source: doc/theses/rob_schluntz_MMath/ctordtor.tex @ 52f6250

ADTarm-ehast-experimentalenumforall-pointer-decayjacob/cs343-translationnew-ast-unique-exprpthread-emulationqualifiedEnum
Last change on this file since 52f6250 was 67982887, checked in by Peter A. Buhr <pabuhr@…>, 6 years ago

specialize thesis directory-names

  • Property mode set to 100644
File size: 64.9 KB
Line 
1%======================================================================
2\chapter{Constructors and Destructors}
3%======================================================================
4
5% TODO now: as an experiment, implement Andrei Alexandrescu's ScopeGuard http://www.drdobbs.com/cpp/generic-change-the-way-you-write-excepti/184403758?pgno=2
6% doesn't seem possible to do this without allowing ttype on generic structs?
7
8Since \CFA is a true systems language, it does not require a garbage collector.
9As well, \CFA is not an object-oriented programming language, \ie, structures cannot have methods.
10While structures can have function pointer members, this is different from methods, since methods have implicit access to structure members and methods cannot be reassigned.
11Nevertheless, one important goal is to reduce programming complexity and increase safety.
12To that end, \CFA provides support for implicit pre/post-execution of routines for objects, via constructors and destructors.
13
14This chapter details the design of constructors and destructors in \CFA, along with their current implementation in the translator.
15Generated code samples have been edited for clarity and brevity.
16
17\section{Design Criteria}
18\label{s:Design}
19In designing constructors and destructors for \CFA, the primary goals were ease of use and maintaining backwards compatibility.
20
21In C, when a variable is defined, its value is initially undefined unless it is explicitly initialized or allocated in the static area.
22\begin{cfacode}
23int main() {
24  int x;        // uninitialized
25  int y = 5;    // initialized to 5
26  x = y;        // assigned 5
27  static int z; // initialized to 0
28}
29\end{cfacode}
30In the example above, @x@ is defined and left uninitialized, while @y@ is defined and initialized to 5.
31Next, @x@ is assigned the value of @y@.
32In the last line, @z@ is implicitly initialized to 0 since it is marked @static@.
33The key difference between assignment and initialization being that assignment occurs on a live object (\ie, an object that contains data).
34It is important to note that this means @x@ could have been used uninitialized prior to being assigned, while @y@ could not be used uninitialized.
35Use of uninitialized variables yields undefined behaviour \cite[p.~558]{C11}, which is a common source of errors in C programs.
36
37Initialization of a declaration is strictly optional, permitting uninitialized variables to exist.
38Furthermore, declaration initialization is limited to expressions, so there is no way to insert arbitrary code before a variable is live, without delaying the declaration.
39Many C compilers give good warnings for uninitialized variables most of the time, but they cannot in all cases.
40\begin{cfacode}
41int f(int *);  // output parameter: never reads, only writes
42int g(int *);  // input parameter: never writes, only reads,
43               // so requires initialized variable
44
45int x, y;
46f(&x);  // okay - only writes to x
47g(&y);  // uses y uninitialized
48\end{cfacode}
49Other languages are able to give errors in the case of uninitialized variable use, but due to backwards compatibility concerns, this is not the case in \CFA.
50
51In C, constructors and destructors are often mimicked by providing routines that create and tear down objects, where the tear down function is typically only necessary if the type modifies the execution environment.
52\begin{cfacode}
53struct array_int {
54  int * x;
55};
56struct array_int create_array(int sz) {
57  return (struct array_int) { calloc(sizeof(int)*sz) };
58}
59void destroy_rh(struct resource_holder * rh) {
60  free(rh->x);
61}
62\end{cfacode}
63This idiom does not provide any guarantees unless the structure is opaque, which then requires that all objects are heap allocated.
64\begin{cfacode}
65struct opqaue_array_int;
66struct opqaue_array_int * create_opqaue_array(int sz);
67void destroy_opaque_array(opaque_array_int *);
68int opaque_get(opaque_array_int *);  // subscript
69
70opaque_array_int * x = create_opaque_array(10);
71int x2 = opaque_get(x, 2);
72\end{cfacode}
73This pattern is cumbersome to use since every access becomes a function call, requiring awkward syntax and a performance cost.
74While useful in some situations, this compromise is too restrictive.
75Furthermore, even with this idiom it is easy to make mistakes, such as forgetting to destroy an object or destroying it multiple times.
76
77A constructor provides a way of ensuring that the necessary aspects of object initialization is performed, from setting up invariants to providing compile- and run-time checks for appropriate initialization parameters.
78This goal is achieved through a \emph{guarantee} that a constructor is called \emph{implicitly} after every object is allocated from a type with associated constructors, as part of an object's \emph{definition}.
79Since a constructor is called on every object of a managed type, it is \emph{impossible} to forget to initialize such objects, as long as all constructors perform some sensible form of initialization.
80
81In \CFA, a constructor is a function with the name @?{}@.
82Like other operators in \CFA, the name represents the syntax used to call the constructor, \eg, @struct S = { ... };@.
83Every 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).
84The @this@ parameter must have a pointer type, whose base type is the type of object that the function constructs.
85There 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.
86There is currently a proposal to add reference types to \CFA.
87Once this proposal has been implemented, the @this@ parameter will become a reference type with the same restrictions.
88
89Consider the definition of a simple type encapsulating a dynamic array of @int@s.
90
91\begin{cfacode}
92struct Array {
93  int * data;
94  int len;
95}
96\end{cfacode}
97
98In C, if the user creates an @Array@ object, the fields @data@ and @len@ are uninitialized, unless an explicit initializer list is present.
99It is the user's responsibility to remember to initialize both of the fields to sensible values, since there are no implicit checks for invalid values or reasonable defaults.
100In \CFA, the user can define a constructor to handle initialization of @Array@ objects.
101
102\begin{cfacode}
103void ?{}(Array * arr){
104  arr->len = 10;    // default size
105  arr->data = malloc(sizeof(int)*arr->len);
106  for (int i = 0; i < arr->len; ++i) {
107    arr->data[i] = 0;
108  }
109}
110Array x;  // allocates storage for Array and calls ?{}(&x)
111\end{cfacode}
112
113This 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.
114This particular form of constructor is called the \emph{default constructor}, because it is called on an object defined without an initializer.
115In other words, a default constructor is a constructor that takes a single argument: the @this@ parameter.
116
117In \CFA, a destructor is a function much like a constructor, except that its name is \lstinline!^?{}! \footnote{Originally, the name @~?{}@ was chosen for destructors, to provide familiarity to \CC programmers. Unforunately, this name causes parsing conflicts with the bitwise-not operator when used with operator syntax (see section \ref{sub:syntax}.)} and it takes only one argument.
118A destructor for the @Array@ type can be defined as:
119\begin{cfacode}
120void ^?{}(Array * arr) {
121  free(arr->data);
122}
123\end{cfacode}
124The destructor is automatically called at deallocation for all objects of type @Array@.
125Hence, the memory associated with an @Array@ is automatically freed when the object's lifetime ends.
126The exact guarantees made by \CFA with respect to the calling of destructors are discussed in section \ref{sub:implicit_dtor}.
