Blame - docs/AutomaticReferenceCounting.html - fp2-dev/platform/external/clang

blob: ff65f3be48016874b365d1d7c9f7515a997df39d [file] [log] [blame]

Benjamin Kramer	665a8dc	2012-01-15 15:26:07 +0000	[diff] [blame]	1	<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
				2	"http://www.w3.org/TR/html4/strict.dtd">
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	3	<html>
				4	<head>
				5	<title>Objective-C Automatic Reference Counting (ARC)</title>
Benjamin Kramer	665a8dc	2012-01-15 15:26:07 +0000	[diff] [blame]	6	<link type="text/css" rel="stylesheet" href="../menu.css">
				7	<link type="text/css" rel="stylesheet" href="../content.css">
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	8	<style type="text/css">
				9	/* Collapse the items in the ToC to the left. */
				10	div#toc ul {
				11	padding-left: 0
				12	}
				13
				14	/* Rationales appear in italic. */
				15	div.rationale {
				16	font-style: italic
				17	}
				18
John McCall	f3d08a6	2011-06-18 07:31:30 +0000	[diff] [blame]	19	div.rationale em {
				20	font-style: normal
				21	}
				22
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	23	/* Revisions are also italicized. */
				24	span.revision {
				25	font-style: italic
				26	}
				27
				28	span.whenRevised {
				29	font-weight: bold;
				30	font-style: normal
				31	}
				32
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	33	div h1 { font-size: 2em; margin: .67em 0 }
				34	div div h1 { font-size: 1.5em; margin: .75em 0 }
				35	div div div h1 { font-size: 1.17em; margin: .83em 0 }
				36	div div div div h1 { margin: 1.12em 0 }
				37
				38	span.term { font-style: italic; font-weight: bold }
				39	</style>
				40
Benjamin Kramer	665a8dc	2012-01-15 15:26:07 +0000	[diff] [blame]	41	<script type="text/javascript">
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	42	/// A little script to recursively build a table of contents.
				43	function buildTOC(div, toc, ancestry) {
				44	var children = div.childNodes;
				45	var len = children.length;
				46
				47	var childNumber = 0;
				48
				49	var list = null;
				50	for (var i = 0; i < len; ++i) {
				51	var child = children[i];
				52	if (child.nodeName != "DIV") continue;
				53	if (child.getAttribute("class") == "rationale") continue;
				54	if (child.id == "toc") continue;
				55
				56	// Okay, we're actually going to build a list node.
				57	if (list === null) list = document.createElement("ul");
				58
				59	var childAncestry = ancestry + ++childNumber + ".";
				60
				61	var headerNode = child.childNodes[1];
				62	var title = headerNode.innerHTML;
				63	headerNode.insertBefore(document.createTextNode(childAncestry + " "),
				64	headerNode.firstChild);
				65
				66	var item = document.createElement("li");
				67	item.appendChild(document.createTextNode(childAncestry + " "));
				68
				69	var anchor = document.createElement("a");
				70	anchor.href = "#" + child.id;
				71	anchor.innerHTML = title;
				72	item.appendChild(anchor);
				73
				74	buildTOC(child, item, childAncestry);
				75
				76	list.appendChild(item);
				77	}
				78	if (list) toc.appendChild(list);
				79	}
				80
				81	function onLoad() {
				82	var toc = document.getElementById("toc");
				83	var content = document.getElementById("content");
				84	buildTOC(content, toc, "");
				85	}
				86	window.onload = onLoad;
				87
				88	</script>
				89	</head>
				90	<body>
				91
				92	<!--#include virtual="../menu.html.incl"-->
				93
				94	<div id="content">
				95	<h1>Automatic Reference Counting</h1>
				96
				97	<div id="toc">
				98	</div>
				99
				100	<div id="meta">
				101	<h1>About this document</h1>
				102
				103	<div id="meta.purpose">
				104	<h1>Purpose</h1>
				105
				106	<p>The first and primary purpose of this document is to serve as a
				107	complete technical specification of Automatic Reference Counting.
				108	Given a core Objective-C compiler and runtime, it should be possible
				109	to write a compiler and runtime which implements these new
				110	semantics.</p>
				111
				112	<p>The secondary purpose is to act as a rationale for why ARC was
				113	designed in this way. This should remain tightly focused on the
				114	technical design and should not stray into marketing speculation.</p>
				115
				116	</div> <!-- meta.purpose -->
				117
				118	<div id="meta.background">
				119	<h1>Background</h1>
				120
				121	<p>This document assumes a basic familiarity with C.</p>
				122
				123	<p><span class="term">Blocks</span> are a C language extension for
				124	creating anonymous functions. Users interact with and transfer block
				125	objects using <span class="term">block pointers</span>, which are
				126	represented like a normal pointer. A block may capture values from
				127	local variables; when this occurs, memory must be dynamically
				128	allocated. The initial allocation is done on the stack, but the
				129	runtime provides a <tt>Block_copy</tt> function which, given a block
				130	pointer, either copies the underlying block object to the heap,
				131	setting its reference count to 1 and returning the new block pointer,
				132	or (if the block object is already on the heap) increases its
				133	reference count by 1. The paired function is <tt>Block_release</tt>,
				134	which decreases the reference count by 1 and destroys the object if
				135	the count reaches zero and is on the heap.</p>
				136
				137	<p>Objective-C is a set of language extensions, significant enough to
				138	be considered a different language. It is a strict superset of C.
				139	The extensions can also be imposed on C++, producing a language called
				140	Objective-C++. The primary feature is a single-inheritance object
				141	system; we briefly describe the modern dialect.</p>
				142
				143	<p>Objective-C defines a new type kind, collectively called
				144	the <span class="term">object pointer types</span>. This kind has two
				145	notable builtin members, <tt>id</tt> and <tt>Class</tt>; <tt>id</tt>
				146	is the final supertype of all object pointers. The validity of
				147	conversions between object pointer types is not checked at runtime.
				148	Users may define <span class="term">classes</span>; each class is a
				149	type, and the pointer to that type is an object pointer type. A class
				150	may have a superclass; its pointer type is a subtype of its
				151	superclass's pointer type. A class has a set
				152	of <span class="term">ivars</span>, fields which appear on all
				153	instances of that class. For every class <i>T</i> there's an
				154	associated metaclass; it has no fields, its superclass is the
				155	metaclass of <i>T</i>'s superclass, and its metaclass is a global
				156	class. Every class has a global object whose class is the
				157	class's metaclass; metaclasses have no associated type, so pointers to
				158	this object have type <tt>Class</tt>.</p>
				159
				160	<p>A class declaration (<tt>@interface</tt>) declares a set
				161	of <span class="term">methods</span>. A method has a return type, a
				162	list of argument types, and a <span class="term">selector</span>: a
				163	name like <tt>foo:bar:baz:</tt>, where the number of colons
				164	corresponds to the number of formal arguments. A method may be an
				165	instance method, in which case it can be invoked on objects of the
				166	class, or a class method, in which case it can be invoked on objects
				167	of the metaclass. A method may be invoked by providing an object
				168	(called the <span class="term">receiver</span>) and a list of formal
				169	arguments interspersed with the selector, like so:</p>
				170
				171	<pre>[receiver foo: fooArg bar: barArg baz: bazArg]</pre>
				172
				173	<p>This looks in the dynamic class of the receiver for a method with
				174	this name, then in that class's superclass, etc., until it finds
				175	something it can execute. The receiver <q>expression</q> may also be
				176	the name of a class, in which case the actual receiver is the class
				177	object for that class, or (within method definitions) it may
				178	be <tt>super</tt>, in which case the lookup algorithm starts with the
				179	static superclass instead of the dynamic class. The actual methods
				180	dynamically found in a class are not those declared in the
				181	<tt>@interface</tt>, but those defined in a separate
				182	<tt>@implementation</tt> declaration; however, when compiling a
				183	call, typechecking is done based on the methods declared in the
				184	<tt>@interface</tt>.</p>
				185
				186	<p>Method declarations may also be grouped into
				187	<span class="term">protocols</span>, which are not inherently
				188	associated with any class, but which classes may claim to follow.
				189	Object pointer types may be qualified with additional protocols that
				190	the object is known to support.</p>
				191
				192	<p><span class="term">Class extensions</span> are collections of ivars
				193	and methods, designed to allow a class's <tt>@interface</tt> to be
				194	split across multiple files; however, there is still a primary
				195	implementation file which must see the <tt>@interface</tt>s of all
				196	class extensions.
				197	<span class="term">Categories</span> allow methods (but not ivars) to
				198	be declared <i>post hoc</i> on an arbitrary class; the methods in the
				199	category's <tt>@implementation</tt> will be dynamically added to that
				200	class's method tables which the category is loaded at runtime,
				201	replacing those methods in case of a collision.</p>
				202
				203	<p>In the standard environment, objects are allocated on the heap, and
				204	their lifetime is manually managed using a reference count. This is
				205	done using two instance methods which all classes are expected to
				206	implement: <tt>retain</tt> increases the object's reference count by
				207	1, whereas <tt>release</tt> decreases it by 1 and calls the instance
				208	method <tt>dealloc</tt> if the count reaches 0. To simplify certain
				209	operations, there is also an <span class="term">autorelease
				210	pool</span>, a thread-local list of objects to call <tt>release</tt>
				211	on later; an object can be added to this pool by
				212	calling <tt>autorelease</tt> on it.</p>
				213
				214	<p>Block pointers may be converted to type <tt>id</tt>; block objects
				215	are laid out in a way that makes them compatible with Objective-C
				216	objects. There is a builtin class that all block objects are
				217	considered to be objects of; this class implements <tt>retain</tt> by
				218	adjusting the reference count, not by calling <tt>Block_copy</tt>.</p>
				219
				220	</div> <!-- meta.background -->
				221
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	222	<div id="meta.evolution">
				223	<h1>Evolution</h1>
				224
				225	<p>ARC is under continual evolution, and this document must be updated
				226	as the language progresses.</p>
				227
				228	<p>If a change increases the expressiveness of the language, for
				229	example by lifting a restriction or by adding new syntax, the change
				230	will be annotated with a revision marker, like so:</p>
				231
				232	<blockquote>
				233	ARC applies to Objective-C pointer types, block pointer types, and
				234	<span class="revision"><span class="whenRevised">[beginning Apple
				235	8.0, LLVM 3.8]</span> BPTRs declared within <code>extern
				236	"BCPL"</code> blocks</span>.
				237	</blockquote>
				238
				239	<p>For now, it is sensible to version this document by the releases of
				240	its sole implementation (and its host project), clang.
				241	<q>LLVM X.Y</q> refers to an open-source release of clang from the
				242	LLVM project. <q>Apple X.Y</q> refers to an Apple-provided release of
				243	the Apple LLVM Compiler. Other organizations that prepare their own,
				244	separately-versioned clang releases and wish to maintain similar
				245	information in this document should send requests to cfe-dev.</p>
				246
				247	<p>If a change decreases the expressiveness of the language, for
				248	example by imposing a new restriction, this should be taken as an
				249	oversight in the original specification and something to be avoided
				250	in all versions. Such changes are generally to be avoided.</p>
				251
				252	</div> <!-- meta.evolution -->
				253
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	254	</div> <!-- meta -->
				255
				256	<div id="general">
				257	<h1>General</h1>
				258
				259	<p>Automatic Reference Counting implements automatic memory management
				260	for Objective-C objects and blocks, freeing the programmer from the
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	261	need to explicitly insert retains and releases. It does not provide a
				262	cycle collector; users must explicitly manage the lifetime of their
				263	objects, breaking cycles manually or with weak or unsafe
				264	references.</p>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	265
				266	<p>ARC may be explicitly enabled with the compiler
				267	flag <tt>-fobjc-arc</tt>. It may also be explicitly disabled with the
				268	compiler flag <tt>-fno-objc-arc</tt>. The last of these two flags
				269	appearing on the compile line <q>wins</q>.</p>
				270
				271	<p>If ARC is enabled, <tt>__has_feature(objc_arc)</tt> will expand to
				272	1 in the preprocessor. For more information about <tt>__has_feature</tt>,
				273	see the <a href="LanguageExtensions.html#__has_feature_extension">language
				274	extensions</a> document.</p>
				275
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	276	</div> <!-- general -->
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	277
				278	<div id="objects">
				279	<h1>Retainable object pointers</h1>
				280
				281	<p>This section describes retainable object pointers, their basic
				282	operations, and the restrictions imposed on their use under ARC. Note
				283	in particular that it covers the rules for pointer <em>values</em>
				284	(patterns of bits indicating the location of a pointed-to object), not
				285	pointer
				286	<em>objects</em> (locations in memory which store pointer values).
				287	The rules for objects are covered in the next section.</p>
				288
				289	<p>A <span class="term">retainable object pointer</span>
				290	(or <q>retainable pointer</q>) is a value of
				291	a <span class="term">retainable object pointer type</span>
				292	(<q>retainable type</q>). There are three kinds of retainable object
				293	pointer types:</p>
				294	<ul>
				295	<li>block pointers (formed by applying the caret (<tt>^</tt>)
				296	declarator sigil to a function type)</li>
				297	<li>Objective-C object pointers (<tt>id</tt>, <tt>Class</tt>, <tt>NSFoo*</tt>, etc.)</li>
				298	<li>typedefs marked with <tt>__attribute__((NSObject))</tt></li>
				299	</ul>
				300
				301	<p>Other pointer types, such as <tt>int*</tt> and <tt>CFStringRef</tt>,
				302	are not subject to ARC's semantics and restrictions.</p>
				303
				304	<div class="rationale">
				305
				306	<p>Rationale: We are not at liberty to require
				307	all code to be recompiled with ARC; therefore, ARC must interoperate
				308	with Objective-C code which manages retains and releases manually. In
				309	general, there are three requirements in order for a
				310	compiler-supported reference-count system to provide reliable
				311	interoperation:</p>
				312
				313	<ul>
				314	<li>The type system must reliably identify which objects are to be
				315	managed. An <tt>int*</tt> might be a pointer to a <tt>malloc</tt>'ed
				316	array, or it might be a interior pointer to such an array, or it might
				317	point to some field or local variable. In contrast, values of the
				318	retainable object pointer types are never interior.</li>
				319	<li>The type system must reliably indicate how to
				320	manage objects of a type. This usually means that the type must imply
				321	a procedure for incrementing and decrementing retain counts.
