5 This document is the primary reference for the Rust programming language. It
6 provides three kinds of material:
8 - Chapters that informally describe each language construct and their use.
9 - Chapters that informally describe the memory model, concurrency model,
10 runtime services, linkage model and debugging facilities.
11 - Appendix chapters providing rationale and references to languages that
12 influenced the design.
14 This document does not serve as an introduction to the language. Background
15 familiarity with the language is assumed. A separate [book] is available to
16 help acquire such background familiarity.
18 This document also does not serve as a reference to the [standard] library
19 included in the language distribution. Those libraries are documented
20 separately by extracting documentation attributes from their source code. Many
21 of the features that one might expect to be language features are library
22 features in Rust, so what you're looking for may be there, not here.
24 You may also be interested in the [grammar].
26 [book]: book/index.html
27 [standard]: std/index.html
28 [grammar]: grammar.html
32 ## Unicode productions
34 A few productions in Rust's grammar permit Unicode code points outside the
35 ASCII range. We define these productions in terms of character properties
36 specified in the Unicode standard, rather than in terms of ASCII-range code
37 points. The grammar has a [Special Unicode Productions][unicodeproductions]
38 section that lists these productions.
40 [unicodeproductions]: grammar.html#special-unicode-productions
42 ## String table productions
44 Some rules in the grammar — notably [unary
45 operators](#unary-operator-expressions), [binary
46 operators](#binary-operator-expressions), and [keywords][keywords] — are
47 given in a simplified form: as a listing of a table of unquoted, printable
48 whitespace-separated strings. These cases form a subset of the rules regarding
49 the [token](#tokens) rule, and are assumed to be the result of a
50 lexical-analysis phase feeding the parser, driven by a DFA, operating over the
51 disjunction of all such string table entries.
53 [keywords]: grammar.html#keywords
55 When such a string enclosed in double-quotes (`"`) occurs inside the grammar,
56 it is an implicit reference to a single member of such a string table
57 production. See [tokens](#tokens) for more information.
63 Rust input is interpreted as a sequence of Unicode code points encoded in UTF-8.
64 Most Rust grammar rules are defined in terms of printable ASCII-range
65 code points, but a small number are defined in terms of Unicode properties or
66 explicit code point lists. [^inputformat]
68 [^inputformat]: Substitute definitions for the special Unicode productions are
69 provided to the grammar verifier, restricted to ASCII range, when verifying the
70 grammar in this document.
74 An identifier is any nonempty Unicode[^non_ascii_idents] string of the following form:
76 [^non_ascii_idents]: Non-ASCII characters in identifiers are currently feature
77 gated. This is expected to improve soon.
81 * The first character has property `XID_start`
82 * The remaining characters have property `XID_continue`
86 * The first character is `_`
87 * The identifier is more than one character, `_` alone is not an identifier
88 * The remaining characters have property `XID_continue`
90 that does _not_ occur in the set of [keywords][keywords].
92 > **Note**: `XID_start` and `XID_continue` as character properties cover the
93 > character ranges used to form the more familiar C and Java language-family
98 Comments in Rust code follow the general C++ style of line (`//`) and
99 block (`/* ... */`) comment forms. Nested block comments are supported.
101 Line comments beginning with exactly _three_ slashes (`///`), and block
102 comments (`/** ... */`), are interpreted as a special syntax for `doc`
103 [attributes](#attributes). That is, they are equivalent to writing
104 `#[doc="..."]` around the body of the comment, i.e., `/// Foo` turns into
107 Line comments beginning with `//!` and block comments `/*! ... */` are
108 doc comments that apply to the parent of the comment, rather than the item
109 that follows. That is, they are equivalent to writing `#![doc="..."]` around
110 the body of the comment. `//!` comments are usually used to document
111 modules that occupy a source file.
113 Non-doc comments are interpreted as a form of whitespace.
117 Whitespace is any non-empty string containing only the following characters:
119 - `U+0020` (space, `' '`)
120 - `U+0009` (tab, `'\t'`)
121 - `U+000A` (LF, `'\n'`)
122 - `U+000D` (CR, `'\r'`)
124 Rust is a "free-form" language, meaning that all forms of whitespace serve only
125 to separate _tokens_ in the grammar, and have no semantic significance.
127 A Rust program has identical meaning if each whitespace element is replaced
128 with any other legal whitespace element, such as a single space character.
132 Tokens are primitive productions in the grammar defined by regular
133 (non-recursive) languages. "Simple" tokens are given in [string table
134 production](#string-table-productions) form, and occur in the rest of the
135 grammar as double-quoted strings. Other tokens have exact rules given.
139 A literal is an expression consisting of a single token, rather than a sequence
140 of tokens, that immediately and directly denotes the value it evaluates to,
141 rather than referring to it by name or some other evaluation rule. A literal is
142 a form of constant expression, so is evaluated (primarily) at compile time.
146 ##### Characters and strings
148 | | Example | `#` sets | Characters | Escapes |
149 |----------------------------------------------|-----------------|------------|-------------|---------------------|
150 | [Character](#character-literals) | `'H'` | `N/A` | All Unicode | [Quote](#quote-escapes) & [Byte](#byte-escapes) & [Unicode](#unicode-escapes) |
151 | [String](#string-literals) | `"hello"` | `N/A` | All Unicode | [Quote](#quote-escapes) & [Byte](#byte-escapes) & [Unicode](#unicode-escapes) |
152 | [Raw](#raw-string-literals) | `r#"hello"#` | `0...` | All Unicode | `N/A` |
153 | [Byte](#byte-literals) | `b'H'` | `N/A` | All ASCII | [Quote](#quote-escapes) & [Byte](#byte-escapes) |
154 | [Byte string](#byte-string-literals) | `b"hello"` | `N/A` | All ASCII | [Quote](#quote-escapes) & [Byte](#byte-escapes) |
155 | [Raw byte string](#raw-byte-string-literals) | `br#"hello"#` | `0...` | All ASCII | `N/A` |
161 | `\x7F` | 8-bit character code (exactly 2 digits) |
163 | `\r` | Carriage return |
168 ##### Unicode escapes
171 | `\u{7FFF}` | 24-bit Unicode character code (up to 6 digits) |
176 | `\'` | Single quote |
177 | `\"` | Double quote |
181 | [Number literals](#number-literals)`*` | Example | Exponentiation | Suffixes |
182 |----------------------------------------|---------|----------------|----------|
183 | Decimal integer | `98_222` | `N/A` | Integer suffixes |
184 | Hex integer | `0xff` | `N/A` | Integer suffixes |
185 | Octal integer | `0o77` | `N/A` | Integer suffixes |
186 | Binary integer | `0b1111_0000` | `N/A` | Integer suffixes |
187 | Floating-point | `123.0E+77` | `Optional` | Floating-point suffixes |
189 `*` All number literals allow `_` as a visual separator: `1_234.0E+18f64`
192 | Integer | Floating-point |
193 |---------|----------------|
194 | `u8`, `i8`, `u16`, `i16`, `u32`, `i32`, `u64`, `i64`, `isize`, `usize` | `f32`, `f64` |
196 #### Character and string literals
198 ##### Character literals
200 A _character literal_ is a single Unicode character enclosed within two
201 `U+0027` (single-quote) characters, with the exception of `U+0027` itself,
202 which must be _escaped_ by a preceding `U+005C` character (`\`).
204 ##### String literals
206 A _string literal_ is a sequence of any Unicode characters enclosed within two
207 `U+0022` (double-quote) characters, with the exception of `U+0022` itself,
208 which must be _escaped_ by a preceding `U+005C` character (`\`).
210 Line-break characters are allowed in string literals. Normally they represent
211 themselves (i.e. no translation), but as a special exception, when an unescaped
212 `U+005C` character (`\`) occurs immediately before the newline (`U+000A`), the
213 `U+005C` character, the newline, and all whitespace at the beginning of the
214 next line are ignored. Thus `a` and `b` are equal:
224 ##### Character escapes
226 Some additional _escapes_ are available in either character or non-raw string
227 literals. An escape starts with a `U+005C` (`\`) and continues with one of the
230 * An _8-bit code point escape_ starts with `U+0078` (`x`) and is
231 followed by exactly two _hex digits_. It denotes the Unicode code point
232 equal to the provided hex value.
233 * A _24-bit code point escape_ starts with `U+0075` (`u`) and is followed
234 by up to six _hex digits_ surrounded by braces `U+007B` (`{`) and `U+007D`
235 (`}`). It denotes the Unicode code point equal to the provided hex value.
236 * A _whitespace escape_ is one of the characters `U+006E` (`n`), `U+0072`
237 (`r`), or `U+0074` (`t`), denoting the Unicode values `U+000A` (LF),
238 `U+000D` (CR) or `U+0009` (HT) respectively.
239 * The _backslash escape_ is the character `U+005C` (`\`) which must be
240 escaped in order to denote *itself*.
242 ##### Raw string literals
244 Raw string literals do not process any escapes. They start with the character
245 `U+0072` (`r`), followed by zero or more of the character `U+0023` (`#`) and a
246 `U+0022` (double-quote) character. The _raw string body_ can contain any sequence
247 of Unicode characters and is terminated only by another `U+0022` (double-quote)
248 character, followed by the same number of `U+0023` (`#`) characters that preceded
249 the opening `U+0022` (double-quote) character.
251 All Unicode characters contained in the raw string body represent themselves,
252 the characters `U+0022` (double-quote) (except when followed by at least as
253 many `U+0023` (`#`) characters as were used to start the raw string literal) or
254 `U+005C` (`\`) do not have any special meaning.
256 Examples for string literals:
259 "foo"; r"foo"; // foo
260 "\"foo\""; r#""foo""#; // "foo"
263 r##"foo #"# bar"##; // foo #"# bar
265 "\x52"; "R"; r"R"; // R
266 "\\x52"; r"\x52"; // \x52
269 #### Byte and byte string literals
273 A _byte literal_ is a single ASCII character (in the `U+0000` to `U+007F`
274 range) or a single _escape_ preceded by the characters `U+0062` (`b`) and
275 `U+0027` (single-quote), and followed by the character `U+0027`. If the character
276 `U+0027` is present within the literal, it must be _escaped_ by a preceding
277 `U+005C` (`\`) character. It is equivalent to a `u8` unsigned 8-bit integer
280 ##### Byte string literals
282 A non-raw _byte string literal_ is a sequence of ASCII characters and _escapes_,
283 preceded by the characters `U+0062` (`b`) and `U+0022` (double-quote), and
284 followed by the character `U+0022`. If the character `U+0022` is present within
285 the literal, it must be _escaped_ by a preceding `U+005C` (`\`) character.
286 Alternatively, a byte string literal can be a _raw byte string literal_, defined
287 below. A byte string literal of length `n` is equivalent to a `&'static [u8; n]` borrowed fixed-sized array
288 of unsigned 8-bit integers.
290 Some additional _escapes_ are available in either byte or non-raw byte string
291 literals. An escape starts with a `U+005C` (`\`) and continues with one of the
294 * A _byte escape_ escape starts with `U+0078` (`x`) and is
295 followed by exactly two _hex digits_. It denotes the byte
296 equal to the provided hex value.
297 * A _whitespace escape_ is one of the characters `U+006E` (`n`), `U+0072`
298 (`r`), or `U+0074` (`t`), denoting the bytes values `0x0A` (ASCII LF),
299 `0x0D` (ASCII CR) or `0x09` (ASCII HT) respectively.
300 * The _backslash escape_ is the character `U+005C` (`\`) which must be
301 escaped in order to denote its ASCII encoding `0x5C`.
303 ##### Raw byte string literals
305 Raw byte string literals do not process any escapes. They start with the
306 character `U+0062` (`b`), followed by `U+0072` (`r`), followed by zero or more
307 of the character `U+0023` (`#`), and a `U+0022` (double-quote) character. The
308 _raw string body_ can contain any sequence of ASCII characters and is terminated
309 only by another `U+0022` (double-quote) character, followed by the same number of
310 `U+0023` (`#`) characters that preceded the opening `U+0022` (double-quote)
311 character. A raw byte string literal can not contain any non-ASCII byte.
313 All characters contained in the raw string body represent their ASCII encoding,
314 the characters `U+0022` (double-quote) (except when followed by at least as
315 many `U+0023` (`#`) characters as were used to start the raw string literal) or
316 `U+005C` (`\`) do not have any special meaning.