127
128As discussed previously, the distinction between initialization and assignment is important.
129Consider the following example.
130\begin{cfacode}[numbers=left]
131Array x, y;
132Array z = x;  // initialization
133y = x;        // assignment
134\end{cfacode}
135By the previous definition of the default constructor for @Array@, @x@ and @y@ are initialized to valid arrays of length 10 after their respective definitions.
136On line 2, @z@ is initialized with the value of @x@, while on line 3, @y@ is assigned the value of @x@.
137The 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.
138In particular, these cases cannot be handled the same way because in the former case @z@ has no array, while @y@ does.
139A \emph{copy constructor} is used to perform initialization using another object of the same type.
140
141\begin{cfacode}[emph={other}, emphstyle=\color{red}]
142void ?{}(Array * arr, Array other) {  // copy constructor
143  arr->len = other.len;               // initialization
144  arr->data = malloc(sizeof(int)*arr->len)
145  for (int i = 0; i < arr->len; ++i) {
146    arr->data[i] = other.data[i];     // copy from other object
147  }
148}
149Array ?=?(Array * arr, Array other) { // assignment
150  ^?{}(arr);                          // explicitly call destructor
151  ?{}(arr, other);                    // explicitly call constructor
152  return *arr;
153}
154\end{cfacode}
155The two functions above handle the cases of initialization and assignment.
156The first function is called a copy constructor, because it constructs its argument by copying the values from another object of the same type.
157The second function is the standard copy-assignment operator.
158\CFA does not currently have the concept of reference types, so the most appropriate type for the source object in copy constructors and assignment operators is a value type.
159Appropriate care is taken in the implementation to avoid recursive calls to the copy constructor.
160The four functions (default constructor, destructor, copy constructor, and assignment operator) are special in that they safely control the state of most objects.
161
162It is possible to define a constructor that takes any combination of parameters to provide additional initialization options.
163For example, a reasonable extension to the array type would be a constructor that allocates the array to a given initial capacity and initializes the elements of the array to a given @fill@ value.
164\begin{cfacode}
165void ?{}(Array * arr, int capacity, int fill) {
166  arr->len = capacity;
167  arr->data = malloc(sizeof(int)*arr->len);
168  for (int i = 0; i < arr->len; ++i) {
169    arr->data[i] = fill;
170  }
171}
172\end{cfacode}
173
174In \CFA, constructors are called implicitly in initialization contexts.
175\begin{cfacode}
176Array x, y = { 20, 0xdeadbeef }, z = y;
177\end{cfacode}
178Constructor 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.
179One downside of reusing C initialization syntax is that it is not 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 in the program.
180
181This example generates the following code
182\begin{cfacode}
183Array x;
184?{}(&x);                  // implicit default construct
185Array y;
186?{}(&y, 20, 0xdeadbeef);  // explicit fill construct
187Array z;
188?{}(&z, y);               // copy construct
189^?{}(&z);                 // implicit destruct
190^?{}(&y);                 // implicit destruct
191^?{}(&x);                 // implicit destruct
192\end{cfacode}
193Due 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.
194This loophole is minor and exists in \CC as well.
195Destructors are implicitly called in reverse declaration-order so that objects with dependencies are destructed before the objects they are dependent on.
196
197\subsection{Calling Syntax}
198\label{sub:syntax}
199There are several ways to construct an object in \CFA.
200As previously introduced, every variable is automatically constructed at its definition, which is the most natural way to construct an object.
201\begin{cfacode}
202struct A { ... };
203void ?{}(A *);
204void ?{}(A *, A);
205void ?{}(A *, int, int);
206
207A a1;             // default constructed
208A a2 = { 0, 0 };  // constructed with 2 ints
209A a3 = a1;        // copy constructed
210// implicitly destruct a3, a2, a1, in that order
211\end{cfacode}
212Since constructors and destructors are just functions, the second way is to call the function directly.
213\begin{cfacode}
214struct A { int a; };
215void ?{}(A *);
216void ?{}(A *, A);
217void ^?{}(A *);
218
219A x;               // implicitly default constructed: ?{}(&x)
220A * y = malloc();  // copy construct: ?{}(&y, malloc())
221
222^?{}(&x);   // explicit destroy x, in different order
223?{}(&x);    // explicit construct x, second construction
224^?{}(y);    // explicit destroy y
225?{}(y, x);  // explit construct y from x, second construction
226
227// implicit ^?{}(&y);
228// implicit ^?{}(&x);
229\end{cfacode}
230Calling a constructor or destructor directly is a flexible feature that allows complete control over the management of storage.
231In particular, constructors double as a placement syntax.
232\begin{cfacode}
233struct A { ... };
234struct memory_pool { ... };
235void ?{}(memory_pool *, size_t);
236
237memory_pool pool = { 1024 };  // create an arena of size 1024
238
239A * a = allocate(&pool);      // allocate from memory pool
240?{}(a);                       // construct an A in place
241
242for (int i = 0; i < 10; i++) {
243  // reuse storage rather than reallocating
244  ^?{}(a);
245  ?{}(a);
246  // use a ...
247}
248^?{}(a);
249deallocate(&pool, a);         // return to memory pool
250\end{cfacode}
251Finally, constructors and destructors support \emph{operator syntax}.
252Like other operators in \CFA, the function name mirrors the use-case, in that the question marks are placeholders for the first $N$ arguments.
253This syntactic form is similar to the new initialization syntax in \CCeleven, except that it is used in expression contexts, rather than declaration contexts.
254\begin{cfacode}
255struct A { ... };
256struct B { A a; };
257
258A x, y, * z = &x;
259(&x){}          // default construct
260(&x){ y }       // copy construct
261(&x){ 1, 2, 3 } // construct with 3 arguments
262z{ y };         // copy construct x through a pointer
263^(&x){}         // destruct
264
265void ?{}(B * b) {
266  (&b->a){ 11, 17, 13 };  // construct a member
267}
268\end{cfacode}
269Constructor operator syntax has relatively high precedence, requiring parentheses around an address-of expression.
270Destructor operator syntax is actually an statement, and requires parentheses for symmetry with constructor syntax.
271
272One of these three syntactic forms should appeal to either C or \CC programmers using \CFA.
273
274\subsection{Constructor Expressions}
275In \CFA, it is possible to use a constructor as an expression.
276Like other operators, the function name @?{}@ matches its operator syntax.
277For example, @(&x){}@ calls the default constructor on the variable @x@, and produces @&x@ as a result.
278A key example for this capability is the use of constructor expressions to initialize the result of a call to @malloc@.
279\begin{cfacode}
280struct X { ... };
281void ?{}(X *, double);
282X * x = malloc(){ 1.5 };
283\end{cfacode}
284In this example, @malloc@ dynamically allocates storage and initializes it using a constructor, all before assigning it into the variable @x@.
285Intuitively, the expression-resolver determines that @malloc@ returns some type @T *@, as does the constructor expression since it returns the type of its argument.
286This type flows outwards to the declaration site where the expected type is known to be @X *@, thus the first argument to the constructor must be @X *@, narrowing the search space.
287
288If 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.