				322	Supporting single-ownership objects requires a lot more explicit
				323	mediation in the language.</li>
				324	<li>There must be reliable conventions for whether and
				325	when <q>ownership</q> is passed between caller and callee, for both
				326	arguments and return values. Objective-C methods follow such a
				327	convention very reliably, at least for system libraries on Mac OS X,
				328	and functions always pass objects at +0. The C-based APIs for Core
				329	Foundation objects, on the other hand, have much more varied transfer
				330	semantics.</li>
				331	</ul>
				332	</div> <!-- rationale -->
				333
				334	<p>The use of <tt>__attribute__((NSObject))</tt> typedefs is not
				335	recommended. If it's absolutely necessary to use this attribute, be
				336	very explicit about using the typedef, and do not assume that it will
				337	be preserved by language features like <tt>__typeof</tt> and C++
				338	template argument substitution.</p>
				339
				340	<div class="rationale"><p>Rationale: any compiler operation which
				341	incidentally strips type <q>sugar</q> from a type will yield a type
				342	without the attribute, which may result in unexpected
				343	behavior.</p></div>
				344
				345	<div id="objects.retains">
				346	<h1>Retain count semantics</h1>
				347
				348	<p>A retainable object pointer is either a <span class="term">null
				349	pointer</span> or a pointer to a valid object. Furthermore, if it has
				350	block pointer type and is not <tt>null</tt> then it must actually be a
				351	pointer to a block object, and if it has <tt>Class</tt> type (possibly
				352	protocol-qualified) then it must actually be a pointer to a class
				353	object. Otherwise ARC does not enforce the Objective-C type system as
				354	long as the implementing methods follow the signature of the static
				355	type. It is undefined behavior if ARC is exposed to an invalid
				356	pointer.</p>
				357
				358	<p>For ARC's purposes, a valid object is one with <q>well-behaved</q>
				359	retaining operations. Specifically, the object must be laid out such
				360	that the Objective-C message send machinery can successfully send it
				361	the following messages:</p>
				362
				363	<ul>
				364	<li><tt>retain</tt>, taking no arguments and returning a pointer to
				365	the object.</li>
				366	<li><tt>release</tt>, taking no arguments and returning <tt>void</tt>.</li>
				367	<li><tt>autorelease</tt>, taking no arguments and returning a pointer
				368	to the object.</li>
				369	</ul>
				370
				371	<p>The behavior of these methods is constrained in the following ways.
				372	The term <span class="term">high-level semantics</span> is an
				373	intentionally vague term; the intent is that programmers must
				374	implement these methods in a way such that the compiler, modifying
				375	code in ways it deems safe according to these constraints, will not
				376	violate their requirements. For example, if the user puts logging
				377	statements in <tt>retain</tt>, they should not be surprised if those
				378	statements are executed more or less often depending on optimization
				379	settings. These constraints are not exhaustive of the optimization
				380	opportunities: values held in local variables are subject to
				381	additional restrictions, described later in this document.</p>
				382
				383	<p>It is undefined behavior if a computation history featuring a send
				384	of <tt>retain</tt> followed by a send of <tt>release</tt> to the same
				385	object, with no intervening <tt>release</tt> on that object, is not
				386	equivalent under the high-level semantics to a computation
				387	history in which these sends are removed. Note that this implies that
				388	these methods may not raise exceptions.</p>
				389
				390	<p>It is undefined behavior if a computation history features any use
				391	whatsoever of an object following the completion of a send
				392	of <tt>release</tt> that is not preceded by a send of <tt>retain</tt>
				393	to the same object.</p>
				394
				395	<p>The behavior of <tt>autorelease</tt> must be equivalent to sending
				396	<tt>release</tt> when one of the autorelease pools currently in scope
				397	is popped. It may not throw an exception.</p>
				398
				399	<p>When the semantics call for performing one of these operations on a
				400	retainable object pointer, if that pointer is <tt>null</tt> then the
				401	effect is a no-op.</p>
				402
				403	<p>All of the semantics described in this document are subject to
				404	additional <a href="#optimization">optimization rules</a> which permit
				405	the removal or optimization of operations based on local knowledge of
				406	data flow. The semantics describe the high-level behaviors that the
				407	compiler implements, not an exact sequence of operations that a
				408	program will be compiled into.</p>
				409
				410	</div> <!-- objects.retains -->
				411
				412	<div id="objects.operands">
				413	<h1>Retainable object pointers as operands and arguments</h1>
				414
				415	<p>In general, ARC does not perform retain or release operations when
				416	simply using a retainable object pointer as an operand within an
				417	expression. This includes:</p>
				418	<ul>
				419	<li>loading a retainable pointer from an object with non-weak
				420	<a href="#ownership">ownership</a>,</li>
				421	<li>passing a retainable pointer as an argument to a function or
				422	method, and</li>
				423	<li>receiving a retainable pointer as the result of a function or
				424	method call.</li>
				425	</ul>
				426
				427	<div class="rationale"><p>Rationale: while this might seem
				428	uncontroversial, it is actually unsafe when multiple expressions are
				429	evaluated in <q>parallel</q>, as with binary operators and calls,
				430	because (for example) one expression might load from an object while
				431	another writes to it. However, C and C++ already call this undefined
				432	behavior because the evaluations are unsequenced, and ARC simply
				433	exploits that here to avoid needing to retain arguments across a large
				434	number of calls.</p></div>
				435
				436	<p>The remainder of this section describes exceptions to these rules,
				437	how those exceptions are detected, and what those exceptions imply
				438	semantically.</p>
				439
				440	<div id="objects.operands.consumed">
				441	<h1>Consumed parameters</h1>
				442
				443	<p>A function or method parameter of retainable object pointer type
				444	may be marked as <span class="term">consumed</span>, signifying that
				445	the callee expects to take ownership of a +1 retain count. This is
				446	done by adding the <tt>ns_consumed</tt> attribute to the parameter
				447	declaration, like so:</p>
				448
				449	<pre>void foo(__attribute((ns_consumed)) id x);
				450	- (void) foo: (id) __attribute((ns_consumed)) x;</pre>
				451
				452	<p>This attribute is part of the type of the function or method, not
				453	the type of the parameter. It controls only how the argument is
				454	passed and received.</p>
				455
				456	<p>When passing such an argument, ARC retains the argument prior to
				457	making the call.</p>
				458
				459	<p>When receiving such an argument, ARC releases the argument at the
				460	end of the function, subject to the usual optimizations for local
				461	values.</p>
				462
				463	<div class="rationale"><p>Rationale: this formalizes direct transfers
				464	of ownership from a caller to a callee. The most common scenario here
				465	is passing the <tt>self</tt> parameter to <tt>init</tt>, but it is
				466	useful to generalize. Typically, local optimization will remove any
				467	extra retains and releases: on the caller side the retain will be
				468	merged with a +1 source, and on the callee side the release will be
				469	rolled into the initialization of the parameter.</p></div>
				470
				471	<p>The implicit <tt>self</tt> parameter of a method may be marked as
				472	consumed by adding <tt>__attribute__((ns_consumes_self))</tt> to the
John McCall	be16b89	2011-06-18 08:15:19 +0000	[diff] [blame]	473	method declaration. Methods in the <tt>init</tt>
				474	<a href="#family">family</a> are treated as if they were implicitly
				475	marked with this attribute.</p>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	476
John McCall	be16b89	2011-06-18 08:15:19 +0000	[diff] [blame]	477	<p>It is undefined behavior if an Objective-C message send to a method
				478	with <tt>ns_consumed</tt> parameters (other than self) is made with a
				479	null receiver. It is undefined behavior if the method to which an
				480	Objective-C message send statically resolves to has a different set
				481	of <tt>ns_consumed</tt> parameters than the method it dynamically
				482	resolves to. It is undefined behavior if a block or function call is
				483	made through a static type with a different set of <tt>ns_consumed</tt>
				484	parameters than the implementation of the called block or function.</p>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	485
John McCall	be16b89	2011-06-18 08:15:19 +0000	[diff] [blame]	486	<div class="rationale"><p>Rationale: consumed parameters with null
				487	receiver are a guaranteed leak. Mismatches with consumed parameters
				488	will cause over-retains or over-releases, depending on the direction.
				489	The rule about function calls is really just an application of the
				490	existing C/C++ rule about calling functions through an incompatible
				491	function type, but it's useful to state it explicitly.</p></div>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	492
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	493	</div> <!-- objects.operands.consumed -->
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	494
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	495	<div id="objects.operands.retained-returns">
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	496	<h1>Retained return values</h1>
				497
				498	<p>A function or method which returns a retainable object pointer type
				499	may be marked as returning a retained value, signifying that the
				500	caller expects to take ownership of a +1 retain count. This is done
				501	by adding the <tt>ns_returns_retained</tt> attribute to the function or
				502	method declaration, like so:</p>
				503
				504	<pre>id foo(void) __attribute((ns_returns_retained));
				505	- (id) foo __attribute((ns_returns_retained));</pre>
				506
				507	<p>This attribute is part of the type of the function or method.</p>
				508
				509	<p>When returning from such a function or method, ARC retains the
				510	value at the point of evaluation of the return statement, before
				511	leaving all local scopes.</p>
				512
				513	<p>When receiving a return result from such a function or method, ARC
				514	releases the value at the end of the full-expression it is contained
				515	within, subject to the usual optimizations for local values.</p>
				516
				517	<div class="rationale"><p>Rationale: this formalizes direct transfers of
				518	ownership from a callee to a caller. The most common scenario this
				519	models is the retained return from <tt>init</tt>, <tt>alloc</tt>,
				520	<tt>new</tt>, and <tt>copy</tt> methods, but there are other cases in
				521	the frameworks. After optimization there are typically no extra
				522	retains and releases required.</p></div>
				523
				524	<p>Methods in
				525	the <tt>alloc</tt>, <tt>copy</tt>, <tt>init</tt>, <tt>mutableCopy</tt>,
				526	and <tt>new</tt> <a href="#family">families</a> are implicitly marked
				527	<tt>__attribute__((ns_returns_retained))</tt>. This may be suppressed
				528	by explicitly marking the
				529	method <tt>__attribute__((ns_returns_not_retained))</tt>.</p>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	530
John McCall	be16b89	2011-06-18 08:15:19 +0000	[diff] [blame]	531	<p>It is undefined behavior if the method to which an Objective-C
				532	message send statically resolves has different retain semantics on its
				533	result from the method it dynamically resolves to. It is undefined
				534	behavior if a block or function call is made through a static type
				535	with different retain semantics on its result from the implementation
				536	of the called block or function.</p>
				537
				538	<div class="rationale"><p>Rationale: Mismatches with returned results
				539	will cause over-retains or over-releases, depending on the direction.
				540	Again, the rule about function calls is really just an application of
				541	the existing C/C++ rule about calling functions through an
				542	incompatible function type.</p></div>
				543
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	544	</div> <!-- objects.operands.retained-returns -->
John McCall	be16b89	2011-06-18 08:15:19 +0000	[diff] [blame]	545
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	546	<div id="objects.operands.other-returns">
				547	<h1>Unretained return values</h1>
				548
				549	<p>A method or function which returns a retainable object type but
				550	does not return a retained value must ensure that the object is
				551	still valid across the return boundary.</p>
				552
				553	<p>When returning from such a function or method, ARC retains the
				554	value at the point of evaluation of the return statement, then leaves
				555	all local scopes, and then balances out the retain while ensuring that
				556	the value lives across the call boundary. In the worst case, this may
				557	involve an <tt>autorelease</tt>, but callers must not assume that the
				558	value is actually in the autorelease pool.</p>
				559
				560	<p>ARC performs no extra mandatory work on the caller side, although
				561	it may elect to do something to shorten the lifetime of the returned
				562	value.</p>
				563
				564	<div class="rationale"><p>Rationale: it is common in non-ARC code to not
				565	return an autoreleased value; therefore the convention does not force
				566	either path. It is convenient to not be required to do unnecessary
				567	retains and autoreleases; this permits optimizations such as eliding
				568	retain/autoreleases when it can be shown that the original pointer
				569	will still be valid at the point of return.</p></div>
				570
				571	<p>A method or function may be marked
				572	with <tt>__attribute__((ns_returns_autoreleased))</tt> to indicate
				573	that it returns a pointer which is guaranteed to be valid at least as
				574	long as the innermost autorelease pool. There are no additional
				575	semantics enforced in the definition of such a method; it merely
				576	enables optimizations in callers.</p>
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	577
				578	</div> <!-- objects.operands.other-returns -->
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	579
				580	<div id="objects.operands.casts">
				581	<h1>Bridged casts</h1>
				582
				583	<p>A <span class="term">bridged cast</span> is a C-style cast
				584	annotated with one of three keywords:</p>
				585
				586	<ul>
				587	<li><tt>(__bridge T) op</tt> casts the operand to the destination
				588	type <tt>T</tt>. If <tt>T</tt> is a retainable object pointer type,
				589	then <tt>op</tt> must have a non-retainable pointer type.
				590	If <tt>T</tt> is a non-retainable pointer type, then <tt>op</tt> must
				591	have a retainable object pointer type. Otherwise the cast is
				592	ill-formed. There is no transfer of ownership, and ARC inserts
				593	no retain operations.</li>
				594
				595	<li><tt>(__bridge_retained T) op</tt> casts the operand, which must
				596	have retainable object pointer type, to the destination type, which
				597	must be a non-retainable pointer type. ARC retains the value, subject
				598	to the usual optimizations on local values, and the recipient is
				599	responsible for balancing that +1.</li>
				600
				601	<li><tt>(__bridge_transfer T) op</tt> casts the operand, which must
				602	have non-retainable pointer type, to the destination type, which must
				603	be a retainable object pointer type. ARC will release the value at
				604	the end of the enclosing full-expression, subject to the usual
				605	optimizations on local values.</li>
				606	</ul>
				607
				608	<p>These casts are required in order to transfer objects in and out of
				609	ARC control; see the rationale in the section
				610	on <a href="#objects.restrictions.conversion">conversion of retainable
				611	object pointers</a>.</p>
				612
				613	<p>Using a <tt>__bridge_retained</tt> or <tt>__bridge_transfer</tt>
				614	cast purely to convince ARC to emit an unbalanced retain or release,
				615	respectively, is poor form.</p>
				616
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	617	</div> <!-- objects.operands.casts -->
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	618
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	619	</div> <!-- objects.operands -->
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	620
				621	<div id="objects.restrictions">
				622	<h1>Restrictions</h1>
				623
				624	<div id="objects.restrictions.conversion">
				625	<h1>Conversion of retainable object pointers</h1>
				626
				627	<p>In general, a program which attempts to implicitly or explicitly
				628	convert a value of retainable object pointer type to any
				629	non-retainable type, or vice-versa, is ill-formed. For example, an
Fariborz Jahanian	a26b2e5	2011-07-06 21:58:44 +0000	[diff] [blame]	630	Objective-C object pointer shall not be converted to <tt>void*</tt>.