318 Examples for byte string literals:
321 b"foo"; br"foo"; // foo
322 b"\"foo\""; br#""foo""#; // "foo"
325 br##"foo #"# bar"##; // foo #"# bar
327 b"\x52"; b"R"; br"R"; // R
328 b"\\x52"; br"\x52"; // \x52
333 A _number literal_ is either an _integer literal_ or a _floating-point
334 literal_. The grammar for recognizing the two kinds of literals is mixed.
336 ##### Integer literals
338 An _integer literal_ has one of four forms:
340 * A _decimal literal_ starts with a *decimal digit* and continues with any
341 mixture of *decimal digits* and _underscores_.
342 * A _hex literal_ starts with the character sequence `U+0030` `U+0078`
343 (`0x`) and continues as any mixture of hex digits and underscores.
344 * An _octal literal_ starts with the character sequence `U+0030` `U+006F`
345 (`0o`) and continues as any mixture of octal digits and underscores.
346 * A _binary literal_ starts with the character sequence `U+0030` `U+0062`
347 (`0b`) and continues as any mixture of binary digits and underscores.
349 Like any literal, an integer literal may be followed (immediately,
350 without any spaces) by an _integer suffix_, which forcibly sets the
351 type of the literal. The integer suffix must be the name of one of the
352 integral types: `u8`, `i8`, `u16`, `i16`, `u32`, `i32`, `u64`, `i64`,
355 The type of an _unsuffixed_ integer literal is determined by type inference:
357 * If an integer type can be _uniquely_ determined from the surrounding
358 program context, the unsuffixed integer literal has that type.
360 * If the program context under-constrains the type, it defaults to the
361 signed 32-bit integer `i32`.
363 * If the program context over-constrains the type, it is considered a
366 Examples of integer literals of various forms:
373 0o70_i16; // type i16
374 0b1111_1111_1001_0000_i32; // type i32
375 0usize; // type usize
378 ##### Floating-point literals
380 A _floating-point literal_ has one of two forms:
382 * A _decimal literal_ followed by a period character `U+002E` (`.`). This is
383 optionally followed by another decimal literal, with an optional _exponent_.
384 * A single _decimal literal_ followed by an _exponent_.
386 Like integer literals, a floating-point literal may be followed by a
387 suffix, so long as the pre-suffix part does not end with `U+002E` (`.`).
388 The suffix forcibly sets the type of the literal. There are two valid
389 _floating-point suffixes_, `f32` and `f64` (the 32-bit and 64-bit floating point
390 types), which explicitly determine the type of the literal.
392 The type of an _unsuffixed_ floating-point literal is determined by
395 * If a floating-point type can be _uniquely_ determined from the
396 surrounding program context, the unsuffixed floating-point literal
399 * If the program context under-constrains the type, it defaults to `f64`.
401 * If the program context over-constrains the type, it is considered a
404 Examples of floating-point literals of various forms:
407 123.0f64; // type f64
410 12E+99_f64; // type f64
411 let x: f64 = 2.; // type f64
414 This last example is different because it is not possible to use the suffix
415 syntax with a floating point literal ending in a period. `2.f64` would attempt
416 to call a method named `f64` on `2`.
418 The representation semantics of floating-point numbers are described in
419 ["Machine Types"](#machine-types).
421 #### Boolean literals
423 The two values of the boolean type are written `true` and `false`.
427 Symbols are a general class of printable [tokens](#tokens) that play structural
428 roles in a variety of grammar productions. They are a
429 set of remaining miscellaneous printable tokens that do not
430 otherwise appear as [unary operators](#unary-operator-expressions), [binary
431 operators](#binary-operator-expressions), or [keywords][keywords].
432 They are catalogued in [the Symbols section][symbols] of the Grammar document.
434 [symbols]: grammar.html#symbols
439 A _path_ is a sequence of one or more path components _logically_ separated by
440 a namespace qualifier (`::`). If a path consists of only one component, it may
441 refer to either an [item](#items) or a [variable](#variables) in a local control
442 scope. If a path has multiple components, it refers to an item.
444 Every item has a _canonical path_ within its crate, but the path naming an item
445 is only meaningful within a given crate. There is no global namespace across
446 crates; an item's canonical path merely identifies it within the crate.
448 Two examples of simple paths consisting of only identifier components:
455 Path components are usually [identifiers](#identifiers), but they may
456 also include angle-bracket-enclosed lists of type arguments. In
457 [expression](#expressions) context, the type argument list is given
458 after a `::` namespace qualifier in order to disambiguate it from a
459 relational expression involving the less-than symbol (`<`). In type
460 expression context, the final namespace qualifier is omitted.
462 Two examples of paths with type arguments:
465 # struct HashMap<K, V>(K,V);
467 # fn id<T>(t: T) -> T { t }
468 type T = HashMap<i32,String>; // Type arguments used in a type expression
469 let x = id::<i32>(10); // Type arguments used in a call expression
473 Paths can be denoted with various leading qualifiers to change the meaning of
476 * Paths starting with `::` are considered to be global paths where the
477 components of the path start being resolved from the crate root. Each
478 identifier in the path must resolve to an item.
486 ::a::foo(); // call a's foo function
492 * Paths starting with the keyword `super` begin resolution relative to the
493 parent module. Each further identifier must resolve to an item.
501 super::a::foo(); // call a's foo function
507 * Paths starting with the keyword `self` begin resolution relative to the
508 current module. Each further identifier must resolve to an item.
518 Additionally keyword `super` may be repeated several times after the first
519 `super` or `self` to refer to ancestor modules.
528 super::super::foo(); // call a's foo function
529 self::super::super::foo(); // call a's foo function
539 A number of minor features of Rust are not central enough to have their own
540 syntax, and yet are not implementable as functions. Instead, they are given
541 names, and invoked through a consistent syntax: `some_extension!(...)`.
543 Users of `rustc` can define new syntax extensions in two ways:
545 * [Compiler plugins][plugin] can include arbitrary Rust code that
546 manipulates syntax trees at compile time. Note that the interface
547 for compiler plugins is considered highly unstable.
549 * [Macros](book/macros.html) define new syntax in a higher-level,
554 `macro_rules` allows users to define syntax extension in a declarative way. We
555 call such extensions "macros by example" or simply "macros" — to be distinguished
556 from the "procedural macros" defined in [compiler plugins][plugin].
558 Currently, macros can expand to expressions, statements, items, or patterns.
560 (A `sep_token` is any token other than `*` and `+`. A `non_special_token` is
561 any token other than a delimiter or `$`.)
563 The macro expander looks up macro invocations by name, and tries each macro
564 rule in turn. It transcribes the first successful match. Matching and
565 transcription are closely related to each other, and we will describe them
570 The macro expander matches and transcribes every token that does not begin with
571 a `$` literally, including delimiters. For parsing reasons, delimiters must be
572 balanced, but they are otherwise not special.
574 In the matcher, `$` _name_ `:` _designator_ matches the nonterminal in the Rust
575 syntax named by _designator_. Valid designators are:
577 * `item`: an [item](#items)
578 * `block`: a [block](#block-expressions)
579 * `stmt`: a [statement](#statements)
580 * `pat`: a [pattern](#match-expressions)
581 * `expr`: an [expression](#expressions)
582 * `ty`: a [type](#types)
583 * `ident`: an [identifier](#identifiers)
584 * `path`: a [path](#paths)
585 * `tt`: either side of the `=>` in macro rules
586 * `meta`: the contents of an [attribute](#attributes)
588 In the transcriber, the
589 designator is already known, and so only the name of a matched nonterminal comes
590 after the dollar sign.
592 In both the matcher and transcriber, the Kleene star-like operator indicates
593 repetition. The Kleene star operator consists of `$` and parentheses, optionally
594 followed by a separator token, followed by `*` or `+`. `*` means zero or more
595 repetitions, `+` means at least one repetition. The parentheses are not matched or
596 transcribed. On the matcher side, a name is bound to _all_ of the names it
597 matches, in a structure that mimics the structure of the repetition encountered
598 on a successful match. The job of the transcriber is to sort that structure
601 The rules for transcription of these repetitions are called "Macro By Example".
602 Essentially, one "layer" of repetition is discharged at a time, and all of them
603 must be discharged by the time a name is transcribed. Therefore, `( $( $i:ident
604 ),* ) => ( $i )` is an invalid macro, but `( $( $i:ident ),* ) => ( $( $i:ident
605 ),* )` is acceptable (if trivial).
607 When Macro By Example encounters a repetition, it examines all of the `$`
608 _name_ s that occur in its body. At the "current layer", they all must repeat
609 the same number of times, so ` ( $( $i:ident ),* ; $( $j:ident ),* ) => ( $(
610 ($i,$j) ),* )` is valid if given the argument `(a,b,c ; d,e,f)`, but not
611 `(a,b,c ; d,e)`. The repetition walks through the choices at that layer in
612 lockstep, so the former input transcribes to `(a,d), (b,e), (c,f)`.
614 Nested repetitions are allowed.
616 ### Parsing limitations
618 The parser used by the macro system is reasonably powerful, but the parsing of
619 Rust syntax is restricted in two ways:
621 1. Macro definitions are required to include suitable separators after parsing
622 expressions and other bits of the Rust grammar. This implies that
623 a macro definition like `$i:expr [ , ]` is not legal, because `[` could be part
624 of an expression. A macro definition like `$i:expr,` or `$i:expr;` would be legal,
625 however, because `,` and `;` are legal separators. See [RFC 550] for more information.
626 2. The parser must have eliminated all ambiguity by the time it reaches a `$`
627 _name_ `:` _designator_. This requirement most often affects name-designator
628 pairs when they occur at the beginning of, or immediately after, a `$(...)*`;
629 requiring a distinctive token in front can solve the problem.
631 [RFC 550]: https://github.com/rust-lang/rfcs/blob/master/text/0550-macro-future-proofing.md
633 # Crates and source files
635 Although Rust, like any other language, can be implemented by an interpreter as
636 well as a compiler, the only existing implementation is a compiler,
638 always been designed to be compiled. For these reasons, this section assumes a
641 Rust's semantics obey a *phase distinction* between compile-time and
642 run-time.[^phase-distinction] Semantic rules that have a *static
643 interpretation* govern the success or failure of compilation, while
645 that have a *dynamic interpretation* govern the behavior of the program at
648 [^phase-distinction]: This distinction would also exist in an interpreter.
649 Static checks like syntactic analysis, type checking, and lints should
650 happen before the program is executed regardless of when it is executed.
652 The compilation model centers on artifacts called _crates_. Each compilation
653 processes a single crate in source form, and if successful, produces a single
654 crate in binary form: either an executable or some sort of
655 library.[^cratesourcefile]
657 [^cratesourcefile]: A crate is somewhat analogous to an *assembly* in the
658 ECMA-335 CLI model, a *library* in the SML/NJ Compilation Manager, a *unit*
659 in the Owens and Flatt module system, or a *configuration* in Mesa.
661 A _crate_ is a unit of compilation and linking, as well as versioning,
662 distribution and runtime loading. A crate contains a _tree_ of nested
663 [module](#modules) scopes. The top level of this tree is a module that is
664 anonymous (from the point of view of paths within the module) and any item
665 within a crate has a canonical [module path](#paths) denoting its location
666 within the crate's module tree.
668 The Rust compiler is always invoked with a single source file as input, and
669 always produces a single output crate. The processing of that source file may
670 result in other source files being loaded as modules. Source files have the
673 A Rust source file describes a module, the name and location of which —
674 in the module tree of the current crate — are defined from outside the
675 source file: either by an explicit `mod_item` in a referencing source file, or
676 by the name of the crate itself. Every source file is a module, but not every
677 module needs its own source file: [module definitions](#modules) can be nested
680 Each source file contains a sequence of zero or more `item` definitions, and
681 may optionally begin with any number of [attributes](#items-and-attributes)
682 that apply to the containing module, most of which influence the behavior of
683 the compiler. The anonymous crate module can have additional attributes that
684 apply to the crate as a whole.
687 // Specify the crate name.
688 #![crate_name = "projx"]
690 // Specify the type of output artifact.
691 #![crate_type = "lib"]
693 // Turn on a warning.
694 // This can be done in any module, not just the anonymous crate module.
695 #![warn(non_camel_case_types)]
698 A crate that contains a `main` function can be compiled to an executable. If a
699 `main` function is present, its return type must be `()`
700 ("[unit](#tuple-types)") and it must take no arguments.
702 # Items and attributes
704 Crates contain [items](#items), each of which may have some number of
705 [attributes](#attributes) attached to it.
709 An _item_ is a component of a crate. Items are organized within a crate by a
710 nested set of [modules](#modules). Every crate has a single "outermost"
711 anonymous module; all further items within the crate have [paths](#paths)
712 within the module tree of the crate.