289\begin{cfacode}
290X * x = malloc();
291x{ 1.5 };
292\end{cfacode}
293Not 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.
294This feature is implemented via a transformation producing the value of the first argument of the constructor, since constructors do not themselves have a return value.
295Since 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.
296The previous example generates the following code.
297\begin{cfacode}
298struct X *_tmp_ctor;
299struct X *x = ?{}(  // construct result of malloc
300  _tmp_ctor=malloc_T( // store result of malloc
301    sizeof(struct X),
302    _Alignof(struct X)
303  ),
304  1.5
305), _tmp_ctor; // produce constructed result of malloc
306\end{cfacode}
307It 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.
308
309While it is possible to use operator syntax with destructors, destructors invalidate their argument, thus operator syntax with destructors is void-typed expression.
310
311\subsection{Function Generation}
312In \CFA, every type is defined to have the core set of four special functions described previously.
313Having 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.
314In addition to simplifying the definition of the language, it also simplifies the analysis that the translator must perform.
315If the translator can expect these functions to exist, then it can unconditionally attempt to resolve them.
316Moreover, the existence of a standard interface allows polymorphic code to interoperate with new types seamlessly.
317While automatic generation of assignment functions is present in previous versions of \CFA, the the implementation has been largely rewritten to accomodate constructors and destructors.
318
319To 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).
320This default is intended to maintain backwards compatibility and performance, by not imposing unexpected operations for a C programmer, as a zero-default behaviour would.
321However, it is possible for a user to define such constructors so that variables are safely zeroed by default, if desired.
322%%%%%%%%%%%%%%%%%%%%%%%%%% line width %%%%%%%%%%%%%%%%%%%%%%%%%%
323\begin{cfacode}
324void ?{}(int * i) { *i = 0; }
325forall(dtype T) void ?{}(T ** p) { *p = 0; }  // any pointer type
326void f() {
327  int x;    // initialized to 0
328  int * p;  // initialized to 0
329}
330\end{cfacode}
331%%%%%%%%%%%%%%%%%%%%%%%%%% line width %%%%%%%%%%%%%%%%%%%%%%%%%%
332
333There are several options for user-defined types: structures, unions, and enumerations.
334To aid in ease of use, the standard set of four functions is automatically generated for a user-defined type after its definition is completed.
335By auto-generating these functions, it is ensured that legacy C code continues to work correctly in every context where \CFA expects these functions to exist, since they are generated for every complete type.
336As well, these functions are always generated, since they may be needed by polymorphic functions.
337With that said, the generated functions are not called implicitly unless they are non-trivial, and are never exported, making it simple for the optimizer to strip them away when they are not used.
338
339The generated functions for enumerations are the simplest.
340Since enumerations in C are essentially just another integral type, the generated functions behave in the same way that the built-in functions for the basic types work.
341For example, given the enumeration
342\begin{cfacode}
343enum Colour {
344  R, G, B
345};
346\end{cfacode}
347The following functions are automatically generated.
348\begin{cfacode}
349void ?{}(enum Colour *_dst){
350  // default constructor does nothing
351}
352void ?{}(enum Colour *_dst, enum Colour _src){
353  *_dst=_src;  // bitwise copy
354}
355void ^?{}(enum Colour *_dst){
356  // destructor does nothing
357}
358enum Colour ?=?(enum Colour *_dst, enum Colour _src){
359  return *_dst=_src; // bitwise copy
360}
361\end{cfacode}
362In the future, \CFA will introduce strongly-typed enumerations, like those in \CC, wherein enumerations create a new type distinct from @int@ so that integral values require an explicit cast to be stored in an enumeration variable.
363The existing generated routines are sufficient to express this restriction, since they are currently set up to take in values of that enumeration type.
364Changes 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@.
365In this way, it is still possible to add an @int@ to an enumeration, but the resulting value is an @int@, meaning it cannot be reassigned to an enumeration without a cast.
366
367For structures, the situation is more complicated.
368Given 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 @$i$@.
369That is, a default constructor for @S@ default constructs the members of @S@, the copy constructor copy constructs them, and so on.
370For example, given the structure definition
371\begin{cfacode}
372struct A {
373  B b;
374  C c;
375}
376\end{cfacode}
377The following functions are implicitly generated.
378\begin{cfacode}
379void ?{}(A * this) {
380  ?{}(&this->b);  // default construct each field
381  ?{}(&this->c);
382}
383void ?{}(A * this, A other) {
384  ?{}(&this->b, other.b);  // copy construct each field
385  ?{}(&this->c, other.c);
386}
387A ?=?(A * this, A other) {
388  ?=?(&this->b, other.b);  // assign each field
389  ?=?(&this->c, other.c);
390}
391void ^?{}(A * this) {
392  ^?{}(&this->c);  // destruct each field
393  ^?{}(&this->b);
394}
395\end{cfacode}
396It is important to note that the destructors are called in reverse declaration order to prevent conflicts in the event there are dependencies among members.
397
398In addition to the standard set, a set of \emph{field constructors} is also generated for structures.
399The field constructors are constructors that consume a prefix of the structure's member-list.
400That 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.
401The addition of field constructors allows structures in \CFA to be used naturally in the same ways as used in C (\ie, to initialize any prefix of the structure), \eg, @A a0 = { b }, a1 = { b, c }@.
402Extending the previous example, the following constructors are implicitly generated for @A@.
403\begin{cfacode}
404void ?{}(A * this, B b) {
405  ?{}(&this->b, b);
406  ?{}(&this->c);
407}
408void ?{}(A * this, B b, C c) {
409  ?{}(&this->b, b);
410  ?{}(&this->c, c);
411}
412\end{cfacode}
413
414For unions, the default constructor and destructor do nothing, as it is not obvious which member, if any, should be constructed.
415For copy constructor and assignment operations, a bitwise @memcpy@ is applied.
416In 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.
417An alternative 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.
418This approach ultimately feels subtle and unsafe.
419Another option is to, like \CC, disallow unions from containing members that are themselves managed types.
420This restriction is a reasonable approach from a safety standpoint, but is not very C-like.
421Since 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.
422It 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.
423
424For example, given the union
425\begin{cfacode}
426union X {
427  Y y;
428  Z z;
429};
430\end{cfacode}
431The following functions are automatically generated.
432\begin{cfacode}
433void ?{}(union X *_dst){  // default constructor
434}
435void ?{}(union X *_dst, union X _src){  // copy constructor
436  __builtin_memcpy(_dst, &_src, sizeof(union X ));
437}
438void ^?{}(union X *_dst){  // destructor
439}
440union X ?=?(union X *_dst, union X _src){  // assignment
441  __builtin_memcpy(_dst, &_src, sizeof(union X));
442  return _src;
443}
444void ?{}(union X *_dst, struct Y src){  // construct first field
445  __builtin_memcpy(_dst, &src, sizeof(struct Y));
446}
447\end{cfacode}
448
449% 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
450In \CCeleven, unions may have managed members, with the caveat that 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.
451This restriction could easily be added into \CFA once \emph{deleted} functions are added.