David Blaikie	0ff6495	2011-11-24 00:37:54 +0000	[diff] [blame]	631	As an exception, cast to <tt>intptr_t</tt> is allowed because such
Fariborz Jahanian	a26b2e5	2011-07-06 21:58:44 +0000	[diff] [blame]	632	casts are not transferring ownership. The <a href="#objects.operands.casts">bridged
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	633	casts</a> may be used to perform these conversions where
				634	necessary.</p>
				635
				636	<div class="rationale"><p>Rationale: we cannot ensure the correct
				637	management of the lifetime of objects if they may be freely passed
				638	around as unmanaged types. The bridged casts are provided so that the
				639	programmer may explicitly describe whether the cast transfers control
				640	into or out of ARC.</p></div>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	641
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	642	<p>However, the following exceptions apply.</p>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	643
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	644	</div> <!-- objects.restrictions.conversion -->
				645
				646	<div id="objects.restrictions.conversion-exception-known">
				647	<h1>Conversion to retainable object pointer type of
				648	expressions with known semantics</h1>
				649
				650	<p><span class="revision"><span class="whenRevised">[beginning Apple
				651	4.0, LLVM 3.1]</span> These exceptions have been greatly expanded;
				652	they previously applied only to a much-reduced subset which is
				653	difficult to categorize but which included null pointers, message
				654	sends (under the given rules), and the various global constants.</span></p>
				655
				656	<p>An unbridged conversion to a retainable object pointer type from a
				657	type other than a retainable object pointer type is ill-formed, as
				658	discussed above, unless the operand of the cast has a syntactic form
				659	which is known retained, known unretained, or known
				660	retain-agnostic.</p>
				661
				662	<p>An expression is <span class="term">known retain-agnostic</span> if
				663	it is:</p>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	664	<ul>
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	665	<li>an Objective-C string literal,</li>
				666	<li>a load from a <tt>const</tt> system global variable of
				667	<a href="#misc.c-retainable">C retainable pointer type</a>, or</li>
				668	<li>a null pointer constant.</li>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	669	</ul>
				670
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	671	<p>An expression is <span class="term">known unretained</span> if it
				672	is an rvalue of <a href="#misc.c-retainable">C retainable
				673	pointer type</a> and it is:</p>
				674	<ul>
				675	<li>a direct call to a function, and either that function has the
				676	<tt>cf_returns_not_retained</tt> attribute or it is an
				677	<a href="#misc.c-retainable.audit">audited</a> function that does not
				678	have the <tt>cf_returns_retained</tt> attribute and does not follow
				679	the create/copy naming convention,</li>
				680	<li>a message send, and the declared method either has
				681	the <tt>cf_returns_not_retained</tt> attribute or it has neither
				682	the <tt>cf_returns_retained</tt> attribute nor a
				683	<a href="#family">selector family</a> that implies a retained
				684	result.</li>
				685	</ul>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	686
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	687	<p>An expression is <span class="term">known retained</span> if it is
				688	an rvalue of <a href="#misc.c-retainable">C retainable pointer type</a>
				689	and it is:</p>
				690	<ul>
				691	<li>a message send, and the declared method either has the
				692	<tt>cf_returns_retained</tt> attribute, or it does not have
				693	the <tt>cf_returns_not_retained</tt> attribute but it does have a
				694	<a href="#family">selector family</a> that implies a retained
				695	result.</li>
				696	</ul>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	697
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	698	<p>Furthermore:</p>
				699	<ul>
				700	<li>a comma expression is classified according to its right-hand side,</li>
				701	<li>a statement expression is classified according to its result
				702	expression, if it has one,</li>
				703	<li>an lvalue-to-rvalue conversion applied to an Objective-C property
				704	lvalue is classified according to the underlying message send, and</li>
				705	<li>a conditional operator is classified according to its second and
				706	third operands, if they agree in classification, or else the other
				707	if one is known retain-agnostic.</li>
				708	</ul>
				709
				710	<p>If the cast operand is known retained, the conversion is treated as
				711	a <tt>__bridge_transfer</tt> cast. If the cast operand is known
				712	unretained or known retain-agnostic, the conversion is treated as
				713	a <tt>__bridge</tt> cast.</p>
				714
				715	<div class="rationale"><p>Rationale: Bridging casts are annoying.
				716	Absent the ability to completely automate the management of CF
				717	objects, however, we are left with relatively poor attempts to reduce
				718	the need for a glut of explicit bridges. Hence these rules.</p>
				719
				720	<p>We've so far consciously refrained from implicitly turning retained
				721	CF results from function calls into <tt>__bridge_transfer</tt> casts.
				722	The worry is that some code patterns — for example, creating a
				723	CF value, assigning it to an ObjC-typed local, and then
				724	calling <tt>CFRelease</tt> when done — are a bit too likely to
				725	be accidentally accepted, leading to mysterious behavior.</p></div>
				726
				727	</div> <!-- objects.restrictions.conversion-exception-known -->
				728
				729	<div id="objects.restrictions.conversion-exception-contextual">
				730	<h1>Conversion from retainable object pointer type in certain contexts</h1>
				731
				732	<p><span class="revision"><span class="whenRevised">[beginning Apple
				733	4.0, LLVM 3.1]</span></span></p>
				734
				735	<p>If an expression of retainable object pointer type is explicitly
				736	cast to a <a href="#misc.c-retainable">C retainable pointer type</a>,
				737	the program is ill-formed as discussed above unless the result is
				738	immediately used:</p>
				739
				740	<ul>
				741	<li>to initialize a parameter in an Objective-C message send where the
				742	parameter is not marked with the <tt>cf_consumed</tt> attribute, or</li>
				743	<li>to initialize a parameter in a direct call to
				744	an <a href="#misc.c-retainable.audit">audited</a> function where the
				745	parameter is not marked with the <tt>cf_consumed</tt> attribute.</li>
				746	</ul>
				747
				748	<div class="rationale"><p>Rationale: Consumed parameters are left out
				749	because ARC would naturally balance them with a retain, which was
				750	judged too treacherous. This is in part because several of the most
				751	common consuming functions are in the <tt>Release</tt> family, and it
				752	would be quite unfortunate for explicit releases to be silently
				753	balanced out in this way.</p></div>
				754
				755	</div> <!-- objects.restrictions.conversion-exception-contextual -->
				756
				757	</div> <!-- objects.restrictions -->
				758
				759	</div> <!-- objects -->
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	760
				761	<div id="ownership">
				762	<h1>Ownership qualification</h1>
				763
				764	<p>This section describes the behavior of <em>objects</em> of
				765	retainable object pointer type; that is, locations in memory which
				766	store retainable object pointers.</p>
				767
				768	<p>A type is a <span class="term">retainable object owner type</span>
				769	if it is a retainable object pointer type or an array type whose
				770	element type is a retainable object owner type.</p>
				771
				772	<p>An <span class="term">ownership qualifier</span> is a type
Douglas Gregor	4020cae	2011-06-17 23:16:24 +0000	[diff] [blame]	773	qualifier which applies only to retainable object owner types. An array type is
				774	ownership-qualified according to its element type, and adding an ownership
				775	qualifier to an array type so qualifies its element type.</p>
				776
				777	<p>A program is ill-formed if it attempts to apply an ownership qualifier
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	778	to a type which is already ownership-qualified, even if it is the same
Douglas Gregor	4020cae	2011-06-17 23:16:24 +0000	[diff] [blame]	779	qualifier. There is a single exception to this rule: an ownership qualifier
				780	may be applied to a substituted template type parameter, which overrides the
				781	ownership qualifier provided by the template argument.</p>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	782
				783	<p>Except as described under
				784	the <a href="#ownership.inference">inference rules</a>, a program is
				785	ill-formed if it attempts to form a pointer or reference type to a
				786	retainable object owner type which lacks an ownership qualifier.</p>
				787
				788	<div class="rationale"><p>Rationale: these rules, together with the
				789	inference rules, ensure that all objects and lvalues of retainable
Douglas Gregor	4020cae	2011-06-17 23:16:24 +0000	[diff] [blame]	790	object pointer type have an ownership qualifier. The ability to override an ownership qualifier during template substitution is required to counteract the <a href="#ownership.inference.template_arguments">inference of <tt>__strong</tt> for template type arguments</a>. </p></div>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	791
				792	<p>There are four ownership qualifiers:</p>
				793
				794	<ul>
				795	<li><tt>__autoreleasing</tt></li>
				796	<li><tt>__strong</tt></li>
				797	<li><tt>__unsafe_unretained</tt></li>
				798	<li><tt>__weak</tt></li>
				799	</ul>
				800
				801	<p>A type is <span class="term">nontrivially ownership-qualified</span>
				802	if it is qualified with <tt>__autoreleasing</tt>, <tt>__strong</tt>, or
				803	<tt>__weak</tt>.</p>
				804
				805	<div id="ownership.spelling">
				806	<h1>Spelling</h1>
				807
				808	<p>The names of the ownership qualifiers are reserved for the
				809	implementation. A program may not assume that they are or are not
				810	implemented with macros, or what those macros expand to.</p>
				811
				812	<p>An ownership qualifier may be written anywhere that any other type
				813	qualifier may be written.</p>
				814
				815	<p>If an ownership qualifier appears in
				816	the <i>declaration-specifiers</i>, the following rules apply:</p>
				817
				818	<ul>
				819	<li>if the type specifier is a retainable object owner type, the
				820	qualifier applies to that type;</li>
				821	<li>if the outermost non-array part of the declarator is a pointer or
				822	block pointer, the qualifier applies to that type;</li>
				823	<li>otherwise the program is ill-formed.</li>
				824	</ul>
				825
				826	<p>If an ownership qualifier appears on the declarator name, or on the
				827	declared object, it is applied to outermost pointer or block-pointer
				828	type.</p>
				829
				830	<p>If an ownership qualifier appears anywhere else in a declarator, it
				831	applies to the type there.</p>
				832
John McCall	4f26495	2011-07-13 23:15:32 +0000	[diff] [blame]	833	<div id="ownership.spelling.property">
				834	<h1>Property declarations</h1>
				835
				836	<p>A property of retainable object pointer type may have ownership.
				837	If the property's type is ownership-qualified, then the property has
				838	that ownership. If the property has one of the following modifiers,
				839	then the property has the corresponding ownership. A property is
				840	ill-formed if it has conflicting sources of ownership, or if it has
				841	redundant ownership modifiers, or if it has <tt>__autoreleasing</tt>
				842	ownership.</p>
				843
				844	<ul>
				845	<li><tt>assign</tt> implies <tt>__unsafe_unretained</tt> ownership.</li>
				846	<li><tt>copy</tt> implies <tt>__strong</tt> ownership, as well as the
				847	usual behavior of copy semantics on the setter.</li>
				848	<li><tt>retain</tt> implies <tt>__strong</tt> ownership.</li>
				849	<li><tt>strong</tt> implies <tt>__strong</tt> ownership.</li>
				850	<li><tt>unsafe_unretained</tt> implies <tt>__unsafe_unretained</tt>
				851	ownership.</li>
				852	<li><tt>weak</tt> implies <tt>__weak</tt> ownership.</li>
				853	</ul>
				854
				855	<p>With the exception of <tt>weak</tt>, these modifiers are available
				856	in non-ARC modes.</p>
				857
				858	<p>A property's specified ownership is preserved in its metadata, but
				859	otherwise the meaning is purely conventional unless the property is
				860	synthesized. If a property is synthesized, then the
				861	<span class="term">associated instance variable</span> is the
				862	instance variable which is named, possibly implicitly, by the
				863	<tt>@synthesize</tt> declaration. If the associated instance variable
				864	already exists, then its ownership qualification must equal the
				865	ownership of the property; otherwise, the instance variable is created
				866	with that ownership qualification.</p>
				867
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	868	<p>A property of retainable object pointer type which is synthesized
				869	without a source of ownership has the ownership of its associated
				870	instance variable, if it already exists; otherwise,
				871	<span class="revision"><span class="whenRevised">[beginning Apple 3.1,
				872	LLVM 3.1]</span> its ownership is implicitly <tt>strong</tt></span>.
				873	Prior to this revision, it was ill-formed to synthesize such a
				874	property.</p>
				875
				876	<div class="rationale"><p>Rationale: using <tt>strong</tt> by default
				877	is safe and consistent with the generic ARC rule about
				878	<a href="#ownership.inference.variables">inferring ownership</a>. It
				879	is, unfortunately, inconsistent with the non-ARC rule which states
				880	that such properties are implicitly <tt>assign</tt>. However, that
				881	rule is clearly untenable in ARC, since it leads to default-unsafe
				882	code. The main merit to banning the properties is to avoid confusion
				883	with non-ARC practice, which did not ultimately strike us as
				884	sufficient to justify requiring extra syntax and (more importantly)
				885	forcing novices to understand ownership rules just to declare a
				886	property when the default is so reasonable. Changing the rule away
				887	from non-ARC practice was acceptable because we had conservatively
				888	banned the synthesis in order to give ourselves exactly this
				889	leeway.</p></div>
				890
John McCall	4f26495	2011-07-13 23:15:32 +0000	[diff] [blame]	891	</div> <!-- ownership.spelling.property -->
				892
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	893	</div> <!-- ownership.spelling -->
				894
				895	<div id="ownership.semantics">
				896	<h1>Semantics</h1>
				897
				898	<p>There are five <span class="term">managed operations</span> which
				899	may be performed on an object of retainable object pointer type. Each
				900	qualifier specifies different semantics for each of these operations.