714 Items are entirely determined at compile-time, generally remain fixed during
715 execution, and may reside in read-only memory.
717 There are several kinds of item:
719 * [`extern crate` declarations](#extern-crate-declarations)
720 * [`use` declarations](#use-declarations)
721 * [modules](#modules)
722 * [functions](#functions)
723 * [type definitions](grammar.html#type-definitions)
724 * [structs](#structs)
725 * [enumerations](#enumerations)
726 * [constant items](#constant-items)
727 * [static items](#static-items)
729 * [implementations](#implementations)
731 Some items form an implicit scope for the declaration of sub-items. In other
732 words, within a function or module, declarations of items can (in many cases)
733 be mixed with the statements, control blocks, and similar artifacts that
734 otherwise compose the item body. The meaning of these scoped items is the same
735 as if the item was declared outside the scope — it is still a static item
736 — except that the item's *path name* within the module namespace is
737 qualified by the name of the enclosing item, or is private to the enclosing
738 item (in the case of functions). The grammar specifies the exact locations in
739 which sub-item declarations may appear.
743 All items except modules, constants and statics may be *parameterized* by type.
744 Type parameters are given as a comma-separated list of identifiers enclosed in
745 angle brackets (`<...>`), after the name of the item and before its definition.
746 The type parameters of an item are considered "part of the name", not part of
747 the type of the item. A referencing [path](#paths) must (in principle) provide
748 type arguments as a list of comma-separated types enclosed within angle
749 brackets, in order to refer to the type-parameterized item. In practice, the
750 type-inference system can usually infer such argument types from context. There
751 are no general type-parametric types, only type-parametric items. That is, Rust
752 has no notion of type abstraction: there are no higher-ranked (or "forall") types
753 abstracted over other types, though higher-ranked types do exist for lifetimes.
757 A module is a container for zero or more [items](#items).
759 A _module item_ is a module, surrounded in braces, named, and prefixed with the
760 keyword `mod`. A module item introduces a new, named module into the tree of
761 modules making up a crate. Modules can nest arbitrarily.
763 An example of a module:
767 type Complex = (f64, f64);
768 fn sin(f: f64) -> f64 {
772 fn cos(f: f64) -> f64 {
776 fn tan(f: f64) -> f64 {
783 Modules and types share the same namespace. Declaring a named type with
784 the same name as a module in scope is forbidden: that is, a type definition,
785 trait, struct, enumeration, or type parameter can't shadow the name of a module
786 in scope, or vice versa.
788 A module without a body is loaded from an external file, by default with the
789 same name as the module, plus the `.rs` extension. When a nested submodule is
790 loaded from an external file, it is loaded from a subdirectory path that
791 mirrors the module hierarchy.
794 // Load the `vec` module from `vec.rs`
798 // Load the `local_data` module from `thread/local_data.rs`
799 // or `thread/local_data/mod.rs`.
804 The directories and files used for loading external file modules can be
805 influenced with the `path` attribute.
808 #[path = "thread_files"]
810 // Load the `local_data` module from `thread_files/tls.rs`
816 #### Extern crate declarations
818 An _`extern crate` declaration_ specifies a dependency on an external crate.
819 The external crate is then bound into the declaring scope as the `ident`
820 provided in the `extern_crate_decl`.
822 The external crate is resolved to a specific `soname` at compile time, and a
823 runtime linkage requirement to that `soname` is passed to the linker for
824 loading at runtime. The `soname` is resolved at compile time by scanning the
825 compiler's library path and matching the optional `crateid` provided against
826 the `crateid` attributes that were declared on the external crate when it was
827 compiled. If no `crateid` is provided, a default `name` attribute is assumed,
828 equal to the `ident` given in the `extern_crate_decl`.
830 Three examples of `extern crate` declarations:
835 extern crate std; // equivalent to: extern crate std as std;
837 extern crate std as ruststd; // linking to 'std' under another name
840 #### Use declarations
842 A _use declaration_ creates one or more local name bindings synonymous with
843 some other [path](#paths). Usually a `use` declaration is used to shorten the
844 path required to refer to a module item. These declarations may appear in
845 [modules](#modules) and [blocks](grammar.html#block-expressions), usually at the top.
847 > **Note**: Unlike in many languages,
848 > `use` declarations in Rust do *not* declare linkage dependency with external crates.
849 > Rather, [`extern crate` declarations](#extern-crate-declarations) declare linkage dependencies.
851 Use declarations support a number of convenient shortcuts:
853 * Rebinding the target name as a new local name, using the syntax `use p::q::r as x;`
854 * Simultaneously binding a list of paths differing only in their final element,
855 using the glob-like brace syntax `use a::b::{c,d,e,f};`
856 * Binding all paths matching a given prefix, using the asterisk wildcard syntax
858 * Simultaneously binding a list of paths differing only in their final element
859 and their immediate parent module, using the `self` keyword, such as
860 `use a::b::{self, c, d};`
862 An example of `use` declarations:
865 use std::option::Option::{Some, None};
866 use std::collections::hash_map::{self, HashMap};
869 fn bar(map1: HashMap<String, usize>, map2: hash_map::HashMap<String, usize>){}
872 // Equivalent to 'foo(vec![std::option::Option::Some(1.0f64),
873 // std::option::Option::None]);'
874 foo(vec![Some(1.0f64), None]);
876 // Both `hash_map` and `HashMap` are in scope.
877 let map1 = HashMap::new();
878 let map2 = hash_map::HashMap::new();
883 Like items, `use` declarations are private to the containing module, by
884 default. Also like items, a `use` declaration can be public, if qualified by
885 the `pub` keyword. Such a `use` declaration serves to _re-export_ a name. A
886 public `use` declaration can therefore _redirect_ some public name to a
887 different target definition: even a definition with a private canonical path,
888 inside a different module. If a sequence of such redirections form a cycle or
889 cannot be resolved unambiguously, they represent a compile-time error.
891 An example of re-exporting:
896 pub use quux::foo::{bar, baz};
905 In this example, the module `quux` re-exports two public names defined in
908 Also note that the paths contained in `use` items are relative to the crate
909 root. So, in the previous example, the `use` refers to `quux::foo::{bar,
910 baz}`, and not simply to `foo::{bar, baz}`. This also means that top-level
911 module declarations should be at the crate root if direct usage of the declared
912 modules within `use` items is desired. It is also possible to use `self` and
913 `super` at the beginning of a `use` item to refer to the current and direct
914 parent modules respectively. All rules regarding accessing declared modules in
915 `use` declarations apply to both module declarations and `extern crate`
918 An example of what will and will not work for `use` items:
921 # #![allow(unused_imports)]
922 use foo::baz::foobaz; // good: foo is at the root of the crate
930 use foo::example::iter; // good: foo is at crate root
931 // use example::iter; // bad: example is not at the crate root
932 use self::baz::foobaz; // good: self refers to module 'foo'
933 use foo::bar::foobar; // good: foo is at crate root
940 use super::bar::foobar; // good: super refers to module 'foo'
950 A _function item_ defines a sequence of [statements](#statements) and a
951 final [expression](#expressions), along with a name and a set of
952 parameters. Other than a name, all these are optional.
953 Functions are declared with the keyword `fn`. Functions may declare a
954 set of *input* [*variables*](#variables) as parameters, through which the caller
955 passes arguments into the function, and the *output* [*type*](#types)
956 of the value the function will return to its caller on completion.
958 A function may also be copied into a first-class *value*, in which case the
959 value has the corresponding [*function type*](#function-types), and can be used
960 otherwise exactly as a function item (with a minor additional cost of calling
961 the function indirectly).
963 Every control path in a function logically ends with a `return` expression or a
964 diverging expression. If the outermost block of a function has a
965 value-producing expression in its final-expression position, that expression is
966 interpreted as an implicit `return` expression applied to the final-expression.
968 An example of a function:
971 fn add(x: i32, y: i32) -> i32 {
976 As with `let` bindings, function arguments are irrefutable patterns, so any
977 pattern that is valid in a let binding is also valid as an argument.
980 fn first((value, _): (i32, i32)) -> i32 { value }
984 #### Generic functions
986 A _generic function_ allows one or more _parameterized types_ to appear in its
987 signature. Each type parameter must be explicitly declared in an
988 angle-bracket-enclosed and comma-separated list, following the function name.
991 // foo is generic over A and B
993 fn foo<A, B>(x: A, y: B) {
996 Inside the function signature and body, the name of the type parameter can be
997 used as a type name. [Trait](#traits) bounds can be specified for type parameters
998 to allow methods with that trait to be called on values of that type. This is
999 specified using the `where` syntax:
1002 fn foo<T>(x: T) where T: Debug {
1005 When a generic function is referenced, its type is instantiated based on the
1006 context of the reference. For example, calling the `foo` function here:
1009 use std::fmt::Debug;
1011 fn foo<T>(x: &[T]) where T: Debug {
1019 will instantiate type parameter `T` with `i32`.
1021 The type parameters can also be explicitly supplied in a trailing
1022 [path](#paths) component after the function name. This might be necessary if
1023 there is not sufficient context to determine the type parameters. For example,
1024 `mem::size_of::<u32>() == 4`.
1026 #### Diverging functions
1028 A special kind of function can be declared with a `!` character where the
1029 output type would normally be. For example:
1032 fn my_err(s: &str) -> ! {
1038 We call such functions "diverging" because they never return a value to the
1039 caller. Every control path in a diverging function must end with a `panic!()` or
1040 a call to another diverging function on every control path. The `!` annotation
1041 does *not* denote a type.
1043 It might be necessary to declare a diverging function because as mentioned
1044 previously, the typechecker checks that every control path in a function ends
1045 with a [`return`](#return-expressions) or diverging expression. So, if `my_err`
1046 were declared without the `!` annotation, the following code would not
1050 # fn my_err(s: &str) -> ! { panic!() }
1052 fn f(i: i32) -> i32 {
1057 my_err("Bad number!");
1062 This will not compile without the `!` annotation on `my_err`, since the `else`
1063 branch of the conditional in `f` does not return an `i32`, as required by the
1064 signature of `f`. Adding the `!` annotation to `my_err` informs the
1065 typechecker that, should control ever enter `my_err`, no further type judgments
1066 about `f` need to hold, since control will never resume in any context that
1067 relies on those judgments. Thus the return type on `f` only needs to reflect
1068 the `if` branch of the conditional.
1070 #### Extern functions
1072 Extern functions are part of Rust's foreign function interface, providing the
1073 opposite functionality to [external blocks](#external-blocks). Whereas
1074 external blocks allow Rust code to call foreign code, extern functions with
1075 bodies defined in Rust code _can be called by foreign code_. They are defined
1076 in the same way as any other Rust function, except that they have the `extern`
1080 // Declares an extern fn, the ABI defaults to "C"
1081 extern fn new_i32() -> i32 { 0 }
1083 // Declares an extern fn with "stdcall" ABI
1084 extern "stdcall" fn new_i32_stdcall() -> i32 { 0 }
1087 Unlike normal functions, extern fns have type `extern "ABI" fn()`. This is the
1088 same type as the functions declared in an extern block.
1091 # extern fn new_i32() -> i32 { 0 }
1092 let fptr: extern "C" fn() -> i32 = new_i32;
1095 Extern functions may be called directly from Rust code as Rust uses large,
1096 contiguous stack segments like C.
1100 A _type alias_ defines a new name for an existing [type](#types). Type
1101 aliases are declared with the keyword `type`. Every value has a single,
1102 specific type, but may implement several different traits, or be compatible with
1103 several different type constraints.
1105 For example, the following defines the type `Point` as a synonym for the type
1106 `(u8, u8)`, the type of pairs of unsigned 8 bit integers:
1109 type Point = (u8, u8);
1110 let p: Point = (41, 68);
1115 A _struct_ is a nominal [struct type](#struct-types) defined with the
1118 An example of a `struct` item and its use:
1121 struct Point {x: i32, y: i32}
1122 let p = Point {x: 10, y: 11};
1126 A _tuple struct_ is a nominal [tuple type](#tuple-types), also defined with
1127 the keyword `struct`. For example:
1130 struct Point(i32, i32);
1131 let p = Point(10, 11);
1132 let px: i32 = match p { Point(x, _) => x };
1135 A _unit-like struct_ is a struct without any fields, defined by leaving off
1136 the list of fields entirely. Such a struct implicitly defines a constant of
1137 its type with the same name. For example:
1140 # #![feature(braced_empty_structs)]
1142 let c = [Cookie, Cookie {}, Cookie, Cookie {}];
1148 # #![feature(braced_empty_structs)]
1150 const Cookie: Cookie = Cookie {};
1151 let c = [Cookie, Cookie {}, Cookie, Cookie {}];
1154 The precise memory layout of a struct is not specified. One can specify a
1155 particular layout using the [`repr` attribute](#ffi-attributes).