452
453\subsection{Using Constructors and Destructors}
454Implicitly generated constructor and destructor calls ignore the outermost type qualifiers, \eg @const@ and @volatile@, on a type by way of a cast on the first argument to the function.
455For example,
456\begin{cfacode}
457struct S { int i; };
458void ?{}(S *, int);
459void ?{}(S *, S);
460
461const S s = { 11 };
462volatile S s2 = s;
463\end{cfacode}
464Generates the following code
465\begin{cfacode}
466const struct S s;
467?{}((struct S *)&s, 11);
468volatile struct S s2;
469?{}((struct S *)&s2, s);
470\end{cfacode}
471Here, @&s@ and @&s2@ are cast to unqualified pointer types.
472This mechanism allows the same constructors and destructors to be used for qualified objects as for unqualified objects.
473This rule applies only to implicitly generated constructor calls.
474Hence, explicitly re-initializing qualified objects with a constructor requires an explicit cast.
475
476As discussed in Section \ref{sub:c_background}, compound literals create unnamed objects.
477This mechanism can continue to be used seamlessly in \CFA with managed types to create temporary objects.
478The object created by a compound literal is constructed using the provided brace-enclosed initializer-list, and is destructed at the end of the scope it is used in.
479For example,
480\begin{cfacode}
481struct A { int x; };
482void ?{}(A *, int, int);
483{
484  int x = (A){ 10, 20 }.x;
485}
486\end{cfacode}
487is equivalent to
488\begin{cfacode}
489struct A { int x, y; };
490void ?{}(A *, int, int);
491{
492  A _tmp;
493  ?{}(&_tmp, 10, 20);
494  int x = _tmp.x;
495  ^?{}(&tmp);
496}
497\end{cfacode}
498
499Unlike \CC, \CFA provides an escape hatch that allows a user to decide at an object's definition whether it should be managed or not.
500An object initialized with \ateq is guaranteed to be initialized like a C object, and has no implicit destructor call.
501This feature provides all of the freedom that C programmers are used to having to optimize a program, while maintaining safety as a sensible default.
502\begin{cfacode}
503struct A { int * x; };
504// RAII
505void ?{}(A * a) { a->x = malloc(sizeof(int)); }
506void ^?{}(A * a) { free(a->x); }
507
508A a1;           // managed
509A a2 @= { 0 };  // unmanaged
510\end{cfacode}
511In this example, @a1@ is a managed object, and thus is default constructed and destructed at the start/end of @a1@'s lifetime, while @a2@ is an unmanaged object and is not implicitly constructed or destructed.
512Instead, @a2->x@ is initialized to @0@ as if it were a C object, because of the explicit initializer.
513
514In addition to freedom, \ateq provides a simple path for migrating legacy C code to \CFA, in that objects can be moved from C-style initialization to \CFA gradually and individually.
515It 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.
516It is recommended that most objects be managed by sensible constructors and destructors, except where absolutely necessary, such as memory-mapped devices, trigger devices, I/O controllers, etc.
517
518When a user declares any constructor or destructor, the corresponding intrinsic/generated function and all field constructors for that type are hidden, so that they are not found during expression resolution until the user-defined function goes out of scope.
519Furthermore, if the user declares any constructor, then the intrinsic/generated default constructor is also hidden, precluding default construction.
520These semantics closely mirror the rule for implicit declaration of constructors in \CC, wherein the default constructor is implicitly declared if there is no user-declared constructor \cite[p.~186]{ANSI98:C++}.
521\begin{cfacode}
522struct S { int x, y; };
523
524void f() {
525  S s0, s1 = { 0 }, s2 = { 0, 2 }, s3 = s2;  // okay
526  {
527    void ?{}(S * s, int i) { s->x = i*2; } // locally hide autogen ctors
528    S s4;  // error, no default constructor
529    S s5 = { 3 };  // okay, local constructor
530    S s6 = { 4, 5 };  // error, no field constructor
531    S s7 = s5; // okay
532  }
533  S s8, s9 = { 6 }, s10 = { 7, 8 }, s11 = s10;  // okay
534}
535\end{cfacode}
536In 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.
537
538When defining a constructor or destructor for a structure @S@, any members that are not explicitly constructed or destructed are implicitly constructed or destructed automatically.
539If an explicit call is present, then that call is taken in preference to any implicitly generated call.
540A consequence of this rule is that it is possible, unlike \CC, to precisely control the order of construction and destruction of sub-objects on a per-constructor basis, whereas in \CC sub-object initialization and destruction is always performed based on the declaration order.
541\begin{cfacode}
542struct A {
543  B w, x, y, z;
544};
545void ?{}(A * a, int i) {
546  (&a->x){ i };
547  (&a->z){ a->y };
548}
549\end{cfacode}
550Generates the following
551\begin{cfacode}
552void ?{}(A * a, int i) {
553  (&a->w){};   // implicit default ctor
554  (&a->y){};   // implicit default ctor
555  (&a->x){ i };
556  (&a->z){ a->y };
557}
558\end{cfacode}
559Finally, it is illegal for a sub-object to be explicitly constructed for the first time after it is used for the first time.
560If the translator cannot be reasonably sure that an object is constructed prior to its first use, but is constructed afterward, an error is emitted.
561More specifically, the translator searches the body of a constructor to ensure that every sub-object is initialized.
562\begin{cfacode}
563void ?{}(A * a, double x) {
564  f(a->x);
565  (&a->x){ (int)x }; // error, used uninitialized on previous line
566}
567\end{cfacode}
568However, if the translator sees a sub-object used within the body of a constructor, but does not see a constructor call that uses the sub-object 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).
569To override this rule, \ateq can be used to force the translator to trust the programmer's discretion.
570This form of \ateq is not yet implemented.
571\begin{cfacode}
572void ?{}(A * a) {
573  // default constructs all members
574  f(a->x);
575}
576
577void ?{}(A * a, A other) {
578  // copy constructs all members
579  f(a->y);
580}
581
582void ?{}(A * a, int x) {
583  // object forwarded to another constructor,
584  // does not implicitly construct any members
585  (&a){};
586}
587
588void ^?{}(A * a) {
589  ^(&a->x){}; // explicit destructor call
590} // z, y, w implicitly destructed, in this order
591\end{cfacode}
592If at any point, the @this@ parameter is passed directly as the target of another constructor, then it is assumed the other constructor handles the initialization of all of the object's members and no implicit constructor calls are added to the current constructor.
593
594Despite great effort, some forms of C syntax do not work well with constructors in \CFA.
595In 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.
596\begin{cfacode}
597// all legal forward declarations in C
598void f(int, int, int);
599void f(int a, int b, int c);
600void f(int b, int c, int a);
601void f(int c, int a, int b);
602void f(int x, int y, int z);
603
604f(b:10, a:20, c:30);  // which parameter is which?
605\end{cfacode}
606In 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.
607Furthermore, a function prototype can be repeated an arbitrary number of times, each time using different names.
608As 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.
609
610\begin{sloppypar}
611In addition, constructor calls do not support unnamed nesting.