				901	It is still undefined behavior to access an object outside of its
				902	lifetime.</p>
				903
				904	<p>A load or store with <q>primitive semantics</q> has the same
				905	semantics as the respective operation would have on an <tt>void*</tt>
				906	lvalue with the same alignment and non-ownership qualification.</p>
				907
				908	<p><span class="term">Reading</span> occurs when performing a
Benjamin Kramer	665a8dc	2012-01-15 15:26:07 +0000	[diff] [blame]	909	lvalue-to-rvalue conversion on an object lvalue.</p>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	910
				911	<ul>
				912	<li>For <tt>__weak</tt> objects, the current pointee is retained and
				913	then released at the end of the current full-expression. This must
				914	execute atomically with respect to assignments and to the final
				915	release of the pointee.</li>
				916	<li>For all other objects, the lvalue is loaded with primitive
				917	semantics.</li>
				918	</ul>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	919
				920	<p><span class="term">Assignment</span> occurs when evaluating
Benjamin Kramer	665a8dc	2012-01-15 15:26:07 +0000	[diff] [blame]	921	an assignment operator. The semantics vary based on the qualification:</p>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	922	<ul>
				923	<li>For <tt>__strong</tt> objects, the new pointee is first retained;
				924	second, the lvalue is loaded with primitive semantics; third, the new
				925	pointee is stored into the lvalue with primitive semantics; and
				926	finally, the old pointee is released. This is not performed
				927	atomically; external synchronization must be used to make this safe in
				928	the face of concurrent loads and stores.</li>
				929	<li>For <tt>__weak</tt> objects, the lvalue is updated to point to the
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	930	new pointee, unless the new pointee is an object currently undergoing
				931	deallocation, in which case the lvalue is updated to a null pointer.
				932	This must execute atomically with respect to other assignments to the
				933	object, to reads from the object, and to the final release of the new
				934	pointee.</li>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	935	<li>For <tt>__unsafe_unretained</tt> objects, the new pointee is
				936	stored into the lvalue using primitive semantics.</li>
				937	<li>For <tt>__autoreleasing</tt> objects, the new pointee is retained,
				938	autoreleased, and stored into the lvalue using primitive semantics.</li>
				939	</ul>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	940
				941	<p><span class="term">Initialization</span> occurs when an object's
				942	lifetime begins, which depends on its storage duration.
Benjamin Kramer	665a8dc	2012-01-15 15:26:07 +0000	[diff] [blame]	943	Initialization proceeds in two stages:</p>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	944	<ol>
				945	<li>First, a null pointer is stored into the lvalue using primitive
				946	semantics. This step is skipped if the object
				947	is <tt>__unsafe_unretained</tt>.</li>
				948	<li>Second, if the object has an initializer, that expression is
				949	evaluated and then assigned into the object using the usual assignment
				950	semantics.</li>
				951	</ol>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	952
				953	<p><span class="term">Destruction</span> occurs when an object's
				954	lifetime ends. In all cases it is semantically equivalent to
				955	assigning a null pointer to the object, with the proviso that of
				956	course the object cannot be legally read after the object's lifetime
				957	ends.</p>
				958
				959	<p><span class="term">Moving</span> occurs in specific situations
				960	where an lvalue is <q>moved from</q>, meaning that its current pointee
				961	will be used but the object may be left in a different (but still
				962	valid) state. This arises with <tt>__block</tt> variables and rvalue
				963	references in C++. For <tt>__strong</tt> lvalues, moving is equivalent
				964	to loading the lvalue with primitive semantics, writing a null pointer
				965	to it with primitive semantics, and then releasing the result of the
				966	load at the end of the current full-expression. For all other
				967	lvalues, moving is equivalent to reading the object.</p>
				968
				969	</div> <!-- ownership.semantics -->
				970
				971	<div id="ownership.restrictions">
				972	<h1>Restrictions</h1>
				973
John McCall	2fad783	2011-07-07 00:03:42 +0000	[diff] [blame]	974	<div id="ownership.restrictions.weak">
				975	<h1>Weak-unavailable types</h1>
				976
				977	<p>It is explicitly permitted for Objective-C classes to not
				978	support <tt>__weak</tt> references. It is undefined behavior to
				979	perform an operation with weak assignment semantics with a pointer to
				980	an Objective-C object whose class does not support <tt>__weak</tt>
				981	references.</p>
				982
				983	<div class="rationale"><p>Rationale: historically, it has been
				984	possible for a class to provide its own reference-count implementation
				985	by overriding <tt>retain</tt>, <tt>release</tt>, etc. However, weak
				986	references to an object require coordination with its class's
				987	reference-count implementation because, among other things, weak loads
				988	and stores must be atomic with respect to the final release.
				989	Therefore, existing custom reference-count implementations will
				990	generally not support weak references without additional effort. This
				991	is unavoidable without breaking binary compatibility.</p></div>
				992
				993	<p>A class may indicate that it does not support weak references by
				994	providing the <tt>objc_arc_weak_unavailable</tt> attribute on the
				995	class's interface declaration. A retainable object pointer type
				996	is <span class="term">weak-unavailable</span> if is a pointer to an
				997	(optionally protocol-qualified) Objective-C class <tt>T</tt>
				998	where <tt>T</tt> or one of its superclasses has
				999	the <tt>objc_arc_weak_unavailable</tt> attribute. A program is
				1000	ill-formed if it applies the <tt>__weak</tt> ownership qualifier to a
				1001	weak-unavailable type or if the value operand of a weak assignment
				1002	operation has a weak-unavailable type.</p>
				1003	</div> <!-- ownership.restrictions.weak -->
				1004
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1005	<div id="ownership.restrictions.autoreleasing">
John McCall	2fad783	2011-07-07 00:03:42 +0000	[diff] [blame]	1006	<h1>Storage duration of <tt>__autoreleasing</tt> objects</h1>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1007
				1008	<p>A program is ill-formed if it declares an <tt>__autoreleasing</tt>
				1009	object of non-automatic storage duration.</p>
				1010
				1011	<div class="rationale"><p>Rationale: autorelease pools are tied to the
				1012	current thread and scope by their nature. While it is possible to
				1013	have temporary objects whose instance variables are filled with
				1014	autoreleased objects, there is no way that ARC can provide any sort of
				1015	safety guarantee there.</p></div>
				1016
				1017	<p>It is undefined behavior if a non-null pointer is assigned to
				1018	an <tt>__autoreleasing</tt> object while an autorelease pool is in
				1019	scope and then that object is read after the autorelease pool's scope
				1020	is left.</p>
				1021
				1022	</div>
				1023
				1024	<div id="ownership.restrictions.conversion.indirect">
				1025	<h1>Conversion of pointers to ownership-qualified types</h1>
				1026
				1027	<p>A program is ill-formed if an expression of type <tt>T*</tt> is
				1028	converted, explicitly or implicitly, to the type <tt>U*</tt>,
				1029	where <tt>T</tt> and <tt>U</tt> have different ownership
Benjamin Kramer	665a8dc	2012-01-15 15:26:07 +0000	[diff] [blame]	1030	qualification, unless:</p>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1031	<ul>
				1032	<li><tt>T</tt> is qualified with <tt>__strong</tt>,
				1033	<tt>__autoreleasing</tt>, or <tt>__unsafe_unretained</tt>, and
				1034	<tt>U</tt> is qualified with both <tt>const</tt> and
				1035	<tt>__unsafe_unretained</tt>; or</li>
				1036	<li>either <tt>T</tt> or <tt>U</tt> is <tt>cv void</tt>, where
				1037	<tt>cv</tt> is an optional sequence of non-ownership qualifiers; or</li>
				1038	<li>the conversion is requested with a <tt>reinterpret_cast</tt> in
				1039	Objective-C++; or</li>
				1040	<li>the conversion is a
				1041	well-formed <a href="#ownership.restrictions.pass_by_writeback">pass-by-writeback</a>.</li>
				1042	</ul>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1043
Benjamin Kramer	665a8dc	2012-01-15 15:26:07 +0000	[diff] [blame]	1044	<p>The analogous rule applies to <tt>T&</tt> and <tt>U&</tt> in
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1045	Objective-C++.</p>
				1046
				1047	<div class="rationale"><p>Rationale: these rules provide a reasonable
				1048	level of type-safety for indirect pointers, as long as the underlying
				1049	memory is not deallocated. The conversion to <tt>const
				1050	__unsafe_unretained</tt> is permitted because the semantics of reads
				1051	are equivalent across all these ownership semantics, and that's a very
				1052	useful and common pattern. The interconversion with <tt>void*</tt> is
				1053	useful for allocating memory or otherwise escaping the type system,
				1054	but use it carefully. <tt>reinterpret_cast</tt> is considered to be
				1055	an obvious enough sign of taking responsibility for any
				1056	problems.</p></div>
				1057
				1058	<p>It is undefined behavior to access an ownership-qualified object
				1059	through an lvalue of a differently-qualified type, except that any
				1060	non-<tt>__weak</tt> object may be read through
				1061	an <tt>__unsafe_unretained</tt> lvalue.</p>
				1062
				1063	<p>It is undefined behavior if a managed operation is performed on
				1064	a <tt>__strong</tt> or <tt>__weak</tt> object without a guarantee that
				1065	it contains a primitive zero bit-pattern, or if the storage for such
				1066	an object is freed or reused without the object being first assigned a
				1067	null pointer.</p>
				1068
				1069	<div class="rationale"><p>Rationale: ARC cannot differentiate between
				1070	an assignment operator which is intended to <q>initialize</q> dynamic
				1071	memory and one which is intended to potentially replace a value.
				1072	Therefore the object's pointer must be valid before letting ARC at it.
				1073	Similarly, C and Objective-C do not provide any language hooks for
				1074	destroying objects held in dynamic memory, so it is the programmer's
				1075	responsibility to avoid leaks (<tt>__strong</tt> objects) and
				1076	consistency errors (<tt>__weak</tt> objects).</p>
				1077
				1078	<p>These requirements are followed automatically in Objective-C++ when
				1079	creating objects of retainable object owner type with <tt>new</tt>
				1080	or <tt>new[]</tt> and destroying them with <tt>delete</tt>,
				1081	<tt>delete[]</tt>, or a pseudo-destructor expression. Note that
				1082	arrays of nontrivially-ownership-qualified type are not ABI compatible
				1083	with non-ARC code because the element type is non-POD: such arrays
				1084	that are <tt>new[]</tt>'d in ARC translation units cannot
				1085	be <tt>delete[]</tt>'d in non-ARC translation units and
				1086	vice-versa.</p></div>
				1087
				1088	</div>
				1089
				1090	<div id="ownership.restrictions.pass_by_writeback">
				1091	<h1>Passing to an out parameter by writeback</h1>
				1092
				1093	<p>If the argument passed to a parameter of type
				1094	<tt>T __autoreleasing </tt> has type <tt>U oq </tt>,
				1095	where <tt>oq</tt> is an ownership qualifier, then the argument is a
				1096	candidate for <span class="term">pass-by-writeback</span> if:</p>
				1097
				1098	<ul>
Benjamin Kramer	665a8dc	2012-01-15 15:26:07 +0000	[diff] [blame]	1099	<li><tt>oq</tt> is <tt>__strong</tt> or <tt>__weak</tt>, and</li>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1100	<li>it would be legal to initialize a <tt>T __strong *</tt> with
				1101	a <tt>U __strong *</tt>.</li>
				1102	</ul>
				1103
				1104	<p>For purposes of overload resolution, an implicit conversion
				1105	sequence requiring a pass-by-writeback is always worse than an
				1106	implicit conversion sequence not requiring a pass-by-writeback.</p>
				1107
				1108	<p>The pass-by-writeback is ill-formed if the argument expression does
				1109	not have a legal form:</p>
				1110
				1111	<ul>
Benjamin Kramer	665a8dc	2012-01-15 15:26:07 +0000	[diff] [blame]	1112	<li><tt>&var</tt>, where <tt>var</tt> is a scalar variable of
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1113	automatic storage duration with retainable object pointer type</li>
				1114	<li>a conditional expression where the second and third operands are
				1115	both legal forms</li>
				1116	<li>a cast whose operand is a legal form</li>
				1117	<li>a null pointer constant</li>
				1118	</ul>
				1119
				1120	<div class="rationale"><p>Rationale: the restriction in the form of
				1121	the argument serves two purposes. First, it makes it impossible to
				1122	pass the address of an array to the argument, which serves to protect
				1123	against an otherwise serious risk of mis-inferring an <q>array</q>
				1124	argument as an out-parameter. Second, it makes it much less likely
				1125	that the user will see confusing aliasing problems due to the
				1126	implementation, below, where their store to the writeback temporary is
				1127	not immediately seen in the original argument variable.</p></div>
				1128
Benjamin Kramer	665a8dc	2012-01-15 15:26:07 +0000	[diff] [blame]	1129	<p>A pass-by-writeback is evaluated as follows:</p>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1130	<ol>
				1131	<li>The argument is evaluated to yield a pointer <tt>p</tt> of
				1132	type <tt>U oq *</tt>.</li>
				1133	<li>If <tt>p</tt> is a null pointer, then a null pointer is passed as
				1134	the argument, and no further work is required for the pass-by-writeback.</li>
				1135	<li>Otherwise, a temporary of type <tt>T __autoreleasing</tt> is
				1136	created and initialized to a null pointer.</li>
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	1137	<li>If the parameter is not an Objective-C method parameter marked
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1138	<tt>out</tt>, then <tt>*p</tt> is read, and the result is written
				1139	into the temporary with primitive semantics.</li>
				1140	<li>The address of the temporary is passed as the argument to the
				1141	actual call.</li>
				1142	<li>After the call completes, the temporary is loaded with primitive
				1143	semantics, and that value is assigned into <tt>*p</tt>.</li>
Benjamin Kramer	665a8dc	2012-01-15 15:26:07 +0000	[diff] [blame]	1144	</ol>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1145
				1146	<div class="rationale"><p>Rationale: this is all admittedly
				1147	convoluted. In an ideal world, we would see that a local variable is
				1148	being passed to an out-parameter and retroactively modify its type to
				1149	be <tt>__autoreleasing</tt> rather than <tt>__strong</tt>. This would
				1150	be remarkably difficult and not always well-founded under the C type
				1151	system. However, it was judged unacceptably invasive to require
				1152	programmers to write <tt>__autoreleasing</tt> on all the variables
				1153	they intend to use for out-parameters. This was the least bad
				1154	solution.</p></div>
				1155
				1156	</div>
				1157
				1158	<div id="ownership.restrictions.records">
				1159	<h1>Ownership-qualified fields of structs and unions</h1>
				1160
				1161	<p>A program is ill-formed if it declares a member of a C struct or
				1162	union to have a nontrivially ownership-qualified type.</p>
				1163
				1164	<div class="rationale"><p>Rationale: the resulting type would be
				1165	non-POD in the C++ sense, but C does not give us very good language
				1166	tools for managing the lifetime of aggregates, so it is more
				1167	convenient to simply forbid them. It is still possible to manage this
				1168	with a <tt>void*</tt> or an <tt>__unsafe_unretained</tt>
				1169	object.</p></div>
				1170
				1171	<p>This restriction does not apply in Objective-C++. However,
David Blaikie	5090e9f	2011-10-18 05:49:30 +0000	[diff] [blame]	1172	nontrivally ownership-qualified types are considered non-POD: in C++11
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1173	terms, they are not trivially default constructible, copy
				1174	constructible, move constructible, copy assignable, move assignable,
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	1175	or destructible. It is a violation of C++'s One Definition Rule to use
				1176	a class outside of ARC that, under ARC, would have a nontrivially
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1177	ownership-qualified member.</p>
				1178
				1179	<div class="rationale"><p>Rationale: unlike in C, we can express all
				1180	the necessary ARC semantics for ownership-qualified subobjects as
				1181	suboperations of the (default) special member functions for the class.