1159 An _enumeration_ is a simultaneous definition of a nominal [enumerated
1160 type](#enumerated-types) as well as a set of *constructors*, that can be used
1161 to create or pattern-match values of the corresponding enumerated type.
1163 Enumerations are declared with the keyword `enum`.
1165 An example of an `enum` item and its use:
1173 let mut a: Animal = Animal::Dog;
1177 Enumeration constructors can have either named or unnamed fields:
1182 Cat { name: String, weight: f64 },
1185 let mut a: Animal = Animal::Dog("Cocoa".to_string(), 37.2);
1186 a = Animal::Cat { name: "Spotty".to_string(), weight: 2.7 };
1189 In this example, `Cat` is a _struct-like enum variant_,
1190 whereas `Dog` is simply called an enum variant.
1192 Enums have a discriminant. You can assign them explicitly:
1200 If a discriminant isn't assigned, they start at zero, and add one for each
1203 You can cast an enum to get this value:
1206 # enum Foo { Bar = 123 }
1207 let x = Foo::Bar as u32; // x is now 123u32
1210 This only works as long as none of the variants have data attached. If
1211 it were `Bar(i32)`, this is disallowed.
1215 A *constant item* is a named _constant value_ which is not associated with a
1216 specific memory location in the program. Constants are essentially inlined
1217 wherever they are used, meaning that they are copied directly into the relevant
1218 context when used. References to the same constant are not necessarily
1219 guaranteed to refer to the same memory address.
1221 Constant values must not have destructors, and otherwise permit most forms of
1222 data. Constants may refer to the address of other constants, in which case the
1223 address will have the `static` lifetime. The compiler is, however, still at
1224 liberty to translate the constant many times, so the address referred to may not
1227 Constants must be explicitly typed. The type may be `bool`, `char`, a number, or
1228 a type derived from those primitive types. The derived types are references with
1229 the `static` lifetime, fixed-size arrays, tuples, enum variants, and structs.
1232 const BIT1: u32 = 1 << 0;
1233 const BIT2: u32 = 1 << 1;
1235 const BITS: [u32; 2] = [BIT1, BIT2];
1236 const STRING: &'static str = "bitstring";
1238 struct BitsNStrings<'a> {
1243 const BITS_N_STRINGS: BitsNStrings<'static> = BitsNStrings {
1251 A *static item* is similar to a *constant*, except that it represents a precise
1252 memory location in the program. A static is never "inlined" at the usage site,
1253 and all references to it refer to the same memory location. Static items have
1254 the `static` lifetime, which outlives all other lifetimes in a Rust program.
1255 Static items may be placed in read-only memory if they do not contain any
1256 interior mutability.
1258 Statics may contain interior mutability through the `UnsafeCell` language item.
1259 All access to a static is safe, but there are a number of restrictions on
1262 * Statics may not contain any destructors.
1263 * The types of static values must ascribe to `Sync` to allow thread-safe access.
1264 * Statics may not refer to other statics by value, only by reference.
1265 * Constants cannot refer to statics.
1267 Constants should in general be preferred over statics, unless large amounts of
1268 data are being stored, or single-address and mutability properties are required.
1270 #### Mutable statics
1272 If a static item is declared with the `mut` keyword, then it is allowed to
1273 be modified by the program. One of Rust's goals is to make concurrency bugs
1274 hard to run into, and this is obviously a very large source of race conditions
1275 or other bugs. For this reason, an `unsafe` block is required when either
1276 reading or writing a mutable static variable. Care should be taken to ensure
1277 that modifications to a mutable static are safe with respect to other threads
1278 running in the same process.
1280 Mutable statics are still very useful, however. They can be used with C
1281 libraries and can also be bound from C libraries (in an `extern` block).
1284 # fn atomic_add(_: &mut u32, _: u32) -> u32 { 2 }
1286 static mut LEVELS: u32 = 0;
1288 // This violates the idea of no shared state, and this doesn't internally
1289 // protect against races, so this function is `unsafe`
1290 unsafe fn bump_levels_unsafe1() -> u32 {
1296 // Assuming that we have an atomic_add function which returns the old value,
1297 // this function is "safe" but the meaning of the return value may not be what
1298 // callers expect, so it's still marked as `unsafe`
1299 unsafe fn bump_levels_unsafe2() -> u32 {
1300 return atomic_add(&mut LEVELS, 1);
1304 Mutable statics have the same restrictions as normal statics, except that the
1305 type of the value is not required to ascribe to `Sync`.
1309 A _trait_ describes an abstract interface that types can
1310 implement. This interface consists of associated items, which come in
1317 Associated functions whose first parameter is named `self` are called
1318 methods and may be invoked using `.` notation (e.g., `x.foo()`).
1320 All traits define an implicit type parameter `Self` that refers to
1321 "the type that is implementing this interface". Traits may also
1322 contain additional type parameters. These type parameters (including
1323 `Self`) may be constrained by other traits and so forth as usual.
1325 Trait bounds on `Self` are considered "supertraits". These are
1326 required to be acyclic. Supertraits are somewhat different from other
1327 constraints in that they affect what methods are available in the
1328 vtable when the trait is used as a [trait object](#trait-objects).
1330 Traits are implemented for specific types through separate
1331 [implementations](#implementations).
1333 Consider the following trait:
1336 # type Surface = i32;
1337 # type BoundingBox = i32;
1339 fn draw(&self, Surface);
1340 fn bounding_box(&self) -> BoundingBox;
1344 This defines a trait with two methods. All values that have
1345 [implementations](#implementations) of this trait in scope can have their
1346 `draw` and `bounding_box` methods called, using `value.bounding_box()`
1347 [syntax](#method-call-expressions).
1349 Traits can include default implementations of methods, as in:
1354 fn baz(&self) { println!("We called baz."); }
1358 Here the `baz` method has a default implementation, so types that implement
1359 `Foo` need only implement `bar`. It is also possible for implementing types
1360 to override a method that has a default implementation.
1362 Type parameters can be specified for a trait to make it generic. These appear
1363 after the trait name, using the same syntax used in [generic
1364 functions](#generic-functions).
1368 fn len(&self) -> u32;
1369 fn elt_at(&self, n: u32) -> T;
1370 fn iter<F>(&self, F) where F: Fn(T);
1374 It is also possible to define associated types for a trait. Consider the
1375 following example of a `Container` trait. Notice how the type is available
1376 for use in the method signatures:
1382 fn insert(&mut self, Self::E);
1386 In order for a type to implement this trait, it must not only provide
1387 implementations for every method, but it must specify the type `E`. Here's
1388 an implementation of `Container` for the standard library type `Vec`:
1393 # fn empty() -> Self;
1394 # fn insert(&mut self, Self::E);
1396 impl<T> Container for Vec<T> {
1398 fn empty() -> Vec<T> { Vec::new() }
1399 fn insert(&mut self, x: T) { self.push(x); }
1403 Generic functions may use traits as _bounds_ on their type parameters. This
1404 will have two effects:
1406 - Only types that have the trait may instantiate the parameter.
1407 - Within the generic function, the methods of the trait can be
1408 called on values that have the parameter's type.
1413 # type Surface = i32;
1414 # trait Shape { fn draw(&self, Surface); }
1415 fn draw_twice<T: Shape>(surface: Surface, sh: T) {
1421 Traits also define a [trait object](#trait-objects) with the same
1422 name as the trait. Values of this type are created by coercing from a
1423 pointer of some specific type to a pointer of trait type. For example,
1424 `&T` could be coerced to `&Shape` if `T: Shape` holds (and similarly
1425 for `Box<T>`). This coercion can either be implicit or
1426 [explicit](#type-cast-expressions). Here is an example of an explicit
1431 impl Shape for i32 { }
1432 let mycircle = 0i32;
1433 let myshape: Box<Shape> = Box::new(mycircle) as Box<Shape>;
1436 The resulting value is a box containing the value that was cast, along with
1437 information that identifies the methods of the implementation that was used.
1438 Values with a trait type can have [methods called](#method-call-expressions) on
1439 them, for any method in the trait, and can be used to instantiate type
1440 parameters that are bounded by the trait.
1442 Trait methods may be static, which means that they lack a `self` argument.
1443 This means that they can only be called with function call syntax (`f(x)`) and
1444 not method call syntax (`obj.f()`). The way to refer to the name of a static
1445 method is to qualify it with the trait name, treating the trait name like a
1446 module. For example:
1450 fn from_i32(n: i32) -> Self;
1453 fn from_i32(n: i32) -> f64 { n as f64 }
1455 let x: f64 = Num::from_i32(42);
1458 Traits may inherit from other traits. Consider the following example:
1461 trait Shape { fn area(&self) -> f64; }
1462 trait Circle : Shape { fn radius(&self) -> f64; }
1465 The syntax `Circle : Shape` means that types that implement `Circle` must also
1466 have an implementation for `Shape`. Multiple supertraits are separated by `+`,
1467 `trait Circle : Shape + PartialEq { }`. In an implementation of `Circle` for a
1468 given type `T`, methods can refer to `Shape` methods, since the typechecker
1469 checks that any type with an implementation of `Circle` also has an
1470 implementation of `Shape`:
1475 trait Shape { fn area(&self) -> f64; }
1476 trait Circle : Shape { fn radius(&self) -> f64; }
1477 impl Shape for Foo {
1478 fn area(&self) -> f64 {
1482 impl Circle for Foo {
1483 fn radius(&self) -> f64 {
1484 println!("calling area: {}", self.area());
1494 In type-parameterized functions, methods of the supertrait may be called on
1495 values of subtrait-bound type parameters. Referring to the previous example of
1496 `trait Circle : Shape`:
1499 # trait Shape { fn area(&self) -> f64; }
1500 # trait Circle : Shape { fn radius(&self) -> f64; }
1501 fn radius_times_area<T: Circle>(c: T) -> f64 {
1502 // `c` is both a Circle and a Shape
1503 c.radius() * c.area()
1507 Likewise, supertrait methods may also be called on trait objects.
1510 # trait Shape { fn area(&self) -> f64; }
1511 # trait Circle : Shape { fn radius(&self) -> f64; }
1512 # impl Shape for i32 { fn area(&self) -> f64 { 0.0 } }
1513 # impl Circle for i32 { fn radius(&self) -> f64 { 0.0 } }
1514 # let mycircle = 0i32;
1515 let mycircle = Box::new(mycircle) as Box<Circle>;
1516 let nonsense = mycircle.radius() * mycircle.area();
1521 An _implementation_ is an item that implements a [trait](#traits) for a
1524 Implementations are defined with the keyword `impl`.
1527 # #[derive(Copy, Clone)]
1528 # struct Point {x: f64, y: f64};
1529 # type Surface = i32;
1530 # struct BoundingBox {x: f64, y: f64, width: f64, height: f64};
1531 # trait Shape { fn draw(&self, Surface); fn bounding_box(&self) -> BoundingBox; }
1532 # fn do_draw_circle(s: Surface, c: Circle) { }
1538 impl Copy for Circle {}
1540 impl Clone for Circle {
1541 fn clone(&self) -> Circle { *self }
1544 impl Shape for Circle {
1545 fn draw(&self, s: Surface) { do_draw_circle(s, *self); }
1546 fn bounding_box(&self) -> BoundingBox {
1547 let r = self.radius;
1549 x: self.center.x - r,
1550 y: self.center.y - r,
1558 It is possible to define an implementation without referring to a trait. The
1559 methods in such an implementation can only be used as direct calls on the values
1560 of the type that the implementation targets. In such an implementation, the
1561 trait type and `for` after `impl` are omitted. Such implementations are limited
1562 to nominal types (enums, structs, trait objects), and the implementation must
1563 appear in the same crate as the `self` type:
1566 struct Point {x: i32, y: i32}
1570 println!("Point is at ({}, {})", self.x, self.y);
1574 let my_point = Point {x: 10, y:11};
1578 When a trait _is_ specified in an `impl`, all methods declared as part of the
1579 trait must be implemented, with matching types and type parameter counts.
1581 An implementation can take type parameters, which can be different from the
1582 type parameters taken by the trait it implements. Implementation parameters
1583 are written after the `impl` keyword.