612\begin{cfacode}
613struct B { int x; };
614struct C { int y; };
615struct A { B b; C c; };
616void ?{}(A *, B);
617void ?{}(A *, C);
618
619A a = {
620  { 10 },  // construct B? - invalid
621};
622\end{cfacode}
623In C, nesting initializers means that the programmer intends to initialize sub-objects with the nested initializers.
624The reason for this omission is to both simplify the mental model for using constructors, and to make initialization simpler for the expression resolver.
625If 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.
626That is, in the previous example the line marked as an error could mean construct using @?{}(A *, B)@ or with @?{}(A *, C)@, since the inner initializer @{ 10 }@ could be taken as an intermediate object of type @B@ or @C@.
627In 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.
628\end{sloppypar}
629
630More precisely, constructor calls cannot have a nesting depth greater than the number of array dimensions in the type of the initialized object, plus one.
631For example,
632\begin{cfacode}
633struct A;
634void ?{}(A *, int);
635void ?{}(A *, A, A);
636
637A a1[3] = { { 3 }, { 4 }, { 5 } };
638A a2[2][2] = {
639  { { 9 }, { 10 } },  // a2[0]
640  { {14 }, { 15 } }   // a2[1]
641};
642A a3[4] = { // 1 dimension => max depth 2
643  { { 11 }, { 12 } },  // error, three levels deep
644  { 80 }, { 90 }, { 100 }
645}
646\end{cfacode}
647The body of @A@ has been omitted, since only the constructor interfaces are important.
648
649It should be noted that unmanaged objects, i.e. objects that have only trivial constructors, can still make use of designations and nested initializers in \CFA.
650It is simple to overcome this limitation for managed objects by making use of compound literals, so that the arguments to the constructor call are explicitly typed.
651%%%%%%%%%%%%%%%%%%%%%%%%%% line width %%%%%%%%%%%%%%%%%%%%%%%%%%
652\begin{cfacode}
653struct B { int x; };
654struct C { int y; };
655struct A { B b; C c; };
656void ?{}(A *, B);
657void ?{}(A *, C);
658
659A a = {
660  (C){ 10 } // disambiguate with compound literal
661};
662\end{cfacode}
663%%%%%%%%%%%%%%%%%%%%%%%%%% line width %%%%%%%%%%%%%%%%%%%%%%%%%%
664
665\subsection{Implicit Destructors}
666\label{sub:implicit_dtor}
667Destructors are automatically called at the end of the block in which the object is declared.
668In addition to this, destructors are automatically called when statements manipulate control flow to leave a block in which the object is declared, \eg, with return, break, continue, and goto statements.
669The example below demonstrates a simple routine with multiple return statements.
670\begin{cfacode}
671struct A;
672void ^?{}(A *);
673
674void f(int i) {
675  A x;  // construct x
676  {
677    A y; // construct y
678    {
679      A z; // construct z
680      {
681        if (i == 0) return; // destruct x, y, z
682      }
683      if (i == 1) return; // destruct x, y, z
684    } // destruct z
685    if (i == 2) return; // destruct x, y
686  } // destruct y
687} // destruct x
688\end{cfacode}
689
690The next example illustrates the use of simple continue and break statements and the manner that they interact with implicit destructors.
691\begin{cfacode}
692for (int i = 0; i < 10; i++) {
693  A x;
694  if (i == 2) {
695    continue;  // destruct x
696  } else if (i == 3) {
697    break;     // destruct x
698  }
699} // destruct x
700\end{cfacode}
701Since 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.
702In the case where @i@ is @2@, the continue statement runs the loop update expression and attempts to begin the next iteration of the loop.
703Since continue is a C statement, which does not understand destructors, it is transformed into a @goto@ statement that branches to the end of the loop, just before the block's destructors, to ensure that @x@ is destructed.
704When @i@ is @3@, the break statement moves control to just past the end of the loop.
705Unlike the previous case, the destructor for @x@ cannot be reused, so a destructor call for @x@ is inserted just before the break statement.
706
707\CFA also supports labeled break and continue statements, which allow more precise manipulation of control flow.
708Labeled break and continue allow the programmer to specify which control structure to target by using a label attached to a control structure.
709\begin{cfacode}[emph={L1,L2}, emphstyle=\color{red}]
710L1: for (int i = 0; i < 10; i++) {
711  A x;
712  for (int j = 0; j < 10; j++) {
713    A y;
714    if (i == 1) {
715      continue L1; // destruct y
716    } else if (i == 2) {
717      break L1;    // destruct x,y
718    }
719  } // destruct y
720} // destruct X
721\end{cfacode}
722The statement @continue L1@ begins the next iteration of the outer for-loop.
723Since the semantics of continue require the loop update expression to execute, control branches to the 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@.
724Break, on the other hand, requires jumping out of both loops, so the destructors for both @x@ and @y@ are generated and inserted before the @break L1@ statement.
725
726Finally, an example which demonstrates goto.
727Since goto is a general mechanism for jumping to different locations in the program, a more comprehensive approach is required.
728For 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 $L$.
729If 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.
730Then, 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.
731\begin{cfacode}
732int i = 0;
733{
734  L0: ;     // S_L0 = { x }
735    A y;
736  L1: ;     // S_L1 = { x }
737    A x;
738  L2: ;     // S_L2 = { y, x }
739    if (i == 0) {
740      ++i;
741      goto L1;    // S_G = { y, x }
742      // S_G-S_L1 = { x } => destruct x
743    } else if (i == 1) {
744      ++i;
745      goto L2;    // S_G = { y, x }
746      // S_G-S_L2 = {} => destruct nothing
747    } else if (i == 2) {
748      ++i;
749      goto L3;    // S_G = { y, x }
750      // S_G-S_L3 = {}
751    } else if (false) {
752      ++i;
753      A z;
754      goto L3;    // S_G = { z, y, x }
755      // S_G-S_L3 = { z } => destruct z
756    } else {
757      ++i;
758      goto L4;    // S_G = { y, x }
759      // S_G-S_L4 = { y, x } => destruct y, x
760    }
761  L3: ;    // S_L3 = { y, x }
762    goto L2;      // S_G = { y, x }
763    // S_G-S_L2 = {}
764}
765L4: ;  // S_L4 = {}
766if (i == 4) {
767  goto L0;        // S_G = {}
768  // S_G-S_L0 = {}
769}
770\end{cfacode}
771All break and continue statements are implemented in \CFA in terms of goto statements, so the more constrained forms are precisely governed by these rules.
772
773The next example demonstrates the error case.
774\begin{cfacode}
775{
776    goto L1; // S_G = {}
777    // S_L1-S_G = { y } => error
778    A y;
779  L1: ; // S_L1 = { y }
780    A x;
781  L2: ; // S_L2 = { y, x }
782}
783goto L2; // S_G = {}
784// S_L2-S_G = { y, x } => error
785\end{cfacode}
786
787While \CFA supports the GCC computed-goto extension, the behaviour of managed objects in combination with computed-goto is undefined.
788\begin{cfacode}
789void f(int val) {
790  void * l = val == 0 ? &&L1 : &&L2;
791  {
792      A x;
793    L1: ;
794      goto *l;  // branches differently depending on argument
795  }
796  L2: ;
797}
798\end{cfacode}
799Likewise, destructors are not executed at scope-exit due to a computed-goto in \CC, as of g++ version 6.2.