				1182	These functions then become non-trivial. This has the non-obvious
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	1183	result that the class will have a non-trivial copy constructor and
				1184	non-trivial destructor; if this would not normally be true outside of
				1185	ARC, objects of the type will be passed and returned in an
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1186	ABI-incompatible manner.</p></div>
				1187
				1188	</div>
				1189
				1190	</div>
				1191
				1192	<div id="ownership.inference">
				1193	<h1>Ownership inference</h1>
				1194
				1195	<div id="ownership.inference.variables">
				1196	<h1>Objects</h1>
				1197
				1198	<p>If an object is declared with retainable object owner type, but
				1199	without an explicit ownership qualifier, its type is implicitly
				1200	adjusted to have <tt>__strong</tt> qualification.</p>
				1201
				1202	<p>As a special case, if the object's base type is <tt>Class</tt>
				1203	(possibly protocol-qualified), the type is adjusted to
				1204	have <tt>__unsafe_unretained</tt> qualification instead.</p>
				1205
				1206	</div>
				1207
				1208	<div id="ownership.inference.indirect_parameters">
				1209	<h1>Indirect parameters</h1>
				1210
				1211	<p>If a function or method parameter has type <tt>T*</tt>, where
				1212	<tt>T</tt> is an ownership-unqualified retainable object pointer type,
				1213	then:</p>
				1214
				1215	<ul>
				1216	<li>if <tt>T</tt> is <tt>const</tt>-qualified or <tt>Class</tt>, then
				1217	it is implicitly qualified with <tt>__unsafe_unretained</tt>;</li>
				1218	<li>otherwise, it is implicitly qualified
				1219	with <tt>__autoreleasing</tt>.</li>
				1220	</ul>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1221
				1222	<div class="rationale"><p>Rationale: <tt>__autoreleasing</tt> exists
				1223	mostly for this case, the Cocoa convention for out-parameters. Since
				1224	a pointer to <tt>const</tt> is obviously not an out-parameter, we
				1225	instead use a type more useful for passing arrays. If the user
				1226	instead intends to pass in a <em>mutable</em> array, inferring
				1227	<tt>__autoreleasing</tt> is the wrong thing to do; this directs some
				1228	of the caution in the following rules about writeback.</p></div>
				1229
				1230	<p>Such a type written anywhere else would be ill-formed by the
				1231	general rule requiring ownership qualifiers.</p>
				1232
				1233	<p>This rule does not apply in Objective-C++ if a parameter's type is
				1234	dependent in a template pattern and is only <em>instantiated</em> to
				1235	a type which would be a pointer to an unqualified retainable object
				1236	pointer type. Such code is still ill-formed.</p>
				1237
				1238	<div class="rationale"><p>Rationale: the convention is very unlikely
				1239	to be intentional in template code.</p></div>
				1240
				1241	</div> <!-- ownership.inference.indirect_parameters -->
Douglas Gregor	e559ca1	2011-06-17 22:11:49 +0000	[diff] [blame]	1242
				1243	<div id="ownership.inference.template_arguments">
				1244	<h1>Template arguments</h1>
				1245
				1246	<p>If a template argument for a template type parameter is an
				1247	retainable object owner type that does not have an explicit ownership
				1248	qualifier, it is adjusted to have <tt>__strong</tt>
Douglas Gregor	54fb28a	2011-06-17 22:19:27 +0000	[diff] [blame]	1249	qualification. This adjustment occurs regardless of whether the
Douglas Gregor	e559ca1	2011-06-17 22:11:49 +0000	[diff] [blame]	1250	template argument was deduced or explicitly specified. </p>
				1251
				1252	<div class="rationale"><p>Rationale: <tt>__strong</tt> is a useful default for containers (e.g., <tt>std::vector<id></tt>), which would otherwise require explicit qualification. Moreover, unqualified retainable object pointer types are unlikely to be useful within templates, since they generally need to have a qualifier applied to the before being used.</p></div>
				1253
				1254	</div> <!-- ownership.inference.template_arguments -->
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1255	</div> <!-- ownership.inference -->
				1256	</div> <!-- ownership -->
				1257
Douglas Gregor	e559ca1	2011-06-17 22:11:49 +0000	[diff] [blame]	1258
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1259	<div id="family">
				1260	<h1>Method families</h1>
				1261
				1262	<p>An Objective-C method may fall into a <span class="term">method
				1263	family</span>, which is a conventional set of behaviors ascribed to it
				1264	by the Cocoa conventions.</p>
				1265
Benjamin Kramer	665a8dc	2012-01-15 15:26:07 +0000	[diff] [blame]	1266	<p>A method is in a certain method family if:</p>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1267	<ul>
				1268	<li>it has a <tt>objc_method_family</tt> attribute placing it in that
				1269	family; or if not that,</li>
				1270	<li>it does not have an <tt>objc_method_family</tt> attribute placing
				1271	it in a different or no family, and</li>
				1272	<li>its selector falls into the corresponding selector family, and</li>
				1273	<li>its signature obeys the added restrictions of the method family.</li>
Benjamin Kramer	665a8dc	2012-01-15 15:26:07 +0000	[diff] [blame]	1274	</ul>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1275
				1276	<p>A selector is in a certain selector family if, ignoring any leading
				1277	underscores, the first component of the selector either consists
				1278	entirely of the name of the method family or it begins with that name
				1279	followed by a character other than a lowercase letter. For
				1280	example, <tt>_perform:with:</tt> and <tt>performWith:</tt> would fall
				1281	into the <tt>perform</tt> family (if we recognized one),
				1282	but <tt>performing:with</tt> would not.</p>
				1283
				1284	<p>The families and their added restrictions are:</p>
				1285
				1286	<ul>
				1287	<li><tt>alloc</tt> methods must return a retainable object pointer type.</li>
				1288	<li><tt>copy</tt> methods must return a retainable object pointer type.</li>
				1289	<li><tt>mutableCopy</tt> methods must return a retainable object pointer type.</li>
				1290	<li><tt>new</tt> methods must return a retainable object pointer type.</li>
				1291	<li><tt>init</tt> methods must be instance methods and must return an
				1292	Objective-C pointer type. Additionally, a program is ill-formed if it
				1293	declares or contains a call to an <tt>init</tt> method whose return
				1294	type is neither <tt>id</tt> nor a pointer to a super-class or
John McCall	be16b89	2011-06-18 08:15:19 +0000	[diff] [blame]	1295	sub-class of the declaring class (if the method was declared on
				1296	a class) or the static receiver type of the call (if it was declared
Benjamin Kramer	665a8dc	2012-01-15 15:26:07 +0000	[diff] [blame]	1297	on a protocol).
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1298
				1299	<div class="rationale"><p>Rationale: there are a fair number of existing
				1300	methods with <tt>init</tt>-like selectors which nonetheless don't
				1301	follow the <tt>init</tt> conventions. Typically these are either
				1302	accidental naming collisions or helper methods called during
				1303	initialization. Because of the peculiar retain/release behavior
				1304	of <tt>init</tt> methods, it's very important not to treat these
				1305	methods as <tt>init</tt> methods if they aren't meant to be. It was
				1306	felt that implicitly defining these methods out of the family based on
				1307	the exact relationship between the return type and the declaring class
John McCall	be16b89	2011-06-18 08:15:19 +0000	[diff] [blame]	1308	would be much too subtle and fragile. Therefore we identify a small
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1309	number of legitimate-seeming return types and call everything else an
				1310	error. This serves the secondary purpose of encouraging programmers
John McCall	be16b89	2011-06-18 08:15:19 +0000	[diff] [blame]	1311	not to accidentally give methods names in the <tt>init</tt> family.</p>
				1312
				1313	<p>Note that a method with an <tt>init</tt>-family selector which
				1314	returns a non-Objective-C type (e.g. <tt>void</tt>) is perfectly
				1315	well-formed; it simply isn't in the <tt>init</tt> family.</p></div>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1316	</li>
				1317	</ul>
				1318
				1319	<p>A program is ill-formed if a method's declarations,
				1320	implementations, and overrides do not all have the same method
				1321	family.</p>
				1322
				1323	<div id="family.attribute">
				1324	<h1>Explicit method family control</h1>
				1325
				1326	<p>A method may be annotated with the <tt>objc_method_family</tt>
				1327	attribute to precisely control which method family it belongs to. If
				1328	a method in an <tt>@implementation</tt> does not have this attribute,
				1329	but there is a method declared in the corresponding <tt>@interface</tt>
				1330	that does, then the attribute is copied to the declaration in the
				1331	<tt>@implementation</tt>. The attribute is available outside of ARC,
				1332	and may be tested for with the preprocessor query
				1333	<tt>__has_attribute(objc_method_family)</tt>.</p>
				1334
				1335	<p>The attribute is spelled
				1336	<tt>__attribute__((objc_method_family(<i>family</i>)))</tt>.
				1337	If <i>family</i> is <tt>none</tt>, the method has no family, even if
				1338	it would otherwise be considered to have one based on its selector and
				1339	type. Otherwise, <i>family</i> must be one
				1340	of <tt>alloc</tt>, <tt>copy</tt>, <tt>init</tt>,
				1341	<tt>mutableCopy</tt>, or <tt>new</tt>, in which case the method is
				1342	considered to belong to the corresponding family regardless of its
				1343	selector. It is an error if a method that is explicitly added to a
				1344	family in this way does not meet the requirements of the family other
				1345	than the selector naming convention.</p>
				1346
				1347	<div class="rationale"><p>Rationale: the rules codified in this document
				1348	describe the standard conventions of Objective-C. However, as these
				1349	conventions have not heretofore been enforced by an unforgiving
				1350	mechanical system, they are only imperfectly kept, especially as they
				1351	haven't always even been precisely defined. While it is possible to
				1352	define low-level ownership semantics with attributes like
				1353	<tt>ns_returns_retained</tt>, this attribute allows the user to
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	1354	communicate semantic intent, which is of use both to ARC (which, e.g.,
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1355	treats calls to <tt>init</tt> specially) and the static analyzer.</p></div>
				1356	</div>
				1357
				1358	<div id="family.semantics">
				1359	<h1>Semantics of method families</h1>
				1360
				1361	<p>A method's membership in a method family may imply non-standard
				1362	semantics for its parameters and return type.</p>
				1363
				1364	<p>Methods in the <tt>alloc</tt>, <tt>copy</tt>, <tt>mutableCopy</tt>,
				1365	and <tt>new</tt> families — that is, methods in all the
John McCall	be16b89	2011-06-18 08:15:19 +0000	[diff] [blame]	1366	currently-defined families except <tt>init</tt> — implicitly
				1367	<a href="#objects.operands.retained_returns">return a retained
				1368	object</a> as if they were annotated with
				1369	the <tt>ns_returns_retained</tt> attribute. This can be overridden by
				1370	annotating the method with either of
				1371	the <tt>ns_returns_autoreleased</tt> or
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1372	<tt>ns_returns_not_retained</tt> attributes.</p>
				1373
Fariborz Jahanian	acd4aaf	2011-07-06 22:47:46 +0000	[diff] [blame]	1374	<p>Properties also follow same naming rules as methods. This means that
				1375	those in the <tt>alloc</tt>, <tt>copy</tt>, <tt>mutableCopy</tt>,
				1376	and <tt>new</tt> families provide access to
				1377	<a href="#objects.operands.retained_returns">retained objects</a>.
				1378	This can be overridden by annotating the property with
				1379	<tt>ns_returns_not_retained</tt> attribute.</p>
				1380
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1381	<div id="family.semantics.init">
				1382	<h1>Semantics of <tt>init</tt></h1>
John McCall	be16b89	2011-06-18 08:15:19 +0000	[diff] [blame]	1383	<p>Methods in the <tt>init</tt> family implicitly
				1384	<a href="#objects.operands.consumed">consume</a> their <tt>self</tt>
				1385	parameter and <a href="#objects.operands.retained_returns">return a
				1386	retained object</a>. Neither of these properties can be altered
				1387	through attributes.</p>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1388
				1389	<p>A call to an <tt>init</tt> method with a receiver that is either
				1390	<tt>self</tt> (possibly parenthesized or casted) or <tt>super</tt> is
				1391	called a <span class="term">delegate init call</span>. It is an error
				1392	for a delegate init call to be made except from an <tt>init</tt>
				1393	method, and excluding blocks within such methods.</p>
				1394
John McCall	be16b89	2011-06-18 08:15:19 +0000	[diff] [blame]	1395	<p>As an exception to the <a href="misc.self">usual rule</a>, the
				1396	variable <tt>self</tt> is mutable in an <tt>init</tt> method and has
				1397	the usual semantics for a <tt>__strong</tt> variable. However, it is
				1398	undefined behavior and the program is ill-formed, no diagnostic
				1399	required, if an <tt>init</tt> method attempts to use the previous
				1400	value of <tt>self</tt> after the completion of a delegate init call.