1586 # trait Seq<T> { fn dummy(&self, _: T) { } }
1587 impl<T> Seq<T> for Vec<T> {
1590 impl Seq<bool> for u32 {
1591 /* Treat the integer as a sequence of bits */
1597 External blocks form the basis for Rust's foreign function interface.
1598 Declarations in an external block describe symbols in external, non-Rust
1601 Functions within external blocks are declared in the same way as other Rust
1602 functions, with the exception that they may not have a body and are instead
1603 terminated by a semicolon.
1605 Functions within external blocks may be called by Rust code, just like
1606 functions defined in Rust. The Rust compiler automatically translates between
1607 the Rust ABI and the foreign ABI.
1609 A number of [attributes](#attributes) control the behavior of external blocks.
1611 By default external blocks assume that the library they are calling uses the
1612 standard C "cdecl" ABI. Other ABIs may be specified using an `abi` string, as
1616 // Interface to the Windows API
1617 extern "stdcall" { }
1620 The `link` attribute allows the name of the library to be specified. When
1621 specified the compiler will attempt to link against the native library of the
1625 #[link(name = "crypto")]
1629 The type of a function declared in an extern block is `extern "abi" fn(A1, ...,
1630 An) -> R`, where `A1...An` are the declared types of its arguments and `R` is
1631 the declared return type.
1633 It is valid to add the `link` attribute on an empty extern block. You can use
1634 this to satisfy the linking requirements of extern blocks elsewhere in your code
1635 (including upstream crates) instead of adding the attribute to each extern block.
1637 ## Visibility and Privacy
1639 These two terms are often used interchangeably, and what they are attempting to
1640 convey is the answer to the question "Can this item be used at this location?"
1642 Rust's name resolution operates on a global hierarchy of namespaces. Each level
1643 in the hierarchy can be thought of as some item. The items are one of those
1644 mentioned above, but also include external crates. Declaring or defining a new
1645 module can be thought of as inserting a new tree into the hierarchy at the
1646 location of the definition.
1648 To control whether interfaces can be used across modules, Rust checks each use
1649 of an item to see whether it should be allowed or not. This is where privacy
1650 warnings are generated, or otherwise "you used a private item of another module
1651 and weren't allowed to."
1653 By default, everything in Rust is *private*, with one exception. Enum variants
1654 in a `pub` enum are also public by default. When an item is declared as `pub`,
1655 it can be thought of as being accessible to the outside world. For example:
1659 // Declare a private struct
1662 // Declare a public struct with a private field
1667 // Declare a public enum with two public variants
1669 PubliclyAccessibleState,
1670 PubliclyAccessibleState2,
1674 With the notion of an item being either public or private, Rust allows item
1675 accesses in two cases:
1677 1. If an item is public, then it can be used externally through any of its
1679 2. If an item is private, it may be accessed by the current module and its
1682 These two cases are surprisingly powerful for creating module hierarchies
1683 exposing public APIs while hiding internal implementation details. To help
1684 explain, here's a few use cases and what they would entail:
1686 * A library developer needs to expose functionality to crates which link
1687 against their library. As a consequence of the first case, this means that
1688 anything which is usable externally must be `pub` from the root down to the
1689 destination item. Any private item in the chain will disallow external
1692 * A crate needs a global available "helper module" to itself, but it doesn't
1693 want to expose the helper module as a public API. To accomplish this, the
1694 root of the crate's hierarchy would have a private module which then
1695 internally has a "public API". Because the entire crate is a descendant of
1696 the root, then the entire local crate can access this private module through
1699 * When writing unit tests for a module, it's often a common idiom to have an
1700 immediate child of the module to-be-tested named `mod test`. This module
1701 could access any items of the parent module through the second case, meaning
1702 that internal implementation details could also be seamlessly tested from the
1705 In the second case, it mentions that a private item "can be accessed" by the
1706 current module and its descendants, but the exact meaning of accessing an item
1707 depends on what the item is. Accessing a module, for example, would mean
1708 looking inside of it (to import more items). On the other hand, accessing a
1709 function would mean that it is invoked. Additionally, path expressions and
1710 import statements are considered to access an item in the sense that the
1711 import/expression is only valid if the destination is in the current visibility
1714 Here's an example of a program which exemplifies the three cases outlined
1718 // This module is private, meaning that no external crate can access this
1719 // module. Because it is private at the root of this current crate, however, any
1720 // module in the crate may access any publicly visible item in this module.
1721 mod crate_helper_module {
1723 // This function can be used by anything in the current crate
1724 pub fn crate_helper() {}
1726 // This function *cannot* be used by anything else in the crate. It is not
1727 // publicly visible outside of the `crate_helper_module`, so only this
1728 // current module and its descendants may access it.
1729 fn implementation_detail() {}
1732 // This function is "public to the root" meaning that it's available to external
1733 // crates linking against this one.
1734 pub fn public_api() {}
1736 // Similarly to 'public_api', this module is public so external crates may look
1739 use crate_helper_module;
1741 pub fn my_method() {
1742 // Any item in the local crate may invoke the helper module's public
1743 // interface through a combination of the two rules above.
1744 crate_helper_module::crate_helper();
1747 // This function is hidden to any module which is not a descendant of
1749 fn my_implementation() {}
1755 fn test_my_implementation() {
1756 // Because this module is a descendant of `submodule`, it's allowed
1757 // to access private items inside of `submodule` without a privacy
1759 super::my_implementation();
1767 For a Rust program to pass the privacy checking pass, all paths must be valid
1768 accesses given the two rules above. This includes all use statements,
1769 expressions, types, etc.
1771 ### Re-exporting and Visibility
1773 Rust allows publicly re-exporting items through a `pub use` directive. Because
1774 this is a public directive, this allows the item to be used in the current
1775 module through the rules above. It essentially allows public access into the
1776 re-exported item. For example, this program is valid:
1779 pub use self::implementation::api;
1781 mod implementation {
1790 This means that any external crate referencing `implementation::api::f` would
1791 receive a privacy violation, while the path `api::f` would be allowed.
1793 When re-exporting a private item, it can be thought of as allowing the "privacy
1794 chain" being short-circuited through the reexport instead of passing through
1795 the namespace hierarchy as it normally would.
1799 Any item declaration may have an _attribute_ applied to it. Attributes in Rust
1800 are modeled on Attributes in ECMA-335, with the syntax coming from ECMA-334
1801 (C#). An attribute is a general, free-form metadatum that is interpreted
1802 according to name, convention, and language and compiler version. Attributes
1803 may appear as any of:
1805 * A single identifier, the attribute name
1806 * An identifier followed by the equals sign '=' and a literal, providing a
1808 * An identifier followed by a parenthesized list of sub-attribute arguments
1810 Attributes with a bang ("!") after the hash ("#") apply to the item that the
1811 attribute is declared within. Attributes that do not have a bang after the hash
1812 apply to the item that follows the attribute.
1814 An example of attributes:
1817 // General metadata applied to the enclosing module or crate.
1818 #![crate_type = "lib"]
1820 // A function marked as a unit test
1826 // A conditionally-compiled module
1827 #[cfg(target_os="linux")]
1832 // A lint attribute used to suppress a warning/error
1833 #[allow(non_camel_case_types)]
1837 > **Note:** At some point in the future, the compiler will distinguish between
1838 > language-reserved and user-available attributes. Until then, there is
1839 > effectively no difference between an attribute handled by a loadable syntax
1840 > extension and the compiler.
1842 ### Crate-only attributes
1844 - `crate_name` - specify the crate's crate name.
1845 - `crate_type` - see [linkage](#linkage).
1846 - `feature` - see [compiler features](#compiler-features).
1847 - `no_builtins` - disable optimizing certain code patterns to invocations of
1848 library functions that are assumed to exist
1849 - `no_main` - disable emitting the `main` symbol. Useful when some other
1850 object being linked to defines `main`.
1851 - `no_start` - disable linking to the `native` crate, which specifies the
1852 "start" language item.
1853 - `no_std` - disable linking to the `std` crate.
1854 - `plugin` - load a list of named crates as compiler plugins, e.g.
1855 `#![plugin(foo, bar)]`. Optional arguments for each plugin,
1856 i.e. `#![plugin(foo(... args ...))]`, are provided to the plugin's
1857 registrar function. The `plugin` feature gate is required to use
1859 - `recursion_limit` - Sets the maximum depth for potentially
1860 infinitely-recursive compile-time operations like
1861 auto-dereference or macro expansion. The default is
1862 `#![recursion_limit="64"]`.
1864 ### Module-only attributes
1866 - `no_implicit_prelude` - disable injecting `use std::prelude::*` in this
1868 - `path` - specifies the file to load the module from. `#[path="foo.rs"] mod
1869 bar;` is equivalent to `mod bar { /* contents of foo.rs */ }`. The path is
1870 taken relative to the directory that the current module is in.
1872 ### Function-only attributes
1874 - `main` - indicates that this function should be passed to the entry point,
1875 rather than the function in the crate root named `main`.
1876 - `plugin_registrar` - mark this function as the registration point for
1877 [compiler plugins][plugin], such as loadable syntax extensions.
1878 - `start` - indicates that this function should be used as the entry point,
1879 overriding the "start" language item. See the "start" [language
1880 item](#language-items) for more details.
1881 - `test` - indicates that this function is a test function, to only be compiled
1882 in case of `--test`.
1883 - `should_panic` - indicates that this test function should panic, inverting the success condition.
1884 - `cold` - The function is unlikely to be executed, so optimize it (and calls
1887 ### Static-only attributes
1889 - `thread_local` - on a `static mut`, this signals that the value of this
1890 static may change depending on the current thread. The exact consequences of
1891 this are implementation-defined.
1895 On an `extern` block, the following attributes are interpreted:
1897 - `link_args` - specify arguments to the linker, rather than just the library
1898 name and type. This is feature gated and the exact behavior is
1899 implementation-defined (due to variety of linker invocation syntax).
1900 - `link` - indicate that a native library should be linked to for the
1901 declarations in this block to be linked correctly. `link` supports an optional
1902 `kind` key with three possible values: `dylib`, `static`, and `framework`. See
1903 [external blocks](#external-blocks) for more about external blocks. Two
1904 examples: `#[link(name = "readline")]` and
1905 `#[link(name = "CoreFoundation", kind = "framework")]`.
1906 - `linked_from` - indicates what native library this block of FFI items is
1907 coming from. This attribute is of the form `#[linked_from = "foo"]` where
1908 `foo` is the name of a library in either `#[link]` or a `-l` flag. This
1909 attribute is currently required to export symbols from a Rust dynamic library
1910 on Windows, and it is feature gated behind the `linked_from` feature.
1912 On declarations inside an `extern` block, the following attributes are
1915 - `link_name` - the name of the symbol that this function or static should be
1917 - `linkage` - on a static, this specifies the [linkage
1918 type](http://llvm.org/docs/LangRef.html#linkage-types).
1922 - `repr` - on C-like enums, this sets the underlying type used for
1923 representation. Takes one argument, which is the primitive
1924 type this enum should be represented for, or `C`, which specifies that it
1925 should be the default `enum` size of the C ABI for that platform. Note that
1926 enum representation in C is undefined, and this may be incorrect when the C
1927 code is compiled with certain flags.
1931 - `repr` - specifies the representation to use for this struct. Takes a list
1932 of options. The currently accepted ones are `C` and `packed`, which may be
1933 combined. `C` will use a C ABI compatible struct layout, and `packed` will
1934 remove any padding between fields (note that this is very fragile and may
1935 break platforms which require aligned access).
1937 ### Macro-related attributes
1939 - `macro_use` on a `mod` — macros defined in this module will be visible in the
1940 module's parent, after this module has been included.
1942 - `macro_use` on an `extern crate` — load macros from this crate. An optional
1943 list of names `#[macro_use(foo, bar)]` restricts the import to just those
1944 macros named. The `extern crate` must appear at the crate root, not inside
1945 `mod`, which ensures proper function of the [`$crate` macro
1946 variable](book/macros.html#the-variable-crate).
1948 - `macro_reexport` on an `extern crate` — re-export the named macros.
1950 - `macro_export` - export a macro for cross-crate usage.
1952 - `no_link` on an `extern crate` — even if we load this crate for macros, don't
1953 link it into the output.
1955 See the [macros section of the
1956 book](book/macros.html#scoping-and-macro-importexport) for more information on
1960 ### Miscellaneous attributes
1962 - `export_name` - on statics and functions, this determines the name of the
1964 - `link_section` - on statics and functions, this specifies the section of the
1965 object file that this item's contents will be placed into.