800
801\subsection{Implicit Copy Construction}
802\label{s:implicit_copy_construction}
803When 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.
804When a value is returned from a function, the copy constructor is called to pass the value back to the call site.
805Exempt from these rules are intrinsic and built-in functions.
806It 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.
807That is, since the parameter is not marked as an unmanaged object using \ateq, it is copy constructed if it is returned by value or passed as an argument to another function, so to guarantee consistent behaviour, unmanaged objects must be copy constructed when passed as arguments.
808These semantics are important to bear in mind when using unmanaged objects, and could produce unexpected results when mixed with objects that are explicitly constructed.
809\begin{cfacode}
810struct A { ... };
811void ?{}(A *);
812void ?{}(A *, A);
813void ^?{}(A *);
814
815A identity(A x) { // pass by value => need local copy
816  return x;       // return by value => make call-site copy
817}
818
819A y, z @= {};
820identity(y);  // copy construct y into x
821identity(z);  // copy construct z into x
822\end{cfacode}
823Note that unmanaged argument @z@ is logically copy constructed into managed parameter @x@; however, the translator must copy construct into a temporary variable to be passed as an argument, which is also destructed after the call.
824A compiler could by-pass the argument temporaries since it is in control of the calling conventions and knows exactly where the called-function's parameters live.
825
826This generates the following
827\begin{cfacode}
828struct A f(struct A x){
829  struct A _retval_f;    // return value
830  ?{}((&_retval_f), x);  // copy construct return value
831  return _retval_f;
832}
833
834struct A y;
835?{}(&y);                 // default construct
836struct A z = { 0 };      // C default
837
838struct A _tmp_cp1;       // argument 1
839struct A _tmp_cp_ret0;   // return value
840_tmp_cp_ret0=f(
841  (?{}(&_tmp_cp1, y) , _tmp_cp1)  // argument is a comma expression
842), _tmp_cp_ret0;         // return value for cascading
843^?{}(&_tmp_cp_ret0);     // destruct return value
844^?{}(&_tmp_cp1);         // destruct argument 1
845
846struct A _tmp_cp2;       // argument 1
847struct A _tmp_cp_ret1;   // return value
848_tmp_cp_ret1=f(
849  (?{}(&_tmp_cp2, z), _tmp_cp2)  // argument is a common expression
850), _tmp_cp_ret1;         // return value for cascading
851^?{}(&_tmp_cp_ret1);     // destruct return value
852^?{}(&_tmp_cp2);         // destruct argument 1
853^?{}(&y);
854\end{cfacode}
855
856A special syntactic form, such as a variant of \ateq, can be implemented to specify at the call site that an argument should not be copy constructed, to regain some control for the C programmer.
857\begin{cfacode}
858identity(z@);  // do not copy construct argument
859               // - will copy construct/destruct return value
860A@ identity_nocopy(A @ x) {  // argument not copy constructed or destructed
861  return x;  // not copy constructed
862             // return type marked @ => not destructed
863}
864\end{cfacode}
865It should be noted that reference types will allow specifying that a value does not need to be copied, however reference types do not provide a means of preventing implicit copy construction from uses of the reference, so the problem is still present when passing or returning the reference by value.
866
867Adding implicit copy construction imposes the additional runtime cost of the copy constructor for every argument and return value in a function call.
868This cost is necessary to maintain appropriate value semantics when calling a function.
869In the future, return-value-optimization (RVO) can be implemented for \CFA to elide unnecessary copy construction and destruction of temporary objects.
870This cost is not present for types with trivial copy constructors and destructors.
871
872A known issue with this implementation is that the argument and return value temporaries are not guaranteed to have the same address for their entire lifetimes.
873In the previous example, 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.
874This approach works out most of the time, because typically destructors need to only access the fields of the object and recursively destroy.
875It is currently the case that constructors and destructors that use the @this@ pointer as a unique identifier to store data externally do not work correctly for return value objects.
876Thus, it is currently not safe to rely on an object's @this@ pointer to remain constant throughout execution of the program.
877\begin{cfacode}
878A * external_data[32];
879int ext_count;
880struct A;
881void ?{}(A * a) {
882  // ...
883  external_data[ext_count++] = a;
884}
885void ^?{}(A * a) {
886  for (int i = 0; i < ext_count) {
887    if (a == external_data[i]) { // may never be true
888      // ...
889    }
890  }
891}
892
893A makeA() {
894  A x;  // stores &x in external_data
895  return x;
896}
897makeA();  // return temporary has a different address than x
898// equivalent to:
899//   A _tmp;
900//   _tmp = makeA(), _tmp;
901//   ^?{}(&_tmp);
902\end{cfacode}
903In the above example, a global array of pointers is used to keep track of all of the allocated @A@ objects.
904Due to copying on return, the current object being destructed does not exist in the array if an @A@ object is ever returned by value from a function, such as in @makeA@.
905
906This problem could be solved in the translator by changing the function signatures so that the return value is moved into the parameter list.
907For example, the translator could restructure the code like so
908\begin{cfacode}
909void f(struct A x, struct A * _retval_f){
910  ?{}(_retval_f, x);  // construct directly into caller's stack frame
911}
912
913struct A y;
914?{}(&y);
915struct A z = { 0 };
916
917struct A _tmp_cp1;     // argument 1
918struct A _tmp_cp_ret0; // return value
919f((?{}(&_tmp_cp1, y) , _tmp_cp1), &_tmp_cp_ret0), _tmp_cp_ret0;
920^?{}(&_tmp_cp_ret0);   // return value
921^?{}(&_tmp_cp1);       // argument 1
922\end{cfacode}
923This transformation provides @f@ with the address of the return variable so that it can be constructed into directly.
924It is worth pointing out that this kind of signature rewriting already occurs in polymorphic functions that return by value, as discussed in \cite{Bilson03}.
925A 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, \eg
926\begin{cfacode}
927struct A { int v; };
928A x; // unmanaged, since only trivial constructors are available
929{
930  void ?{}(A * a) { ... }
931  void ^?{}(A * a) { ... }
932  A y; // managed
933}
934A z; // unmanaged
935\end{cfacode}
936Hence there is not enough information to determine at function declaration 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.
937Even 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.
938Furthermore, it is not possible to overload C functions, so using @extern "C"@ to declare functions is of limited use.
939
940It would be possible to regain some control by adding an attribute to structures that 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.
941Ideally, structures should be manageable by default, since otherwise the default case becomes more verbose.
942This means that in general, function signatures would have to be rewritten, and in a select few cases the signatures would not be rewritten.
943\begin{cfacode}
944__attribute__((manageable)) struct A { ... };   // can declare ctors
945__attribute__((unmanageable)) struct B { ... }; // cannot declare ctors
946struct C { ... };                               // can declare ctors
947
948A f();  // rewritten void f(A *);
949B g();  // not rewritten
950C h();  // rewritten void h(C *);
951\end{cfacode}
952An alternative is to make the attribute \emph{identifiable}, which states that objects of this type use the @this@ parameter as an identity.
953This strikes more closely to the visible 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.