				1401	It is conventional, but not required, for an <tt>init</tt> method to
				1402	return <tt>self</tt>.</p>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1403
John McCall	be16b89	2011-06-18 08:15:19 +0000	[diff] [blame]	1404	<p>It is undefined behavior for a program to cause two or more calls
				1405	to <tt>init</tt> methods on the same object, except that
				1406	each <tt>init</tt> method invocation may perform at most one delegate
				1407	init call.</p>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1408
John McCall	f3d08a6	2011-06-18 07:31:30 +0000	[diff] [blame]	1409	</div> <!-- family.semantics.init -->
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1410
				1411	<div id="family.semantics.result_type">
				1412	<h1>Related result types</h1>
				1413
				1414	<p>Certain methods are candidates to have <span class="term">related
				1415	result types</span>:</p>
				1416	<ul>
				1417	<li>class methods in the <tt>alloc</tt> and <tt>new</tt> method families</li>
				1418	<li>instance methods in the <tt>init</tt> family</li>
				1419	<li>the instance method <tt>self</tt></li>
				1420	<li>outside of ARC, the instance methods <tt>retain</tt> and <tt>autorelease</tt></li>
				1421	</ul>
				1422
				1423	<p>If the formal result type of such a method is <tt>id</tt> or
				1424	protocol-qualified <tt>id</tt>, or a type equal to the declaring class
				1425	or a superclass, then it is said to have a related result type. In
				1426	this case, when invoked in an explicit message send, it is assumed to
				1427	return a type related to the type of the receiver:</p>
				1428
				1429	<ul>
				1430	<li>if it is a class method, and the receiver is a class
				1431	name <tt>T</tt>, the message send expression has type <tt>T*</tt>;
				1432	otherwise</li>
				1433	<li>if it is an instance method, and the receiver has type <tt>T</tt>,
				1434	the message send expression has type <tt>T</tt>; otherwise</li>
				1435	<li>the message send expression has the normal result type of the
				1436	method.</li>
				1437	</ul>
				1438
				1439	<p>This is a new rule of the Objective-C language and applies outside
				1440	of ARC.</p>
				1441
				1442	<div class="rationale"><p>Rationale: ARC's automatic code emission is
				1443	more prone than most code to signature errors, i.e. errors where a
				1444	call was emitted against one method signature, but the implementing
				1445	method has an incompatible signature. Having more precise type
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	1446	information helps drastically lower this risk, as well as catching
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1447	a number of latent bugs.</p></div>
				1448
				1449	</div> <!-- family.semantics.result_type -->
				1450	</div> <!-- family.semantics -->
				1451	</div> <!-- family -->
				1452
				1453	<div id="optimization">
				1454	<h1>Optimization</h1>
				1455
				1456	<p>ARC applies aggressive rules for the optimization of local
				1457	behavior. These rules are based around a core assumption of
				1458	<span class="term">local balancing</span>: that other code will
				1459	perform retains and releases as necessary (and only as necessary) for
				1460	its own safety, and so the optimizer does not need to consider global
				1461	properties of the retain and release sequence. For example, if a
				1462	retain and release immediately bracket a call, the optimizer can
				1463	delete the retain and release on the assumption that the called
				1464	function will not do a constant number of unmotivated releases
				1465	followed by a constant number of <q>balancing</q> retains, such that
				1466	the local retain/release pair is the only thing preventing the called
				1467	function from ending up with a dangling reference.</p>
				1468
				1469	<p>The optimizer assumes that when a new value enters local control,
				1470	e.g. from a load of a non-local object or as the result of a function
				1471	call, it is instaneously valid. Subsequently, a retain and release of
				1472	a value are necessary on a computation path only if there is a use of
				1473	that value before the release and after any operation which might
				1474	cause a release of the value (including indirectly or non-locally),
				1475	and only if the value is not demonstrably already retained.</p>
				1476
				1477	<p>The complete optimization rules are quite complicated, but it would
				1478	still be useful to document them here.</p>
				1479
John McCall	dc7c5ad	2011-07-22 08:53:00 +0000	[diff] [blame]	1480	<div id="optimization.precise">
				1481	<h1>Precise lifetime semantics</h1>
				1482
				1483	<p>In general, ARC maintains an invariant that a retainable object
				1484	pointer held in a <tt>__strong</tt> object will be retained for the
				1485	full formal lifetime of the object. Objects subject to this invariant
				1486	have <span class="term">precise lifetime semantics</span>.</p>
				1487
				1488	<p>By default, local variables of automatic storage duration do not
				1489	have precise lifetime semantics. Such objects are simply strong
				1490	references which hold values of retainable object pointer type, and
				1491	these values are still fully subject to the optimizations on values
				1492	under local control.</p>
				1493
				1494	<div class="rationale"><p>Rationale: applying these precise-lifetime
				1495	semantics strictly would be prohibitive. Many useful optimizations
				1496	that might theoretically decrease the lifetime of an object would be
				1497	rendered impossible. Essentially, it promises too much.</p></div>
				1498
				1499	<p>A local variable of retainable object owner type and automatic
				1500	storage duration may be annotated with the <tt>objc_precise_lifetime</tt>
				1501	attribute to indicate that it should be considered to be an object
				1502	with precise lifetime semantics.</p>
				1503
				1504	<div class="rationale"><p>Rationale: nonetheless, it is sometimes
				1505	useful to be able to force an object to be released at a precise time,
				1506	even if that object does not appear to be used. This is likely to be
				1507	uncommon enough that the syntactic weight of explicitly requesting
				1508	these semantics will not be burdensome, and may even make the code
				1509	clearer.</p></div>
				1510
				1511	</div> <!-- optimization.precise -->
				1512
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	1513	</div> <!-- optimization -->
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1514
				1515	<div id="misc">
				1516	<h1>Miscellaneous</h1>
				1517
John McCall	f3d08a6	2011-06-18 07:31:30 +0000	[diff] [blame]	1518	<div id="misc.special_methods">
				1519	<h1>Special methods</h1>
				1520
				1521	<div id="misc.special_methods.retain">
				1522	<h1>Memory management methods</h1>
				1523
				1524	<p>A program is ill-formed if it contains a method definition, message
				1525	send, or <tt>@selector</tt> expression for any of the following
Benjamin Kramer	665a8dc	2012-01-15 15:26:07 +0000	[diff] [blame]	1526	selectors:</p>
John McCall	f3d08a6	2011-06-18 07:31:30 +0000	[diff] [blame]	1527	<ul>
				1528	<li><tt>autorelease</tt></li>
				1529	<li><tt>release</tt></li>
				1530	<li><tt>retain</tt></li>
				1531	<li><tt>retainCount</tt></li>
				1532	</ul>
John McCall	f3d08a6	2011-06-18 07:31:30 +0000	[diff] [blame]	1533
				1534	<div class="rationale"><p>Rationale: <tt>retainCount</tt> is banned
				1535	because ARC robs it of consistent semantics. The others were banned
				1536	after weighing three options for how to deal with message sends:</p>
				1537
				1538	<p><b>Honoring</b> them would work out very poorly if a programmer
				1539	naively or accidentally tried to incorporate code written for manual
				1540	retain/release code into an ARC program. At best, such code would do
				1541	twice as much work as necessary; quite frequently, however, ARC and
				1542	the explicit code would both try to balance the same retain, leading
				1543	to crashes. The cost is losing the ability to perform <q>unrooted</q>
				1544	retains, i.e. retains not logically corresponding to a strong
				1545	reference in the object graph.</p>
				1546
				1547	<p><b>Ignoring</b> them would badly violate user expectations about their
				1548	code. While it <em>would</em> make it easier to develop code simultaneously
				1549	for ARC and non-ARC, there is very little reason to do so except for
				1550	certain library developers. ARC and non-ARC translation units share
				1551	an execution model and can seamlessly interoperate. Within a
				1552	translation unit, a developer who faithfully maintains their code in
				1553	non-ARC mode is suffering all the restrictions of ARC for zero
				1554	benefit, while a developer who isn't testing the non-ARC mode is
				1555	likely to be unpleasantly surprised if they try to go back to it.</p>
				1556
				1557	<p><b>Banning</b> them has the disadvantage of making it very awkward
				1558	to migrate existing code to ARC. The best answer to that, given a
				1559	number of other changes and restrictions in ARC, is to provide a
				1560	specialized tool to assist users in that migration.</p>
				1561
				1562	<p>Implementing these methods was banned because they are too integral
				1563	to the semantics of ARC; many tricks which worked tolerably under
				1564	manual reference counting will misbehave if ARC performs an ephemeral
				1565	extra retain or two. If absolutely required, it is still possible to
				1566	implement them in non-ARC code, for example in a category; the
				1567	implementations must obey the <a href="#objects.retains">semantics</a>
				1568	laid out elsewhere in this document.</p>
				1569
				1570	</div>
				1571	</div> <!-- misc.special_methods.retain -->
				1572
				1573	<div id="misc.special_methods.dealloc">
				1574	<h1><tt>dealloc</tt></h1>
				1575
				1576	<p>A program is ill-formed if it contains a message send
				1577	or <tt>@selector</tt> expression for the selector <tt>dealloc</tt>.</p>
				1578
				1579	<div class="rationale"><p>Rationale: there are no legitimate reasons
				1580	to call <tt>dealloc</tt> directly.</p></div>
				1581
				1582	<p>A class may provide a method definition for an instance method
				1583	named <tt>dealloc</tt>. This method will be called after the final
				1584	<tt>release</tt> of the object but before it is deallocated or any of
				1585	its instance variables are destroyed. The superclass's implementation
				1586	of <tt>dealloc</tt> will be called automatically when the method
				1587	returns.</p>
				1588
				1589	<div class="rationale"><p>Rationale: even though ARC destroys instance
				1590	variables automatically, there are still legitimate reasons to write
				1591	a <tt>dealloc</tt> method, such as freeing non-retainable resources.
				1592	Failing to call <tt>[super dealloc]</tt> in such a method is nearly
				1593	always a bug. Sometimes, the object is simply trying to prevent
				1594	itself from being destroyed, but <tt>dealloc</tt> is really far too
				1595	late for the object to be raising such objections. Somewhat more
				1596	legitimately, an object may have been pool-allocated and should not be
				1597	deallocated with <tt>free</tt>; for now, this can only be supported
				1598	with a <tt>dealloc</tt> implementation outside of ARC. Such an
				1599	implementation must be very careful to do all the other work
				1600	that <tt>NSObject</tt>'s <tt>dealloc</tt> would, which is outside the
				1601	scope of this document to describe.</p></div>
				1602
				1603	</div>
				1604
				1605	</div> <!-- misc.special_methods -->
				1606
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1607	<div id="autoreleasepool">
				1608	<h1><tt>@autoreleasepool</tt></h1>
				1609
				1610	<p>To simplify the use of autorelease pools, and to bring them under
				1611	the control of the compiler, a new kind of statement is available in
				1612	Objective-C. It is written <tt>@autoreleasepool</tt> followed by
				1613	a <i>compound-statement</i>, i.e. by a new scope delimited by curly
				1614	braces. Upon entry to this block, the current state of the
				1615	autorelease pool is captured. When the block is exited normally,
				1616	whether by fallthrough or directed control flow (such
				1617	as <tt>return</tt> or <tt>break</tt>), the autorelease pool is
				1618	restored to the saved state, releasing all the objects in it. When
				1619	the block is exited with an exception, the pool is not drained.</p>
				1620
John McCall	f3d08a6	2011-06-18 07:31:30 +0000	[diff] [blame]	1621	<p><tt>@autoreleasepool</tt> may be used in non-ARC translation units,
				1622	with equivalent semantics.</p>
				1623
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1624	<p>A program is ill-formed if it refers to the
				1625	<tt>NSAutoreleasePool</tt> class.</p>
				1626
				1627	<div class="rationale"><p>Rationale: autorelease pools are clearly
				1628	important for the compiler to reason about, but it is far too much to
				1629	expect the compiler to accurately reason about control dependencies
				1630	between two calls. It is also very easy to accidentally forget to
				1631	drain an autorelease pool when using the manual API, and this can
				1632	significantly inflate the process's high-water-mark. The introduction
				1633	of a new scope is unfortunate but basically required for sane
				1634	interaction with the rest of the language. Not draining the pool
				1635	during an unwind is apparently required by the Objective-C exceptions
				1636	implementation.</p></div>
				1637
				1638	</div> <!-- autoreleasepool -->
				1639
				1640	<div id="misc.self">
				1641	<h1><tt>self</tt></h1>
				1642
				1643	<p>The <tt>self</tt> parameter variable of an Objective-C method is
				1644	never actually retained by the implementation. It is undefined
				1645	behavior, or at least dangerous, to cause an object to be deallocated
Ted Kremenek	2bbcd5c	2011-11-14 21:59:25 +0000	[diff] [blame]	1646	during a message send to that object.</p>
				1647
				1648	<p>To make this safe, for Objective-C instance methods <tt>self</tt> is
				1649	implicitly <tt>const</tt> unless the method is in the <a
				1650	href="#family.semantics.init"><tt>init</tt> family</a>. Further, <tt>self</tt>
				1651	is <b>always</b> implicitly <tt>const</tt> within a class method.</p>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1652
				1653	<div class="rationale"><p>Rationale: the cost of
				1654	retaining <tt>self</tt> in all methods was found to be prohibitive, as
				1655	it tends to be live across calls, preventing the optimizer from
				1656	proving that the retain and release are unnecessary — for good
				1657	reason, as it's quite possible in theory to cause an object to be
				1658	deallocated during its execution without this retain and release.