1966 - `no_mangle` - on any item, do not apply the standard name mangling. Set the
1967 symbol for this item to its identifier.
1968 - `simd` - on certain tuple structs, derive the arithmetic operators, which
1969 lower to the target's SIMD instructions, if any; the `simd` feature gate
1970 is necessary to use this attribute.
1971 - `unsafe_destructor_blind_to_params` - on `Drop::drop` method, asserts that the
1972 destructor code (and all potential specializations of that code) will
1973 never attempt to read from nor write to any references with lifetimes
1974 that come in via generic parameters. This is a constraint we cannot
1975 currently express via the type system, and therefore we rely on the
1976 programmer to assert that it holds. Adding this to a Drop impl causes
1977 the associated destructor to be considered "uninteresting" by the
1978 Drop-Check rule, and thus it can help sidestep data ordering
1979 constraints that would otherwise be introduced by the Drop-Check
1980 rule. Such sidestepping of the constraints, if done incorrectly, can
1981 lead to undefined behavior (in the form of reading or writing to data
1982 outside of its dynamic extent), and thus this attribute has the word
1983 "unsafe" in its name. To use this, the
1984 `unsafe_destructor_blind_to_params` feature gate must be enabled.
1985 - `unsafe_no_drop_flag` - on structs, remove the flag that prevents
1986 destructors from being run twice. Destructors might be run multiple times on
1987 the same object with this attribute. To use this, the `unsafe_no_drop_flag` feature
1988 gate must be enabled.
1989 - `doc` - Doc comments such as `/// foo` are equivalent to `#[doc = "foo"]`.
1990 - `rustc_on_unimplemented` - Write a custom note to be shown along with the error
1991 when the trait is found to be unimplemented on a type.
1992 You may use format arguments like `{T}`, `{A}` to correspond to the
1993 types at the point of use corresponding to the type parameters of the
1994 trait of the same name. `{Self}` will be replaced with the type that is supposed
1995 to implement the trait but doesn't. To use this, the `on_unimplemented` feature gate
1998 ### Conditional compilation
2000 Sometimes one wants to have different compiler outputs from the same code,
2001 depending on build target, such as targeted operating system, or to enable
2004 There are two kinds of configuration options, one that is either defined or not
2005 (`#[cfg(foo)]`), and the other that contains a string that can be checked
2006 against (`#[cfg(bar = "baz")]`). Currently, only compiler-defined configuration
2007 options can have the latter form.
2010 // The function is only included in the build when compiling for OSX
2011 #[cfg(target_os = "macos")]
2016 // This function is only included when either foo or bar is defined
2017 #[cfg(any(foo, bar))]
2018 fn needs_foo_or_bar() {
2022 // This function is only included when compiling for a unixish OS with a 32-bit
2024 #[cfg(all(unix, target_pointer_width = "32"))]
2025 fn on_32bit_unix() {
2029 // This function is only included when foo is not defined
2031 fn needs_not_foo() {
2036 This illustrates some conditional compilation can be achieved using the
2037 `#[cfg(...)]` attribute. `any`, `all` and `not` can be used to assemble
2038 arbitrarily complex configurations through nesting.
2040 The following configurations must be defined by the implementation:
2042 * `debug_assertions` - Enabled by default when compiling without optimizations.
2043 This can be used to enable extra debugging code in development but not in
2044 production. For example, it controls the behavior of the standard library's
2045 `debug_assert!` macro.
2046 * `target_arch = "..."` - Target CPU architecture, such as `"x86"`, `"x86_64"`
2047 `"mips"`, `"powerpc"`, `"powerpc64"`, `"arm"`, or `"aarch64"`.
2048 * `target_endian = "..."` - Endianness of the target CPU, either `"little"` or
2050 * `target_env = ".."` - An option provided by the compiler by default
2051 describing the runtime environment of the target platform. Some examples of
2052 this are `musl` for builds targeting the MUSL libc implementation, `msvc` for
2053 Windows builds targeting MSVC, and `gnu` frequently the rest of the time. This
2054 option may also be blank on some platforms.
2055 * `target_family = "..."` - Operating system family of the target, e. g.
2056 `"unix"` or `"windows"`. The value of this configuration option is defined
2057 as a configuration itself, like `unix` or `windows`.
2058 * `target_os = "..."` - Operating system of the target, examples include
2059 `"windows"`, `"macos"`, `"ios"`, `"linux"`, `"android"`, `"freebsd"`, `"dragonfly"`,
2060 `"bitrig"` , `"openbsd"` or `"netbsd"`.
2061 * `target_pointer_width = "..."` - Target pointer width in bits. This is set
2062 to `"32"` for targets with 32-bit pointers, and likewise set to `"64"` for
2064 * `target_vendor = "..."` - Vendor of the target, for example `apple`, `pc`, or
2066 * `test` - Enabled when compiling the test harness (using the `--test` flag).
2067 * `unix` - See `target_family`.
2068 * `windows` - See `target_family`.
2070 You can also set another attribute based on a `cfg` variable with `cfg_attr`:
2076 Will be the same as `#[b]` if `a` is set by `cfg`, and nothing otherwise.
2078 ### Lint check attributes
2080 A lint check names a potentially undesirable coding pattern, such as
2081 unreachable code or omitted documentation, for the static entity to which the
2084 For any lint check `C`:
2086 * `allow(C)` overrides the check for `C` so that violations will go
2088 * `deny(C)` signals an error after encountering a violation of `C`,
2089 * `forbid(C)` is the same as `deny(C)`, but also forbids changing the lint
2091 * `warn(C)` warns about violations of `C` but continues compilation.
2093 The lint checks supported by the compiler can be found via `rustc -W help`,
2094 along with their default settings. [Compiler
2095 plugins](book/compiler-plugins.html#lint-plugins) can provide additional lint checks.
2099 // Missing documentation is ignored here
2100 #[allow(missing_docs)]
2101 pub fn undocumented_one() -> i32 { 1 }
2103 // Missing documentation signals a warning here
2104 #[warn(missing_docs)]
2105 pub fn undocumented_too() -> i32 { 2 }
2107 // Missing documentation signals an error here
2108 #[deny(missing_docs)]
2109 pub fn undocumented_end() -> i32 { 3 }
2113 This example shows how one can use `allow` and `warn` to toggle a particular
2117 #[warn(missing_docs)]
2119 #[allow(missing_docs)]
2121 // Missing documentation is ignored here
2122 pub fn undocumented_one() -> i32 { 1 }
2124 // Missing documentation signals a warning here,
2125 // despite the allow above.
2126 #[warn(missing_docs)]
2127 pub fn undocumented_two() -> i32 { 2 }
2130 // Missing documentation signals a warning here
2131 pub fn undocumented_too() -> i32 { 3 }
2135 This example shows how one can use `forbid` to disallow uses of `allow` for
2139 #[forbid(missing_docs)]
2141 // Attempting to toggle warning signals an error here
2142 #[allow(missing_docs)]
2144 pub fn undocumented_too() -> i32 { 2 }
2150 Some primitive Rust operations are defined in Rust code, rather than being
2151 implemented directly in C or assembly language. The definitions of these
2152 operations have to be easy for the compiler to find. The `lang` attribute
2153 makes it possible to declare these operations. For example, the `str` module
2154 in the Rust standard library defines the string equality function:
2158 pub fn eq_slice(a: &str, b: &str) -> bool {
2163 The name `str_eq` has a special meaning to the Rust compiler, and the presence
2164 of this definition means that it will use this definition when generating calls
2165 to the string equality function.
2167 The set of language items is currently considered unstable. A complete
2168 list of the built-in language items will be added in the future.
2170 ### Inline attributes
2172 The inline attribute suggests that the compiler should place a copy of
2173 the function or static in the caller, rather than generating code to
2174 call the function or access the static where it is defined.
2176 The compiler automatically inlines functions based on internal heuristics.
2177 Incorrectly inlining functions can actually make the program slower, so it
2178 should be used with care.
2180 `#[inline]` and `#[inline(always)]` always cause the function to be serialized
2181 into the crate metadata to allow cross-crate inlining.
2183 There are three different types of inline attributes:
2185 * `#[inline]` hints the compiler to perform an inline expansion.
2186 * `#[inline(always)]` asks the compiler to always perform an inline expansion.
2187 * `#[inline(never)]` asks the compiler to never perform an inline expansion.
2191 The `derive` attribute allows certain traits to be automatically implemented
2192 for data structures. For example, the following will create an `impl` for the
2193 `PartialEq` and `Clone` traits for `Foo`, the type parameter `T` will be given
2194 the `PartialEq` or `Clone` constraints for the appropriate `impl`:
2197 #[derive(PartialEq, Clone)]
2204 The generated `impl` for `PartialEq` is equivalent to
2207 # struct Foo<T> { a: i32, b: T }
2208 impl<T: PartialEq> PartialEq for Foo<T> {
2209 fn eq(&self, other: &Foo<T>) -> bool {
2210 self.a == other.a && self.b == other.b
2213 fn ne(&self, other: &Foo<T>) -> bool {
2214 self.a != other.a || self.b != other.b
2219 ### Compiler Features
2221 Certain aspects of Rust may be implemented in the compiler, but they're not
2222 necessarily ready for every-day use. These features are often of "prototype
2223 quality" or "almost production ready", but may not be stable enough to be
2224 considered a full-fledged language feature.
2226 For this reason, Rust recognizes a special crate-level attribute of the form:
2229 #![feature(feature1, feature2, feature3)]
2232 This directive informs the compiler that the feature list: `feature1`,
2233 `feature2`, and `feature3` should all be enabled. This is only recognized at a
2234 crate-level, not at a module-level. Without this directive, all features are
2235 considered off, and using the features will result in a compiler error.
2237 The currently implemented features of the reference compiler are:
2239 * `advanced_slice_patterns` - See the [match expressions](#match-expressions)
2240 section for discussion; the exact semantics of
2241 slice patterns are subject to change, so some types
2244 * `slice_patterns` - OK, actually, slice patterns are just scary and
2245 completely unstable.
2247 * `asm` - The `asm!` macro provides a means for inline assembly. This is often
2248 useful, but the exact syntax for this feature along with its
2249 semantics are likely to change, so this macro usage must be opted
2252 * `associated_consts` - Allows constants to be defined in `impl` and `trait`
2253 blocks, so that they can be associated with a type or
2254 trait in a similar manner to methods and associated
2257 * `box_patterns` - Allows `box` patterns, the exact semantics of which
2258 is subject to change.
2260 * `box_syntax` - Allows use of `box` expressions, the exact semantics of which
2261 is subject to change.
2263 * `cfg_target_vendor` - Allows conditional compilation using the `target_vendor`
2264 matcher which is subject to change.
2266 * `concat_idents` - Allows use of the `concat_idents` macro, which is in many
2267 ways insufficient for concatenating identifiers, and may be
2268 removed entirely for something more wholesome.
2270 * `custom_attribute` - Allows the usage of attributes unknown to the compiler
2271 so that new attributes can be added in a backwards compatible
2274 * `custom_derive` - Allows the use of `#[derive(Foo,Bar)]` as sugar for
2275 `#[derive_Foo] #[derive_Bar]`, which can be user-defined syntax
2278 * `intrinsics` - Allows use of the "rust-intrinsics" ABI. Compiler intrinsics
2279 are inherently unstable and no promise about them is made.
2281 * `lang_items` - Allows use of the `#[lang]` attribute. Like `intrinsics`,
2282 lang items are inherently unstable and no promise about them
2285 * `link_args` - This attribute is used to specify custom flags to the linker,
2286 but usage is strongly discouraged. The compiler's usage of the
2287 system linker is not guaranteed to continue in the future, and
2288 if the system linker is not used then specifying custom flags
2289 doesn't have much meaning.
2291 * `link_llvm_intrinsics` – Allows linking to LLVM intrinsics via
2292 `#[link_name="llvm.*"]`.
2294 * `linkage` - Allows use of the `linkage` attribute, which is not portable.
2296 * `log_syntax` - Allows use of the `log_syntax` macro attribute, which is a
2297 nasty hack that will certainly be removed.
2299 * `main` - Allows use of the `#[main]` attribute, which changes the entry point
2300 into a Rust program. This capability is subject to change.