954Furthermore, no restrictions would need to be placed on whether objects can be constructed.
955\begin{cfacode}
956__attribute__((identifiable)) struct A { ... };  // can declare ctors
957struct B { ... };                                // can declare ctors
958
959A f();  // rewritten void f(A *);
960B g();  // not rewritten
961\end{cfacode}
962
963Ultimately, both of these are patchwork solutions.
964Since a real compiler has full control over its calling conventions, it can seamlessly allow passing the return parameter without outwardly changing the signature of a routine.
965As such, it has been decided that this issue is not currently a priority and will be fixed when a full \CFA compiler is implemented.
966
967\section{Implementation}
968\subsection{Array Initialization}
969Arrays are a special case in the C type-system.
970Type checking largely ignores size information for C arrays, making it impossible to write a standalone \CFA function that constructs or destructs an array, while maintaining the standard interface for constructors and destructors.
971Instead, \CFA defines the initialization and destruction of an array recursively.
972That is, when an array is defined, each of its elements is constructed in order from element 0 up to element $n-1$.
973When an array is to be implicitly destructed, each of its elements is destructed in reverse order from element $n-1$ down to element 0.
974As in C, it is possible to explicitly provide different initializers for each element of the array through array initialization syntax.
975In this case, each of the initializers is taken in turn to construct a subsequent element of the array.
976If too many initializers are provided, only the initializers up to N are actually used.
977If too few initializers are provided, then the remaining elements are default constructed.
978
979For example, given the following code.
980\begin{cfacode}
981struct X {
982  int x, y, z;
983};
984void f() {
985  X x[10] = { { 1, 2, 3 }, { 4 }, { 7, 8 } };
986}
987\end{cfacode}
988The following code is generated for @f@.
989\begin{cfacode}
990void f(){
991  struct X x[((long unsigned int )10)];
992  // construct x
993  {
994    int _index0 = 0;
995    // construct with explicit initializers
996    {
997      if (_index0<10) ?{}(&x[_index0], 1, 2, 3);
998      ++_index0;
999      if (_index0<10) ?{}(&x[_index0], 4);
1000      ++_index0;
1001      if (_index0<10) ?{}(&x[_index0], 7, 8);
1002      ++_index0;
1003    }
1004
1005    // default construct remaining elements
1006    for (;_index0<10;++_index0) {
1007      ?{}(&x[_index0]);
1008    }
1009  }
1010  // destruct x
1011  {
1012    int _index1 = 10-1;
1013    for (;_index1>=0;--_index1) {
1014      ^?{}(&x[_index1]);
1015    }
1016  }
1017}
1018\end{cfacode}
1019Multidimensional arrays require more complexity.
1020For example, a two dimensional array
1021\begin{cfacode}
1022void g() {
1023  X x[10][10] = {
1024    { { 1, 2, 3 }, { 4 } }, // x[0]
1025    { { 7, 8 } }            // x[1]
1026  };
1027}\end{cfacode}
1028Generates the following
1029\begin{cfacode}
1030void g(){
1031  struct X x[10][10];
1032  // construct x
1033  {
1034    int _index0 = 0;
1035    for (;_index0<10;++_index0) {
1036      {
1037        int _index1 = 0;
1038        // construct with explicit initializers
1039        {
1040          switch ( _index0 ) {
1041            case 0:
1042              // construct first array
1043              if ( _index1<10 ) ?{}(&x[_index0][_index1], 1, 2, 3);
1044              ++_index1;
1045              if ( _index1<10 ) ?{}(&x[_index0][_index1], 4);
1046              ++_index1;
1047              break;
1048            case 1:
1049              // construct second array
1050              if ( _index1<10 ) ?{}(&x[_index0][_index1], 7, 8);
1051              ++_index1;
1052              break;
1053          }
1054        }
1055        // default construct remaining elements
1056        for (;_index1<10;++_index1) {
1057            ?{}(&x[_index0][_index1]);
1058        }
1059      }
1060    }
1061  }
1062  // destruct x
1063  {
1064    int _index2 = 10-1;
1065    for (;_index2>=0;--_index2) {
1066      {
1067        int _index3 = 10-1;
1068        for (;_index3>=0;--_index3) {
1069            ^?{}(&x[_index2][_index3]);
1070        }
1071      }
1072    }
1073  }
1074}
1075\end{cfacode}
1076% 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.
1077% It is simple to remove the increment statements for @_index1@, but it is not simple to remove the
1078%% technically, it's not hard either. I could easily downcast and change the second argument to ?[?], but is it really necessary/worth it??
1079
1080\subsection{Global Initialization}
1081In standard C, global variables can only be initialized to compile-time constant expressions, which places strict limitations on the programmer's ability to control the default values of objects.
1082In \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.
1083By default, objects within a translation unit are constructed in declaration order, and destructed in the reverse order.
1084The default order of construction of objects amongst translation units is unspecified.
1085It is, however, guaranteed that any global objects in the standard library are initialized prior to the initialization of any object in a user program.
1086
1087This 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[6.31.1]{GCCExtensions}.
1088A similar function is generated with the \emph{destructor} attribute, which handles all global destructor calls.
1089At 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.
1090This 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 tear-down routines.
1091
1092For example, given the following global declarations.
1093\begin{cfacode}
1094struct X {
1095  int y, z;
1096};
1097void ?{}(X *);
1098void ?{}(X *, int, int);
1099void ^?{}(X *);
1100
1101X a;
1102X b = { 10, 3 };
1103\end{cfacode}
1104The following code is generated.
1105\begin{cfacode}
1106__attribute__ ((constructor)) static void _init_global_ctor(void){
1107  ?{}(&a);
1108  ?{}(&b, 10, 3);
1109}
1110__attribute__ ((destructor)) static void _destroy_global_ctor(void){
1111  ^?{}(&b);
1112  ^?{}(&a);
1113}
1114\end{cfacode}
1115
1116%   https://gcc.gnu.org/onlinedocs/gcc/C_002b_002b-Attributes.html#C_002b_002b-Attributes
1117% 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
1118GCC provides an attribute @init_priority@ in \CC, which allows specifying the relative priority for initialization of global objects on a per-object basis.
1119A similar attribute can be implemented in \CFA by pulling marked objects into global constructor/destructor-attribute functions with the specified priority.
1120For example,
1121\begin{cfacode}
1122struct A { ... };
1123void ?{}(A *, int);
1124void ^?{}(A *);
1125__attribute__((init_priority(200))) A x = { 123 };
1126\end{cfacode}
1127would generate
1128\begin{cfacode}
1129A x;
1130__attribute__((constructor(200))) __init_x() {
1131  ?{}(&x, 123);  // construct x with priority 200
1132}
1133__attribute__((destructor(200))) __destroy_x() {
1134  ?{}(&x);       // destruct x with priority 200
1135}
1136\end{cfacode}
1137
1138\subsection{Static Local Variables}
1139In standard C, it is possible to mark variables that are local to a function with the @static@ storage class.
1140Unlike 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.
1141Much like global variables, @static@ variables can only be initialized to a \emph{compile-time constant value} so that a compiler is able to create storage for the variable and initialize it at compile-time.