				1659	Since it's extremely uncommon to actually do so, even unintentionally,
				1660	and since there's no natural way for the programmer to remove this
				1661	retain/release pair otherwise (as there is for other parameters by,
				1662	say, making the variable <tt>__unsafe_unretained</tt>), we chose to
				1663	make this optimizing assumption and shift some amount of risk to the
				1664	user.</p></div>
				1665
				1666	</div> <!-- misc.self -->
				1667
				1668	<div id="misc.enumeration">
				1669	<h1>Fast enumeration iteration variables</h1>
				1670
				1671	<p>If a variable is declared in the condition of an Objective-C fast
				1672	enumeration loop, and the variable has no explicit ownership
				1673	qualifier, then it is qualified with <tt>const __strong</tt> and
				1674	objects encountered during the enumeration are not actually
				1675	retained.</p>
				1676
				1677	<div class="rationale"><p>Rationale: this is an optimization made
				1678	possible because fast enumeration loops promise to keep the objects
				1679	retained during enumeration, and the collection itself cannot be
				1680	synchronously modified. It can be overridden by explicitly qualifying
				1681	the variable with <tt>__strong</tt>, which will make the variable
				1682	mutable again and cause the loop to retain the objects it
Benjamin Kramer	665a8dc	2012-01-15 15:26:07 +0000	[diff] [blame]	1683	encounters.</p></div>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1684
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	1685	</div> <!-- misc.enumeration -->
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1686
				1687	<div id="misc.blocks">
				1688	<h1>Blocks</h1>
				1689
				1690	<p>The implicit <tt>const</tt> capture variables created when
				1691	evaluating a block literal expression have the same ownership
				1692	semantics as the local variables they capture. The capture is
				1693	performed by reading from the captured variable and initializing the
				1694	capture variable with that value; the capture variable is destroyed
				1695	when the block literal is, i.e. at the end of the enclosing scope.</p>
				1696
				1697	<p>The <a href="#ownership.inference">inference</a> rules apply
				1698	equally to <tt>__block</tt> variables, which is a shift in semantics
				1699	from non-ARC, where <tt>__block</tt> variables did not implicitly
				1700	retain during capture.</p>
				1701
				1702	<p><tt>__block</tt> variables of retainable object owner type are
				1703	moved off the stack by initializing the heap copy with the result of
Benjamin Kramer	665a8dc	2012-01-15 15:26:07 +0000	[diff] [blame]	1704	moving from the stack copy.</p>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1705
				1706	<p>With the exception of retains done as part of initializing
				1707	a <tt>__strong</tt> parameter variable or reading a <tt>__weak</tt>
				1708	variable, whenever these semantics call for retaining a value of
				1709	block-pointer type, it has the effect of a <tt>Block_copy</tt>. The
				1710	optimizer may remove such copies when it sees that the result is
				1711	used only as an argument to a call.</p>
				1712
				1713	</div> <!-- misc.blocks -->
				1714
				1715	<div id="misc.exceptions">
				1716	<h1>Exceptions</h1>
				1717
				1718	<p>By default in Objective C, ARC is not exception-safe for normal
Benjamin Kramer	665a8dc	2012-01-15 15:26:07 +0000	[diff] [blame]	1719	releases:</p>
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1720	<ul>
				1721	<li>It does not end the lifetime of <tt>__strong</tt> variables when
				1722	their scopes are abnormally terminated by an exception.</li>
				1723	<li>It does not perform releases which would occur at the end of
				1724	a full-expression if that full-expression throws an exception.</li>
				1725	</ul>
				1726
				1727	<p>A program may be compiled with the option
				1728	<tt>-fobjc-arc-exceptions</tt> in order to enable these, or with the
				1729	option <tt>-fno-objc-arc-exceptions</tt> to explicitly disable them,
				1730	with the last such argument <q>winning</q>.</p>
				1731
				1732	<div class="rationale"><p>Rationale: the standard Cocoa convention is
				1733	that exceptions signal programmer error and are not intended to be
				1734	recovered from. Making code exceptions-safe by default would impose
				1735	severe runtime and code size penalties on code that typically does not
				1736	actually care about exceptions safety. Therefore, ARC-generated code
				1737	leaks by default on exceptions, which is just fine if the process is
				1738	going to be immediately terminated anyway. Programs which do care
				1739	about recovering from exceptions should enable the option.</p></div>
				1740
				1741	<p>In Objective-C++, <tt>-fobjc-arc-exceptions</tt> is enabled by
				1742	default.</p>
				1743
				1744	<div class="rationale"><p>Rationale: C++ already introduces pervasive
				1745	exceptions-cleanup code of the sort that ARC introduces. C++
				1746	programmers who have not already disabled exceptions are much more
				1747	likely to actual require exception-safety.</p></div>
				1748
				1749	<p>ARC does end the lifetimes of <tt>__weak</tt> objects when an
				1750	exception terminates their scope unless exceptions are disabled in the
				1751	compiler.</p>
				1752
				1753	<div class="rationale"><p>Rationale: the consequence of a
				1754	local <tt>__weak</tt> object not being destroyed is very likely to be
				1755	corruption of the Objective-C runtime, so we want to be safer here.
				1756	Of course, potentially massive leaks are about as likely to take down
				1757	the process as this corruption is if the program does try to recover
				1758	from exceptions.</p></div>
				1759
				1760	</div> <!-- misc.exceptions -->
				1761
John McCall	dc7c5ad	2011-07-22 08:53:00 +0000	[diff] [blame]	1762	<div id="misc.interior">
				1763	<h1>Interior pointers</h1>
				1764
				1765	<p>An Objective-C method returning a non-retainable pointer may be
				1766	annotated with the <tt>objc_returns_inner_pointer</tt> attribute to
				1767	indicate that it returns a handle to the internal data of an object,
				1768	and that this reference will be invalidated if the object is
				1769	destroyed. When such a message is sent to an object, the object's
				1770	lifetime will be extended until at least the earliest of:</p>
				1771
				1772	<ul>
				1773	<li>the last use of the returned pointer, or any pointer derived from
				1774	it, in the calling function or</li>
				1775	<li>the autorelease pool is restored to a previous state.</li>
				1776	</ul>
				1777
				1778	<div class="rationale"><p>Rationale: not all memory and resources are
				1779	managed with reference counts; it is common for objects to manage
				1780	private resources in their own, private way. Typically these
				1781	resources are completely encapsulated within the object, but some
				1782	classes offer their users direct access for efficiency. If ARC is not
				1783	aware of methods that return such <q>interior</q> pointers, its
				1784	optimizations can cause the owning object to be reclaimed too soon.
				1785	This attribute informs ARC that it must tread lightly.</p>
				1786
				1787	<p>The extension rules are somewhat intentionally vague. The
				1788	autorelease pool limit is there to permit a simple implementation to
				1789	simply retain and autorelease the receiver. The other limit permits
				1790	some amount of optimization. The phrase <q>derived from</q> is
				1791	intended to encompass the results both of pointer transformations,
				1792	such as casts and arithmetic, and of loading from such derived
				1793	pointers; furthermore, it applies whether or not such derivations are
				1794	applied directly in the calling code or by other utility code (for
				1795	example, the C library routine <tt>strchr</tt>). However, the
				1796	implementation never need account for uses after a return from the
				1797	code which calls the method returning an interior pointer.</p></div>
				1798
				1799	<p>As an exception, no extension is required if the receiver is loaded
				1800	directly from a <tt>__strong</tt> object
				1801	with <a href="#optimization.precise">precise lifetime semantics</a>.</p>
				1802
				1803	<div class="rationale"><p>Rationale: implicit autoreleases carry the
				1804	risk of significantly inflating memory use, so it's important to
				1805	provide users a way of avoiding these autoreleases. Tying this to
				1806	precise lifetime semantics is ideal, as for local variables this
				1807	requires a very explicit annotation, which allows ARC to trust the
				1808	user with good cheer.</p></div>
				1809
				1810	</div> <!-- misc.interior -->
				1811
John McCall	9da31cb	2012-03-27 07:42:12 +0000	[diff] [blame]	1812	<div id="misc.c-retainable">
				1813	<h1>C retainable pointer types</h1>
				1814
				1815	<p>A type is a <span class="term">C retainable pointer type</span>
				1816	if it is a pointer to (possibly qualified) <tt>void</tt> or a
				1817	pointer to a (possibly qualifier) <tt>struct</tt> or <tt>class</tt>
				1818	type.</p>
				1819
				1820	<div class="rationale"><p>Rationale: ARC does not manage pointers of
				1821	CoreFoundation type (or any of the related families of retainable C
				1822	pointers which interoperate with Objective-C for retain/release
				1823	operation). In fact, ARC does not even know how to distinguish these
				1824	types from arbitrary C pointer types. The intent of this concept is
				1825	to filter out some obviously non-object types while leaving a hook for
				1826	later tightening if a means of exhaustively marking CF types is made
				1827	available.</p></div>
				1828
				1829	<div id="misc.c-retainable.audit">
				1830	<h1>Auditing of C retainable pointer interfaces</h1>
				1831
				1832	<p><span class="revision"><span class="whenRevised">[beginning Apple 4.0, LLVM 3.1]</span></span></p>
				1833
				1834	<p>A C function may be marked with the <tt>cf_audited_transfer</tt>
				1835	attribute to express that, except as otherwise marked with attributes,
				1836	it obeys the parameter (consuming vs. non-consuming) and return
				1837	(retained vs. non-retained) conventions for a C function of its name,
				1838	namely:</p>
				1839
				1840	<ul>
				1841	<li>A parameter of C retainable pointer type is assumed to not be
				1842	consumed unless it is marked with the <tt>cf_consumed</tt> attribute, and</li>
				1843	<li>A result of C retainable pointer type is assumed to not be
				1844	returned retained unless the function is either
				1845	marked <tt>cf_returns_retained</tt> or it follows
				1846	the create/copy naming convention and is not
				1847	marked <tt>cf_returns_not_retained</tt>.</li>
				1848	</ul>
				1849
				1850	<p>A function obeys the <span class="term">create/copy</span> naming
				1851	convention if its name contains as a substring:</p>
				1852	<ul>
				1853	<li>either <q>Create</q> or <q>Copy</q> not followed by a lowercase letter, or</li>
				1854	<li>either <q>create</q> or <q>copy</q> not followed by a lowercase
				1855	letter and not preceded by any letter, whether uppercase or lowercase.</li>
				1856	</ul>
				1857
				1858	<p>A second attribute, <tt>cf_unknown_transfer</tt>, signifies that a
				1859	function's transfer semantics cannot be accurately captured using any
				1860	of these annotations. A program is ill-formed if it annotates the
				1861	same function with both <tt>cf_audited_transfer</tt>
				1862	and <tt>cf_unknown_transfer</tt>.</p>
				1863
				1864	<p>A pragma is provided to faciliate the mass annotation of interfaces:</p>
				1865
				1866	<pre>#pragma arc_cf_code_audited begin
				1867	...
				1868	#pragma arc_cf_code_audited end</pre>
				1869
				1870	<p>All C functions declared within the extent of this pragma are
				1871	treated as if annotated with the <tt>cf_audited_transfer</tt>
				1872	attribute unless they otherwise have the <tt>cf_unknown_transfer</tt>
				1873	attribute. The pragma is accepted in all language modes. A program
				1874	is ill-formed if it attempts to change files, whether by including a
				1875	file or ending the current file, within the extent of this pragma.</p>
				1876
				1877	<p>It is possible to test for all the features in this section with
				1878	<tt>__has_feature(arc_cf_code_audited)</tt>.</p>
				1879
				1880	<div class="rationale"><p>Rationale: A significant inconvenience in
				1881	ARC programming is the necessity of interacting with APIs based around
				1882	C retainable pointers. These features are designed to make it
				1883	relatively easy for API authors to quickly review and annotate their
				1884	interfaces, in turn improving the fidelity of tools such as the static
				1885	analyzer and ARC. The single-file restriction on the pragma is
				1886	designed to eliminate the risk of accidentally annotating some other
				1887	header's interfaces.</p></div>
				1888
				1889	</div> <!-- misc.c-retainable.audit -->
				1890
				1891	</div> <!-- misc.c-retainable -->
				1892
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	1893	</div> <!-- misc -->
John McCall	98a48cf	2011-06-19 09:36:02 +0000	[diff] [blame]	1894
				1895	<div id="runtime">
				1896	<h1>Runtime support</h1>
				1897
John McCall	085d09d	2011-06-19 10:12:24 +0000	[diff] [blame]	1898	<p>This section describes the interaction between the ARC runtime and
				1899	the code generated by the ARC compiler. This is not part of the ARC
				1900	language specification; instead, it is effectively a language-specific
				1901	ABI supplement, akin to the <q>Itanium</q> generic ABI for C++.</p>
John McCall	98a48cf	2011-06-19 09:36:02 +0000	[diff] [blame]	1902
				1903	<p>Ownership qualification does not alter the storage requirements for
John McCall	085d09d	2011-06-19 10:12:24 +0000	[diff] [blame]	1904	objects, except that it is undefined behavior if a <tt>__weak</tt>
				1905	object is inadequately aligned for an object of type <tt>id</tt>. The
				1906	other qualifiers may be used on explicitly under-aligned memory.</p>
John McCall	98a48cf	2011-06-19 09:36:02 +0000	[diff] [blame]	1907
				1908	<p>The runtime tracks <tt>__weak</tt> objects which holds non-null
John McCall	3914a30	2011-06-19 09:59:33 +0000	[diff] [blame]	1909	values. It is undefined behavior to direct modify a <tt>__weak</tt>
				1910	object which is being tracked by the runtime except through an
				1911	<a href="#runtime.objc_storeWeak"><tt>objc_storeWeak</tt></a>,
				1912	<a href="#runtime.objc_destroyWeak"><tt>objc_destroyWeak</tt></a>,
				1913	or <a href="#runtime.objc_moveWeak"><tt>objc_moveWeak</tt></a>
John McCall	98a48cf	2011-06-19 09:36:02 +0000	[diff] [blame]	1914	call.</p>
				1915
John McCall	3914a30	2011-06-19 09:59:33 +0000	[diff] [blame]	1916	<p>The runtime must provide a number of new entrypoints which the
				1917	compiler may emit, which are described in the remainder of this
				1918	section.</p>
				1919
				1920	<div class="rationale"><p>Rationale: Several of these functions are
				1921	semantically equivalent to a message send; we emit calls to C
				1922	functions instead because:</p>
				1923	<ul>
				1924	<li>the machine code to do so is significantly smaller,</li>
				1925	<li>it is much easier to recognize the C functions in the ARC optimizer, and</li>
				1926	<li>a sufficient sophisticated runtime may be able to avoid the
				1927	message send in common cases.</li>
				1928	</ul>
				1929
				1930	<p>Several other of these functions are <q>fused</q> operations which
				1931	can be described entirely in terms of other operations. We use the
				1932	fused operations primarily as a code-size optimization, although in
				1933	some cases there is also a real potential for avoiding redundant
				1934	operations in the runtime.</p>
				1935
				1936	</div>
				1937
John McCall	98a48cf	2011-06-19 09:36:02 +0000	[diff] [blame]	1938	<div id="runtime.objc_autorelease">
				1939	<h1><tt>id objc_autorelease(id value);</tt></h1>
				1940	<p><i>Precondition:</i> <tt>value</tt> is null or a pointer to a
				1941	valid object.</p>
				1942	<p>If <tt>value</tt> is null, this call has no effect. Otherwise, it
				1943	adds the object to the innermost autorelease pool exactly as if the
				1944	object had been sent the <tt>autorelease</tt> message.</p>
				1945	<p>Always returns <tt>value</tt>.</p>
				1946	</div> <!-- runtime.objc_autorelease -->
				1947
				1948	<div id="runtime.objc_autoreleasePoolPop">
				1949	<h1><tt>void objc_autoreleasePoolPop(void *pool);</tt></h1>
				1950	<p><i>Precondition:</i> <tt>pool</tt> is the result of a previous call to
				1951	<a href="runtime.objc_autoreleasePoolPush"><tt>objc_autoreleasePoolPush</tt></a>
				1952	on the current thread, where neither <tt>pool</tt> nor any enclosing
				1953	pool have previously been popped.</p>
				1954	<p>Releases all the objects added to the given autorelease pool and
				1955	any autorelease pools it encloses, then sets the current autorelease
				1956	pool to the pool directly enclosing <tt>pool</tt>.</p>
				1957	</div> <!-- runtime.objc_autoreleasePoolPop -->
				1958
				1959	<div id="runtime.objc_autoreleasePoolPush">
				1960	<h1><tt>void *objc_autoreleasePoolPush(void);</tt></h1>
				1961	<p>Creates a new autorelease pool that is enclosed by the current
				1962	pool, makes that the current pool, and returns an opaque <q>handle</q>
				1963	to it.</p>
				1964
				1965	<div class="rationale"><p>Rationale: while the interface is described
				1966	as an explicit hierarchy of pools, the rules allow the implementation
				1967	to just keep a stack of objects, using the stack depth as the opaque
				1968	pool handle.</p></div>
				1969
				1970	</div> <!-- runtime.objc_autoreleasePoolPush -->
				1971
				1972	<div id="runtime.objc_autoreleaseReturnValue">
				1973	<h1><tt>id objc_autoreleaseReturnValue(id value);</tt></h1>
				1974	<p><i>Precondition:</i> <tt>value</tt> is null or a pointer to a
				1975	valid object.</p>
				1976	<p>If <tt>value</tt> is null, this call has no effect. Otherwise, it
				1977	makes a best effort to hand off ownership of a retain count on the
				1978	object to a call
				1979	to <a href="runtime.objc_retainAutoreleasedReturnValue"><tt>objc_retainAutoreleasedReturnValue</tt></a>
				1980	for the same object in an enclosing call frame. If this is not
				1981	possible, the object is autoreleased as above.</p>
				1982	<p>Always returns <tt>value</tt>.</p>
				1983	</div> <!-- runtime.objc_autoreleaseReturnValue -->
				1984
				1985	<div id="runtime.objc_copyWeak">
				1986	<h1><tt>void objc_copyWeak(id dest, id src);</tt></h1>
				1987	<p><i>Precondition:</i> <tt>src</tt> is a valid pointer which either
				1988	contains a null pointer or has been registered as a <tt>__weak</tt>
				1989	object. <tt>dest</tt> is a valid pointer which has not been
				1990	registered as a <tt>__weak</tt> object.</p>
				1991	<p><tt>dest</tt> is initialized to be equivalent to <tt>src</tt>,
				1992	potentially registering it with the runtime. Equivalent to the
				1993	following code:</p>
				1994	<pre>void objc_copyWeak(id dest, id src) {
				1995	objc_release(objc_initWeak(dest, objc_loadWeakRetained(src)));
				1996	}</pre>
				1997	<p>Must be atomic with respect to calls to <tt>objc_storeWeak</tt>
				1998	on <tt>src</tt>.</p>
				1999	</div> <!-- runtime.objc_copyWeak -->
				2000
				2001	<div id="runtime.objc_destroyWeak">
				2002	<h1><tt>void objc_destroyWeak(id *object);</tt></h1>
				2003	<p><i>Precondition:</i> <tt>object</tt> is a valid pointer which
				2004	either contains a null pointer or has been registered as
				2005	a <tt>__weak</tt> object.</p>
				2006	<p><tt>object</tt> is unregistered as a weak object, if it ever was.