2302 * `macro_reexport` - Allows macros to be re-exported from one crate after being imported
2303 from another. This feature was originally designed with the sole
2304 use case of the Rust standard library in mind, and is subject to
2307 * `non_ascii_idents` - The compiler supports the use of non-ascii identifiers,
2308 but the implementation is a little rough around the
2309 edges, so this can be seen as an experimental feature
2310 for now until the specification of identifiers is fully
2313 * `no_std` - Allows the `#![no_std]` crate attribute, which disables the implicit
2314 `extern crate std`. This typically requires use of the unstable APIs
2315 behind the libstd "facade", such as libcore and libcollections. It
2316 may also cause problems when using syntax extensions, including
2319 * `on_unimplemented` - Allows the `#[rustc_on_unimplemented]` attribute, which allows
2320 trait definitions to add specialized notes to error messages
2321 when an implementation was expected but not found.
2323 * `optin_builtin_traits` - Allows the definition of default and negative trait
2324 implementations. Experimental.
2326 * `plugin` - Usage of [compiler plugins][plugin] for custom lints or syntax extensions.
2327 These depend on compiler internals and are subject to change.
2329 * `plugin_registrar` - Indicates that a crate provides [compiler plugins][plugin].
2331 * `quote` - Allows use of the `quote_*!` family of macros, which are
2332 implemented very poorly and will likely change significantly
2333 with a proper implementation.
2335 * `rustc_attrs` - Gates internal `#[rustc_*]` attributes which may be
2336 for internal use only or have meaning added to them in the future.
2338 * `rustc_diagnostic_macros`- A mysterious feature, used in the implementation
2339 of rustc, not meant for mortals.
2341 * `simd` - Allows use of the `#[simd]` attribute, which is overly simple and
2342 not the SIMD interface we want to expose in the long term.
2344 * `simd_ffi` - Allows use of SIMD vectors in signatures for foreign functions.
2345 The SIMD interface is subject to change.
2347 * `start` - Allows use of the `#[start]` attribute, which changes the entry point
2348 into a Rust program. This capability, especially the signature for the
2349 annotated function, is subject to change.
2351 * `thread_local` - The usage of the `#[thread_local]` attribute is experimental
2352 and should be seen as unstable. This attribute is used to
2353 declare a `static` as being unique per-thread leveraging
2354 LLVM's implementation which works in concert with the kernel
2355 loader and dynamic linker. This is not necessarily available
2356 on all platforms, and usage of it is discouraged.
2358 * `trace_macros` - Allows use of the `trace_macros` macro, which is a nasty
2359 hack that will certainly be removed.
2361 * `unboxed_closures` - Rust's new closure design, which is currently a work in
2362 progress feature with many known bugs.
2364 * `unsafe_no_drop_flag` - Allows use of the `#[unsafe_no_drop_flag]` attribute,
2365 which removes hidden flag added to a type that
2366 implements the `Drop` trait. The design for the
2367 `Drop` flag is subject to change, and this feature
2368 may be removed in the future.
2370 * `unmarked_api` - Allows use of items within a `#![staged_api]` crate
2371 which have not been marked with a stability marker.
2372 Such items should not be allowed by the compiler to exist,
2373 so if you need this there probably is a compiler bug.
2375 * `allow_internal_unstable` - Allows `macro_rules!` macros to be tagged with the
2376 `#[allow_internal_unstable]` attribute, designed
2377 to allow `std` macros to call
2378 `#[unstable]`/feature-gated functionality
2379 internally without imposing on callers
2380 (i.e. making them behave like function calls in
2381 terms of encapsulation).
2382 * - `default_type_parameter_fallback` - Allows type parameter defaults to
2383 influence type inference.
2384 * - `braced_empty_structs` - Allows use of empty structs and enum variants with braces.
2386 * - `stmt_expr_attributes` - Allows attributes on expressions and
2387 non-item statements.
2389 * - `deprecated` - Allows using the `#[deprecated]` attribute.
2391 * - `type_ascription` - Allows type ascription expressions `expr: Type`.
2393 * - `abi_vectorcall` - Allows the usage of the vectorcall calling convention
2394 (e.g. `extern "vectorcall" func fn_();`)
2396 If a feature is promoted to a language feature, then all existing programs will
2397 start to receive compilation warnings about `#![feature]` directives which enabled
2398 the new feature (because the directive is no longer necessary). However, if a
2399 feature is decided to be removed from the language, errors will be issued (if
2400 there isn't a parser error first). The directive in this case is no longer
2401 necessary, and it's likely that existing code will break if the feature isn't
2404 If an unknown feature is found in a directive, it results in a compiler error.
2405 An unknown feature is one which has never been recognized by the compiler.
2407 # Statements and expressions
2409 Rust is _primarily_ an expression language. This means that most forms of
2410 value-producing or effect-causing evaluation are directed by the uniform syntax
2411 category of _expressions_. Each kind of expression can typically _nest_ within
2412 each other kind of expression, and rules for evaluation of expressions involve
2413 specifying both the value produced by the expression and the order in which its
2414 sub-expressions are themselves evaluated.
2416 In contrast, statements in Rust serve _mostly_ to contain and explicitly
2417 sequence expression evaluation.
2421 A _statement_ is a component of a block, which is in turn a component of an
2422 outer [expression](#expressions) or [function](#functions).
2424 Rust has two kinds of statement: [declaration
2425 statements](#declaration-statements) and [expression
2426 statements](#expression-statements).
2428 ### Declaration statements
2430 A _declaration statement_ is one that introduces one or more *names* into the
2431 enclosing statement block. The declared names may denote new variables or new
2434 #### Item declarations
2436 An _item declaration statement_ has a syntactic form identical to an
2437 [item](#items) declaration within a module. Declaring an item — a
2438 function, enumeration, struct, type, static, trait, implementation or module
2439 — locally within a statement block is simply a way of restricting its
2440 scope to a narrow region containing all of its uses; it is otherwise identical
2441 in meaning to declaring the item outside the statement block.
2443 > **Note**: there is no implicit capture of the function's dynamic environment when
2444 > declaring a function-local item.
2446 #### `let` statements
2448 A _`let` statement_ introduces a new set of variables, given by a pattern. The
2449 pattern may be followed by a type annotation, and/or an initializer expression.
2450 When no type annotation is given, the compiler will infer the type, or signal
2451 an error if insufficient type information is available for definite inference.
2452 Any variables introduced by a variable declaration are visible from the point of
2453 declaration until the end of the enclosing block scope.
2455 ### Expression statements
2457 An _expression statement_ is one that evaluates an [expression](#expressions)
2458 and ignores its result. The type of an expression statement `e;` is always
2459 `()`, regardless of the type of `e`. As a rule, an expression statement's
2460 purpose is to trigger the effects of evaluating its expression.
2464 An expression may have two roles: it always produces a *value*, and it may have
2465 *effects* (otherwise known as "side effects"). An expression *evaluates to* a
2466 value, and has effects during *evaluation*. Many expressions contain
2467 sub-expressions (operands). The meaning of each kind of expression dictates
2470 * Whether or not to evaluate the sub-expressions when evaluating the expression
2471 * The order in which to evaluate the sub-expressions
2472 * How to combine the sub-expressions' values to obtain the value of the expression
2474 In this way, the structure of expressions dictates the structure of execution.
2475 Blocks are just another kind of expression, so blocks, statements, expressions,
2476 and blocks again can recursively nest inside each other to an arbitrary depth.
2478 #### Lvalues, rvalues and temporaries
2480 Expressions are divided into two main categories: _lvalues_ and _rvalues_.
2481 Likewise within each expression, sub-expressions may occur in _lvalue context_
2482 or _rvalue context_. The evaluation of an expression depends both on its own
2483 category and the context it occurs within.
2485 An lvalue is an expression that represents a memory location. These expressions
2486 are [paths](#path-expressions) (which refer to local variables, function and
2487 method arguments, or static variables), dereferences (`*expr`), [indexing
2488 expressions](#index-expressions) (`expr[expr]`), and [field
2489 references](#field-expressions) (`expr.f`). All other expressions are rvalues.
2491 The left operand of an [assignment](#assignment-expressions) or
2492 [compound-assignment](#compound-assignment-expressions) expression is
2493 an lvalue context, as is the single operand of a unary
2494 [borrow](#unary-operator-expressions). The discriminant or subject of
2495 a [match expression](#match-expressions) may be an lvalue context, if
2496 ref bindings are made, but is otherwise an rvalue context. All other
2497 expression contexts are rvalue contexts.
2499 When an lvalue is evaluated in an _lvalue context_, it denotes a memory
2500 location; when evaluated in an _rvalue context_, it denotes the value held _in_
2501 that memory location.
2503 ##### Temporary lifetimes
2505 When an rvalue is used in an lvalue context, a temporary un-named
2506 lvalue is created and used instead. The lifetime of temporary values
2507 is typically the innermost enclosing statement; the tail expression of
2508 a block is considered part of the statement that encloses the block.
2510 When a temporary rvalue is being created that is assigned into a `let`
2511 declaration, however, the temporary is created with the lifetime of
2512 the enclosing block instead, as using the enclosing statement (the
2513 `let` declaration) would be a guaranteed error (since a pointer to the
2514 temporary would be stored into a variable, but the temporary would be
2515 freed before the variable could be used). The compiler uses simple
2516 syntactic rules to decide which values are being assigned into a `let`
2517 binding, and therefore deserve a longer temporary lifetime.
2519 Here are some examples:
2521 - `let x = foo(&temp())`. The expression `temp()` is an rvalue. As it
2522 is being borrowed, a temporary is created which will be freed after
2523 the innermost enclosing statement (the `let` declaration, in this case).
2524 - `let x = temp().foo()`. This is the same as the previous example,
2525 except that the value of `temp()` is being borrowed via autoref on a
2526 method-call. Here we are assuming that `foo()` is an `&self` method
2527 defined in some trait, say `Foo`. In other words, the expression
2528 `temp().foo()` is equivalent to `Foo::foo(&temp())`.
2529 - `let x = &temp()`. Here, the same temporary is being assigned into
2530 `x`, rather than being passed as a parameter, and hence the
2531 temporary's lifetime is considered to be the enclosing block.
2532 - `let x = SomeStruct { foo: &temp() }`. As in the previous case, the
2533 temporary is assigned into a struct which is then assigned into a
2534 binding, and hence it is given the lifetime of the enclosing block.
2535 - `let x = [ &temp() ]`. As in the previous case, the
2536 temporary is assigned into an array which is then assigned into a
2537 binding, and hence it is given the lifetime of the enclosing block.
2538 - `let ref x = temp()`. In this case, the temporary is created using a ref binding,
2539 but the result is the same: the lifetime is extended to the enclosing block.
2541 #### Moved and copied types
2543 When a [local variable](#variables) is used as an
2544 [rvalue](#lvalues-rvalues-and-temporaries), the variable will be copied
2545 if its type implements `Copy`. All others are moved.
2547 ### Literal expressions
2549 A _literal expression_ consists of one of the [literal](#literals) forms
2550 described earlier. It directly describes a number, character, string, boolean
2551 value, or the unit value.
2555 "hello"; // string type
2556 '5'; // character type
2560 ### Path expressions
2562 A [path](#paths) used as an expression context denotes either a local variable
2563 or an item. Path expressions are [lvalues](#lvalues-rvalues-and-temporaries).
2565 ### Tuple expressions
2567 Tuples are written by enclosing zero or more comma-separated expressions in
2568 parentheses. They are used to create [tuple-typed](#tuple-types) values.
2572 ("a", 4usize, true);
2575 You can disambiguate a single-element tuple from a value in parentheses with a
2579 (0,); // single-element tuple
2580 (0); // zero in parentheses
2583 ### Struct expressions
2585 There are several forms of struct expressions. A _struct expression_
2586 consists of the [path](#paths) of a [struct item](#structs), followed by
2587 a brace-enclosed list of one or more comma-separated name-value pairs,
2588 providing the field values of a new instance of the struct. A field name
2589 can be any identifier, and is separated from its value expression by a colon.
2590 The location denoted by a struct field is mutable if and only if the
2591 enclosing struct is mutable.
2593 A _tuple struct expression_ consists of the [path](#paths) of a [struct
2594 item](#structs), followed by a parenthesized list of one or more
2595 comma-separated expressions (in other words, the path of a struct item
2596 followed by a tuple expression). The struct item must be a tuple struct
2599 A _unit-like struct expression_ consists only of the [path](#paths) of a
2600 [struct item](#structs).
