565\subsection{Static Local Variables}
1566In standard C, it is possible to mark variables that are local to a function with the @static@ storage class.
1567Unlike normal local variables, a @static@ local variable is defined to live for the entire duration of the program, so that each call to the function has access to the same variable with the same address and value as it had in the previous call to the function. % TODO: mention dynamic loading caveat??
1568Much like global variables, in C @static@ variables must be initialized to a \emph{compile-time constant value} so that a compiler is able to create storage for the variable and initialize it before the program begins running.
1570Yet again, this rule is too restrictive for a language with constructors and destructors.
1571Instead, \CFA modifies the definition of a @static@ local variable so that objects are guaranteed to be live from the time control flow reaches their declaration, until the end of the program, since the initializer expression is not necessarily a compile-time constant, but can depend on the current execution state of the function.
1572Since standard C does not allow access to a @static@ local variable before the first time control flow reaches the declaration, this restriction does not preclude any valid C code.
1573Local objects with @static@ storage class are only implicitly constructed and destructed once for the duration of the program.
1574The object is constructed when its declaration is reached for the first time.
1575The object is destructed once at the end of the program.
1577Construction of @static@ local objects is implemented via an accompanying @static bool@ variable, which records whether the variable has already been constructed.
1578A conditional branch checks the value of the companion @bool@, and if the variable has not yet been constructed then the object is constructed.
1579The object's destructor is scheduled to be run when the program terminates using @atexit@, and the companion @bool@'s value is set so that subsequent invocations of the function will not reconstruct the object.
1580Since the parameter to @atexit@ is a parameter-less function, some additional tweaking is required.
1581First, the @static@ variable must be hoisted up to global scope and uniquely renamed to prevent name clashes with other global objects.
1582Second, a function is built which calls the destructor for the newly hoisted variable.
1583Finally, the newly generated function is registered with @atexit@, instead of registering the destructor directly.
1584Since @atexit@ calls functions in the reverse order in which they are registered, @static@ local variables are guaranteed to be destructed in the reverse order that they are constructed, which may differ between multiple executions of the same program.
1586Extending the previous example
1588int f(int x) {
1589  static X a;
1590  static X b = { x, x };  // depends on parameter value
1591  static X c = b;         // depends on local variable
1594Generates the following.
1596static struct X a_static_var0;
1597static void __a_dtor_atexit0(void){
1598  ((void)^?{}(((struct X *)(&a_static_var0))));
1600static struct X b_static_var1;
1601static void __b_dtor_atexit1(void){
1602  ((void)^?{}(((struct X *)(&b_static_var1))));
1604static struct X c_static_var2;
1605static void __c_dtor_atexit2(void){
1606  ((void)^?{}(((struct X *)(&c_static_var2))));
1608int f(int x){
1609  int _retval_f;
1610  __attribute__ ((unused)) static void *_dummy0;
1611  static _Bool __a_uninitialized = 1;
1612  if ( __a_uninitialized ) {
1613    ((void)?{}(((struct X *)(&a_static_var0))));
1614    ((void)(__a_uninitialized=0));
1615    ((void)atexit(__a_dtor_atexit0));
1616  }
1618  __attribute__ ((unused)) static void *_dummy1;
1619  static _Bool __b_uninitialized = 1;
1620  if ( __b_uninitialized ) {
1621    ((void)?{}(((struct X *)(&b_static_var1)), x, x));
1622    ((void)(__b_uninitialized=0));
1623    ((void)atexit(__b_dtor_atexit1));
1624  }
1626  __attribute__ ((unused)) static void *_dummy2;
1627  static _Bool __c_uninitialized = 1;
1628  if ( __c_uninitialized ) {
1629    ((void)?{}(((struct X *)(&c_static_var2)), b_static_var1));
1630    ((void)(__c_uninitialized=0));
1631    ((void)atexit(__c_dtor_atexit2));
1632  }
1636\subsection{Constructor Expressions}
1637In \CFA, it is possible to use a constructor as an expression.
1638Like other operators, the function name @?{}@ matches its operator syntax.
1639For example, @(&x){}@ calls the default constructor on the variable @x@, and produces @&x@ as a result.
1640The significance of constructors as expressions rather than as statements is that the result of a constructor expression can be used as part of a larger expression.
1641A key example is the use of constructor expressions to initialize the result of a call to standard C routine @malloc@.
1643struct X { ... };
1644void ?{}(X *, double);
1645X * x = malloc(sizeof(X)){ 1.5 };
1647In this example, @malloc@ dynamically allocates storage and initializes it using a constructor, all before assigning it into the variable @x@.
1648If this extension is not present, constructing dynamically allocated objects is much more cumbersome, requiring separate initialization of the pointer and initialization of the pointed-to memory.
1650X * x = malloc(sizeof(X));
1651x{ 1.5 };
1653Not only is this verbose, but it is also more error prone, since this form allows maintenance code to easily sneak in between the initialization of @x@ and the initialization of the memory that @x@ points to.
1654This feature is implemented via a transformation produceing the value of the first argument of the constructor, since constructors do not themslves have a return value.
1655Since this transformation results in two instances of the subexpression, care is taken to allocate a temporary variable to hold the result of the subexpression in the case where the subexpression may contain side effects.
1656The previous example generates the following code.
1658struct X *_tmp_ctor;
1659struct X *x = ?{}((_tmp_ctor=((_tmp_cp_ret0=
1660  malloc(sizeof(struct X))), _tmp_cp_ret0))), 1.5), _tmp_ctor);
1662It should be noted that this technique is not exclusive to @malloc@, and allows a user to write a custom allocator that can be idiomatically used in much the same way as a constructed @malloc@ call.
1664It is also possible to use operator syntax with destructors.
1665Unlike constructors, operator syntax with destructors is a statement and thus does not produce a value, since the destructed object is invalidated by the use of a destructor.
1666For example, \lstinline!^(&x){}! calls the destructor on the variable @x@.
Note: See TracBrowser for help on using the repository browser.