				2007	The current value of <tt>object</tt> is left unspecified; otherwise,
				2008	equivalent to the following code:</p>
				2009	<pre>void objc_destroyWeak(id *object) {
				2010	objc_storeWeak(object, nil);
				2011	}</pre>
				2012	<p>Does not need to be atomic with respect to calls
				2013	to <tt>objc_storeWeak</tt> on <tt>object</tt>.</p>
				2014	</div> <!-- runtime.objc_destroyWeak -->
				2015
				2016	<div id="runtime.objc_initWeak">
				2017	<h1><tt>id objc_initWeak(id *object, id value);</tt></h1>
				2018	<p><i>Precondition:</i> <tt>object</tt> is a valid pointer which has
				2019	not been registered as a <tt>__weak</tt> object. <tt>value</tt> is
				2020	null or a pointer to a valid object.</p>
				2021	<p>If <tt>value</tt> is a null pointer or the object to which it
				2022	points has begun deallocation, <tt>object</tt> is zero-initialized.
				2023	Otherwise, <tt>object</tt> is registered as a <tt>__weak</tt> object
				2024	pointing to <tt>value</tt>. Equivalent to the following code:</p>
				2025	<pre>id objc_initWeak(id *object, id value) {
				2026	*object = nil;
				2027	return objc_storeWeak(object, value);
				2028	}</pre>
				2029	<p>Returns the value of <tt>object</tt> after the call.</p>
				2030	<p>Does not need to be atomic with respect to calls
				2031	to <tt>objc_storeWeak</tt> on <tt>object</tt>.</p>
				2032	</div> <!-- runtime.objc_initWeak -->
				2033
				2034	<div id="runtime.objc_loadWeak">
				2035	<h1><tt>id objc_loadWeak(id *object);</tt></h1>
				2036	<p><i>Precondition:</i> <tt>object</tt> is a valid pointer which
				2037	either contains a null pointer or has been registered as
				2038	a <tt>__weak</tt> object.</p>
				2039	<p>If <tt>object</tt> is registered as a <tt>__weak</tt> object, and
				2040	the last value stored into <tt>object</tt> has not yet been
				2041	deallocated or begun deallocation, retains and autoreleases that value
				2042	and returns it. Otherwise returns null. Equivalent to the following
				2043	code:</p>
				2044	<pre>id objc_loadWeak(id *object) {
				2045	return objc_autorelease(objc_loadWeakRetained(object));
				2046	}</pre>
				2047	<p>Must be atomic with respect to calls to <tt>objc_storeWeak</tt>
				2048	on <tt>object</tt>.</p>
				2049	<div class="rationale">Rationale: loading weak references would be
				2050	inherently prone to race conditions without the retain.</div>
				2051	</div> <!-- runtime.objc_loadWeak -->
				2052
				2053	<div id="runtime.objc_loadWeakRetained">
				2054	<h1><tt>id objc_loadWeakRetained(id *object);</tt></h1>
				2055	<p><i>Precondition:</i> <tt>object</tt> is a valid pointer which
				2056	either contains a null pointer or has been registered as
				2057	a <tt>__weak</tt> object.</p>
				2058	<p>If <tt>object</tt> is registered as a <tt>__weak</tt> object, and
				2059	the last value stored into <tt>object</tt> has not yet been
				2060	deallocated or begun deallocation, retains that value and returns it.
				2061	Otherwise returns null.</p>
				2062	<p>Must be atomic with respect to calls to <tt>objc_storeWeak</tt>
				2063	on <tt>object</tt>.</p>
				2064	</div> <!-- runtime.objc_loadWeakRetained -->
				2065
				2066	<div id="runtime.objc_moveWeak">
				2067	<h1><tt>void objc_moveWeak(id dest, id src);</tt></h1>
				2068	<p><i>Precondition:</i> <tt>src</tt> is a valid pointer which either
				2069	contains a null pointer or has been registered as a <tt>__weak</tt>
				2070	object. <tt>dest</tt> is a valid pointer which has not been
				2071	registered as a <tt>__weak</tt> object.</p>
				2072	<p><tt>dest</tt> is initialized to be equivalent to <tt>src</tt>,
				2073	potentially registering it with the runtime. <tt>src</tt> may then be
				2074	left in its original state, in which case this call is equivalent
				2075	to <a href="#runtime.objc_copyWeak"><tt>objc_copyWeak</tt></a>, or it
				2076	may be left as null.</p>
				2077	<p>Must be atomic with respect to calls to <tt>objc_storeWeak</tt>
				2078	on <tt>src</tt>.</p>
				2079	</div> <!-- runtime.objc_moveWeak -->
				2080
John McCall	085d09d	2011-06-19 10:12:24 +0000	[diff] [blame]	2081	<div id="runtime.objc_release">
				2082	<h1><tt>void objc_release(id value);</tt></h1>
				2083	<p><i>Precondition:</i> <tt>value</tt> is null or a pointer to a
				2084	valid object.</p>
				2085	<p>If <tt>value</tt> is null, this call has no effect. Otherwise, it
				2086	performs a release operation exactly as if the object had been sent
				2087	the <tt>release</tt> message.</p>
				2088	</div> <!-- runtime.objc_release -->
				2089
John McCall	98a48cf	2011-06-19 09:36:02 +0000	[diff] [blame]	2090	<div id="runtime.objc_retain">
				2091	<h1><tt>id objc_retain(id value);</tt></h1>
				2092	<p><i>Precondition:</i> <tt>value</tt> is null or a pointer to a
				2093	valid object.</p>
				2094	<p>If <tt>value</tt> is null, this call has no effect. Otherwise, it
				2095	performs a retain operation exactly as if the object had been sent
				2096	the <tt>retain</tt> message.</p>
				2097	<p>Always returns <tt>value</tt>.</p>
				2098	</div> <!-- runtime.objc_retain -->
				2099
				2100	<div id="runtime.objc_retainAutorelease">
				2101	<h1><tt>id objc_retainAutorelease(id value);</tt></h1>
				2102	<p><i>Precondition:</i> <tt>value</tt> is null or a pointer to a
				2103	valid object.</p>
				2104	<p>If <tt>value</tt> is null, this call has no effect. Otherwise, it
				2105	performs a retain operation followed by an autorelease operation.
				2106	Equivalent to the following code:</p>
				2107	<pre>id objc_retainAutorelease(id value) {
				2108	return objc_autorelease(objc_retain(value));
				2109	}</pre>
				2110	<p>Always returns <tt>value</tt>.</p>
				2111	</div> <!-- runtime.objc_retainAutorelease -->
				2112
				2113	<div id="runtime.objc_retainAutoreleaseReturnValue">
				2114	<h1><tt>id objc_retainAutoreleaseReturnValue(id value);</tt></h1>
				2115	<p><i>Precondition:</i> <tt>value</tt> is null or a pointer to a
				2116	valid object.</p>
				2117	<p>If <tt>value</tt> is null, this call has no effect. Otherwise, it
				2118	performs a retain operation followed by the operation described in
				2119	<a href="#runtime.objc_autoreleaseReturnValue"><tt>objc_autoreleaseReturnValue</tt></a>.
				2120	Equivalent to the following code:</p>
				2121	<pre>id objc_retainAutoreleaseReturnValue(id value) {
				2122	return objc_autoreleaseReturnValue(objc_retain(value));
				2123	}</pre>
				2124	<p>Always returns <tt>value</tt>.</p>
				2125	</div> <!-- runtime.objc_retainAutoreleaseReturnValue -->
				2126
				2127	<div id="runtime.objc_retainAutoreleasedReturnValue">
				2128	<h1><tt>id objc_retainAutoreleasedReturnValue(id value);</tt></h1>
				2129	<p><i>Precondition:</i> <tt>value</tt> is null or a pointer to a
				2130	valid object.</p>
				2131	<p>If <tt>value</tt> is null, this call has no effect. Otherwise, it
				2132	attempts to accept a hand off of a retain count from a call to
				2133	<a href="#runtime.objc_autoreleaseReturnValue"><tt>objc_autoreleaseReturnValue</tt></a>
				2134	on <tt>value</tt> in a recently-called function or something it
				2135	calls. If that fails, it performs a retain operation exactly
				2136	like <a href="#runtime.objc_retain"><tt>objc_retain</tt></a>.</p>
				2137	<p>Always returns <tt>value</tt>.</p>
				2138	</div> <!-- runtime.objc_retainAutoreleasedReturnValue -->
				2139
				2140	<div id="runtime.objc_retainBlock">
				2141	<h1><tt>id objc_retainBlock(id value);</tt></h1>
				2142	<p><i>Precondition:</i> <tt>value</tt> is null or a pointer to a
				2143	valid block object.</p>
				2144	<p>If <tt>value</tt> is null, this call has no effect. Otherwise, if
				2145	the block pointed to by <tt>value</tt> is still on the stack, it is
				2146	copied to the heap and the address of the copy is returned. Otherwise
				2147	a retain operation is performed on the block exactly as if it had been
				2148	sent the <tt>retain</tt> message.</p>
				2149	</div> <!-- runtime.objc_retainBlock -->
				2150
John McCall	98a48cf	2011-06-19 09:36:02 +0000	[diff] [blame]	2151	<div id="runtime.objc_storeStrong">
				2152	<h1><tt>id objc_storeStrong(id *object, id value);</tt></h1>
				2153	<p><i>Precondition:</i> <tt>object</tt> is a valid pointer to
				2154	a <tt>__strong</tt> object which is adequately aligned for a
				2155	pointer. <tt>value</tt> is null or a pointer to a valid object.</p>
				2156	<p>Performs the complete sequence for assigning to a <tt>__strong</tt>
				2157	object of non-block type. Equivalent to the following code:</p>
				2158	<pre>id objc_storeStrong(id *object, id value) {
				2159	value = [value retain];
				2160	id oldValue = *object;
				2161	*object = value;
				2162	[oldValue release];
				2163	return value;
				2164	}</pre>
				2165	<p>Always returns <tt>value</tt>.</p>
				2166	</div> <!-- runtime.objc_storeStrong -->
				2167
				2168	<div id="runtime.objc_storeWeak">
				2169	<h1><tt>id objc_storeWeak(id *object, id value);</tt></h1>
				2170	<p><i>Precondition:</i> <tt>object</tt> is a valid pointer which
				2171	either contains a null pointer or has been registered as
				2172	a <tt>__weak</tt> object. <tt>value</tt> is null or a pointer to a
				2173	valid object.</p>
				2174	<p>If <tt>value</tt> is a null pointer or the object to which it
				2175	points has begun deallocation, <tt>object</tt> is assigned null
				2176	and unregistered as a <tt>__weak</tt> object. Otherwise,
				2177	<tt>object</tt> is registered as a <tt>__weak</tt> object or has its
				2178	registration updated to point to <tt>value</tt>.</p>
				2179	<p>Returns the value of <tt>object</tt> after the call.</p>
				2180	</div> <!-- runtime.objc_storeWeak -->
				2181
				2182	</div> <!-- runtime -->
John McCall	8246702	2011-06-15 21:21:53 +0000	[diff] [blame]	2183	</div> <!-- root -->
				2184	</body>
				2185	</html>