1142
1143Yet again, this rule is too restrictive for a language with constructors and destructors.
1144Since the initializer expression is not necessarily a compile-time constant and can depend on the current execution state of the function, \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.
1145Since standard C does not allow access to a @static@ local variable before the first time control flow reaches the declaration, this change does not preclude any valid C code.
1146Local objects with @static@ storage class are only implicitly constructed and destructed once for the duration of the program.
1147The object is constructed when its declaration is reached for the first time.
1148The object is destructed once at the end of the program.
1149
1150Construction of @static@ local objects is implemented via an accompanying @static bool@ variable, which records whether the variable has already been constructed.
1151A conditional branch checks the value of the companion @bool@, and if the variable has not yet been constructed then the object is constructed.
1152The object's destructor is scheduled to be run when the program terminates using @atexit@ \footnote{When using the dynamic linker, it is possible to dynamically load and unload a shared library. Since glibc 2.2.3 \cite{atexit}, functions registered with @atexit@ within the shared library are called when unloading the shared library. As such, static local objects can be destructed using this mechanism even in shared libraries on Linux systems.}, and the companion @bool@'s value is set so that subsequent invocations of the function do not reconstruct the object.
1153Since the parameter to @atexit@ is a parameter-less function, some additional tweaking is required.
1154First, the @static@ variable must be hoisted up to global scope and uniquely renamed to prevent name clashes with other global objects.
1155If necessary, a local structure may need to be hoisted, as well.
1156Second, a function is built that calls the destructor for the newly hoisted variable.
1157Finally, the newly generated function is registered with @atexit@, instead of registering the destructor directly.
1158Since @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.
1159Extending the previous example
1160\begin{cfacode}
1161int f(int x) {
1162  static X a;
1163  static X b = { x, x };  // depends on parameter value
1164  static X c = b;         // depends on local variable
1165}
1166\end{cfacode}
1167Generates the following.
1168\begin{cfacode}
1169static struct X a_static_var0;
1170static void __a_dtor_atexit0(void){
1171  ((void)^?{}(((struct X *)(&a_static_var0))));
1172}
1173static struct X b_static_var1;
1174static void __b_dtor_atexit1(void){
1175  ((void)^?{}(((struct X *)(&b_static_var1))));
1176}
1177static struct X c_static_var2;
1178static void __c_dtor_atexit2(void){
1179  ((void)^?{}(((struct X *)(&c_static_var2))));
1180}
1181int f(int x){
1182  int _retval_f;
1183  __attribute__ ((unused)) static void *_dummy0;
1184  static _Bool __a_uninitialized = 1;
1185  if ( __a_uninitialized ) {
1186    ((void)?{}(((struct X *)(&a_static_var0))));
1187    ((void)(__a_uninitialized=0));
1188    ((void)atexit(__a_dtor_atexit0));
1189  }
1190
1191  __attribute__ ((unused)) static void *_dummy1;
1192  static _Bool __b_uninitialized = 1;
1193  if ( __b_uninitialized ) {
1194    ((void)?{}(((struct X *)(&b_static_var1)), x, x));
1195    ((void)(__b_uninitialized=0));
1196    ((void)atexit(__b_dtor_atexit1));
1197  }
1198
1199  __attribute__ ((unused)) static void *_dummy2;
1200  static _Bool __c_uninitialized = 1;
1201  if ( __c_uninitialized ) {
1202    ((void)?{}(((struct X *)(&c_static_var2)), b_static_var1));
1203    ((void)(__c_uninitialized=0));
1204    ((void)atexit(__c_dtor_atexit2));
1205  }
1206}
1207\end{cfacode}
1208
1209This implementation comes at the runtime cost of an additional branch for every @static@ local variable, each time the function is called.
1210Since initializers are not required to be compile-time constant expressions, they can involve global variables, function arguments, function calls, etc.
1211As a direct consequence, @static@ local variables cannot be initialized with an attribute-constructor routines like global variables can.
1212However, in the case where the variable is unmanaged and has a compile-time constant initializer, a C-compliant initializer is generated and the additional cost is not present.
1213\CC shares the same semantics for its @static@ local variables.
1214
1215\subsection{Polymorphism}
1216As mentioned in section \ref{sub:polymorphism}, \CFA currently has 3 type-classes that are used to designate polymorphic data types: @otype@, @dtype@, and @ftype@.
1217In previous versions of \CFA, @otype@ was syntactic sugar for @dtype@ with known size/alignment information and an assignment function.
1218That is,
1219\begin{cfacode}
1220forall(otype T)
1221void f(T);
1222\end{cfacode}
1223was equivalent to
1224\begin{cfacode}
1225forall(dtype T | sized(T) | { T ?=?(T *, T); })
1226void f(T);
1227\end{cfacode}
1228This allows easily specifying constraints that are common to all complete object-types very simply.
1229
1230Now that \CFA has constructors and destructors, more of a complete object's behaviour can be specified than was previously possible.
1231As such, @otype@ has been augmented to include assertions for a default constructor, copy constructor, and destructor.
1232That is, the previous example is now equivalent to
1233\begin{cfacode}
1234forall(dtype T | sized(T) |
1235  { T ?=?(T *, T); void ?{}(T *); void ?{}(T *, T); void ^?{}(T *); })
1236void f(T);
1237\end{cfacode}
1238These additions allow @f@'s body to create and destroy objects of type @T@, and pass objects of type @T@ as arguments to other functions, following the normal \CFA rules.
1239A point of note here is that objects can be missing default constructors (and eventually other functions through deleted functions), so it is important for \CFA programmers to think carefully about the operations needed by their function, as to not over-constrain the acceptable parameter types and prevent potential reuse.
1240
1241These additional assertion parameters impose a runtime cost on all managed temporary objects created in polymorphic code, even those with trivial constructors and destructors.
1242This cost is necessary because polymorphic code does not know the actual type at compile-time, due to separate compilation.
1243Since trivial constructors and destructors either do not perform operations or are simply bit-wise copy operations, the imposed cost is essentially the cost of the function calls.
1244
1245\section{Summary}
1246
1247When creating a new object of a managed type, it is guaranteed that a constructor is be called to initialize the object at its definition point, and is destructed when the object's lifetime ends.
1248Destructors are called in the reverse order of construction.
1249
1250Every argument passed to a function is copy constructed into a temporary object that is passed by value to the functions and destructed at the end of the statement.
1251Function return values are copy constructed inside the function at the return statement, passed by value to the call-site, and destructed at the call-site at the end of the statement.
1252
1253Every complete object type has a default constructor, copy constructor, assignment operator, and destructor.
1254To accomplish this, these functions are generated as appropriate for new types.
1255User-defined functions shadow built-in and automatically generated functions, so it is possible to specialize the behaviour of a type.
1256Furthermore, default constructors and aggregate field constructors are hidden when \emph{any} constructor is defined.
1257
1258Objects dynamically allocated with @malloc@, \ateq objects, and objects with only trivial constructors and destructors are unmanaged.
1259Unmanaged objects are never the target of an implicit constructor or destructor call.
Note: See TracBrowser for help on using the repository browser.