2602 The following are examples of struct expressions:
2605 # struct Point { x: f64, y: f64 }
2606 # struct TuplePoint(f64, f64);
2607 # mod game { pub struct User<'a> { pub name: &'a str, pub age: u32, pub score: usize } }
2608 # struct Cookie; fn some_fn<T>(t: T) {}
2609 Point {x: 10.0, y: 20.0};
2610 TuplePoint(10.0, 20.0);
2611 let u = game::User {name: "Joe", age: 35, score: 100_000};
2612 some_fn::<Cookie>(Cookie);
2615 A struct expression forms a new value of the named struct type. Note
2616 that for a given *unit-like* struct type, this will always be the same
2619 A struct expression can terminate with the syntax `..` followed by an
2620 expression to denote a functional update. The expression following `..` (the
2621 base) must have the same struct type as the new struct type being formed.
2622 The entire expression denotes the result of constructing a new struct (with
2623 the same type as the base expression) with the given values for the fields that
2624 were explicitly specified and the values in the base expression for all other
2628 # struct Point3d { x: i32, y: i32, z: i32 }
2629 let base = Point3d {x: 1, y: 2, z: 3};
2630 Point3d {y: 0, z: 10, .. base};
2633 ### Block expressions
2635 A _block expression_ is similar to a module in terms of the declarations that
2636 are possible. Each block conceptually introduces a new namespace scope. Use
2637 items can bring new names into scopes and declared items are in scope for only
2640 A block will execute each statement sequentially, and then execute the
2641 expression (if given). If the block ends in a statement, its value is `()`:
2644 let x: () = { println!("Hello."); };
2647 If it ends in an expression, its value and type are that of the expression:
2650 let x: i32 = { println!("Hello."); 5 };
2655 ### Method-call expressions
2657 A _method call_ consists of an expression followed by a single dot, an
2658 identifier, and a parenthesized expression-list. Method calls are resolved to
2659 methods on specific traits, either statically dispatching to a method if the
2660 exact `self`-type of the left-hand-side is known, or dynamically dispatching if
2661 the left-hand-side expression is an indirect [trait object](#trait-objects).
2663 ### Field expressions
2665 A _field expression_ consists of an expression followed by a single dot and an
2666 identifier, when not immediately followed by a parenthesized expression-list
2667 (the latter is a [method call expression](#method-call-expressions)). A field
2668 expression denotes a field of a [struct](#struct-types).
2673 (Struct {a: 10, b: 20}).a;
2676 A field access is an [lvalue](#lvalues-rvalues-and-temporaries) referring to
2677 the value of that field. When the type providing the field inherits mutability,
2678 it can be [assigned](#assignment-expressions) to.
2680 Also, if the type of the expression to the left of the dot is a
2681 pointer, it is automatically dereferenced as many times as necessary
2682 to make the field access possible. In cases of ambiguity, we prefer
2683 fewer autoderefs to more.
2685 ### Array expressions
2687 An [array](#array-and-slice-types) _expression_ is written by enclosing zero
2688 or more comma-separated expressions of uniform type in square brackets.
2690 In the `[expr ';' expr]` form, the expression after the `';'` must be a
2691 constant expression that can be evaluated at compile time, such as a
2692 [literal](#literals) or a [static item](#static-items).
2696 ["a", "b", "c", "d"];
2697 [0; 128]; // array with 128 zeros
2698 [0u8, 0u8, 0u8, 0u8];
2701 ### Index expressions
2703 [Array](#array-and-slice-types)-typed expressions can be indexed by
2704 writing a square-bracket-enclosed expression (the index) after them. When the
2705 array is mutable, the resulting [lvalue](#lvalues-rvalues-and-temporaries) can
2708 Indices are zero-based, and may be of any integral type. Vector access is
2709 bounds-checked at compile-time for constant arrays being accessed with a constant index value.
2710 Otherwise a check will be performed at run-time that will put the thread in a _panicked state_ if it fails.
2715 let x = (["a", "b"])[10]; // compiler error: const index-expr is out of bounds
2718 let y = (["a", "b"])[n]; // panics
2720 let arr = ["a", "b"];
2724 Also, if the type of the expression to the left of the brackets is a
2725 pointer, it is automatically dereferenced as many times as necessary
2726 to make the indexing possible. In cases of ambiguity, we prefer fewer
2729 ### Range expressions
2731 The `..` operator will construct an object of one of the `std::ops::Range` variants.
2734 1..2; // std::ops::Range
2735 3..; // std::ops::RangeFrom
2736 ..4; // std::ops::RangeTo
2737 ..; // std::ops::RangeFull
2740 The following expressions are equivalent.
2743 let x = std::ops::Range {start: 0, end: 10};
2749 ### Unary operator expressions
2751 Rust defines the following unary operators. They are all written as prefix operators,
2752 before the expression they apply to.
2755 : Negation. May only be applied to numeric types.
2757 : Dereference. When applied to a [pointer](#pointer-types) it denotes the
2758 pointed-to location. For pointers to mutable locations, the resulting
2759 [lvalue](#lvalues-rvalues-and-temporaries) can be assigned to.
2760 On non-pointer types, it calls the `deref` method of the `std::ops::Deref`
2761 trait, or the `deref_mut` method of the `std::ops::DerefMut` trait (if
2762 implemented by the type and required for an outer expression that will or
2763 could mutate the dereference), and produces the result of dereferencing the
2764 `&` or `&mut` borrowed pointer returned from the overload method.
2766 : Logical negation. On the boolean type, this flips between `true` and
2767 `false`. On integer types, this inverts the individual bits in the
2768 two's complement representation of the value.
2770 : Borrowing. When applied to an lvalue, these operators produce a
2771 reference (pointer) to the lvalue. The lvalue is also placed into
2772 a borrowed state for the duration of the reference. For a shared
2773 borrow (`&`), this implies that the lvalue may not be mutated, but
2774 it may be read or shared again. For a mutable borrow (`&mut`), the
2775 lvalue may not be accessed in any way until the borrow expires.
2776 If the `&` or `&mut` operators are applied to an rvalue, a
2777 temporary value is created; the lifetime of this temporary value
2778 is defined by [syntactic rules](#temporary-lifetimes).
2780 ### Binary operator expressions
2782 Binary operators expressions are given in terms of [operator
2783 precedence](#operator-precedence).
2785 #### Arithmetic operators
2787 Binary arithmetic expressions are syntactic sugar for calls to built-in traits,
2788 defined in the `std::ops` module of the `std` library. This means that
2789 arithmetic operators can be overridden for user-defined types. The default
2790 meaning of the operators on standard types is given here.
2793 : Addition and array/string concatenation.
2794 Calls the `add` method on the `std::ops::Add` trait.
2797 Calls the `sub` method on the `std::ops::Sub` trait.
2800 Calls the `mul` method on the `std::ops::Mul` trait.
2803 Calls the `div` method on the `std::ops::Div` trait.
2806 Calls the `rem` method on the `std::ops::Rem` trait.
2808 #### Bitwise operators
2810 Like the [arithmetic operators](#arithmetic-operators), bitwise operators are
2811 syntactic sugar for calls to methods of built-in traits. This means that
2812 bitwise operators can be overridden for user-defined types. The default
2813 meaning of the operators on standard types is given here. Bitwise `&`, `|` and
2814 `^` applied to boolean arguments are equivalent to logical `&&`, `||` and `!=`
2815 evaluated in non-lazy fashion.
2819 Calls the `bitand` method of the `std::ops::BitAnd` trait.
2821 : Bitwise inclusive OR.
2822 Calls the `bitor` method of the `std::ops::BitOr` trait.
2824 : Bitwise exclusive OR.
2825 Calls the `bitxor` method of the `std::ops::BitXor` trait.
2828 Calls the `shl` method of the `std::ops::Shl` trait.
2830 : Right shift (arithmetic).
2831 Calls the `shr` method of the `std::ops::Shr` trait.
2833 #### Lazy boolean operators
2835 The operators `||` and `&&` may be applied to operands of boolean type. The
2836 `||` operator denotes logical 'or', and the `&&` operator denotes logical
2837 'and'. They differ from `|` and `&` in that the right-hand operand is only
2838 evaluated when the left-hand operand does not already determine the result of
2839 the expression. That is, `||` only evaluates its right-hand operand when the
2840 left-hand operand evaluates to `false`, and `&&` only when it evaluates to
2843 #### Comparison operators
2845 Comparison operators are, like the [arithmetic
2846 operators](#arithmetic-operators), and [bitwise operators](#bitwise-operators),
2847 syntactic sugar for calls to built-in traits. This means that comparison
2848 operators can be overridden for user-defined types. The default meaning of the
2849 operators on standard types is given here.
2853 Calls the `eq` method on the `std::cmp::PartialEq` trait.
2856 Calls the `ne` method on the `std::cmp::PartialEq` trait.
2859 Calls the `lt` method on the `std::cmp::PartialOrd` trait.
2862 Calls the `gt` method on the `std::cmp::PartialOrd` trait.
2864 : Less than or equal.
2865 Calls the `le` method on the `std::cmp::PartialOrd` trait.
2867 : Greater than or equal.
2868 Calls the `ge` method on the `std::cmp::PartialOrd` trait.
2870 #### Type cast expressions
2872 A type cast expression is denoted with the binary operator `as`.
2874 Executing an `as` expression casts the value on the left-hand side to the type
2875 on the right-hand side.
2877 An example of an `as` expression:
2880 # fn sum(values: &[f64]) -> f64 { 0.0 }
2881 # fn len(values: &[f64]) -> i32 { 0 }
2883 fn average(values: &[f64]) -> f64 {
2884 let sum: f64 = sum(values);
2885 let size: f64 = len(values) as f64;
2890 Some of the conversions which can be done through the `as` operator
2891 can also be done implicitly at various points in the program, such as
2892 argument passing and assignment to a `let` binding with an explicit
2893 type. Implicit conversions are limited to "harmless" conversions that
2894 do not lose information and which have minimal or no risk of
2895 surprising side-effects on the dynamic execution semantics.
2897 #### Assignment expressions
2899 An _assignment expression_ consists of an
2900 [lvalue](#lvalues-rvalues-and-temporaries) expression followed by an equals
2901 sign (`=`) and an [rvalue](#lvalues-rvalues-and-temporaries) expression.
2903 Evaluating an assignment expression [either copies or
2904 moves](#moved-and-copied-types) its right-hand operand to its left-hand
2913 #### Compound assignment expressions
2915 The `+`, `-`, `*`, `/`, `%`, `&`, `|`, `^`, `<<`, and `>>` operators may be
2916 composed with the `=` operator. The expression `lval OP= val` is equivalent to
2917 `lval = lval OP val`. For example, `x = x + 1` may be written as `x += 1`.
2919 Any such expression always has the [`unit`](#tuple-types) type.
2921 #### Operator precedence
2923 The precedence of Rust binary operators is ordered as follows, going from
2926 ```{.text .precedence}
2940 Operators at the same precedence level are evaluated left-to-right. [Unary
2941 operators](#unary-operator-expressions) have the same precedence level and are
2942 stronger than any of the binary operators.
2944 ### Grouped expressions
2946 An expression enclosed in parentheses evaluates to the result of the enclosed
2947 expression. Parentheses can be used to explicitly specify evaluation order
2948 within an expression.
2950 An example of a parenthesized expression:
2953 let x: i32 = (2 + 3) * 4;
2957 ### Call expressions
2959 A _call expression_ invokes a function, providing zero or more input variables
2960 and an optional location to move the function's output into. If the function
2961 eventually returns, then the expression completes.
2963 Some examples of call expressions:
2966 # fn add(x: i32, y: i32) -> i32 { 0 }
2968 let x: i32 = add(1i32, 2i32);
2969 let pi: Result<f32, _> = "3.14".parse();
2972 ### Lambda expressions
2974 A _lambda expression_ (sometimes called an "anonymous function expression")
2975 defines a function and denotes it as a value, in a single expression. A lambda
2976 expression is a pipe-symbol-delimited (`|`) list of identifiers followed by an
2979 A lambda expression denotes a function that maps a list of parameters
2980 (`ident_list`) onto the expression that follows the `ident_list`. The
2981 identifiers in the `ident_list` are the parameters to the function. These
2982 parameters' types need not be specified, as the compiler infers them from
2985 Lambda expressions are most useful when passing functions as arguments to other
2986 functions, as an abbreviation for defining and capturing a separate function.
2988 Significantly, lambda expressions _capture their environment_, which regular
2989 [function definitions](#functions) do not. The exact type of capture depends
2990 on the [function type](#function-types) inferred for the lambda expression. In
2991 the simplest and least-expensive form (analogous to a ```|| { }``` expression),
2992 the lambda expression captures its environment by reference, effectively
2993 borrowing pointers to all outer variables mentioned inside the function.
2994 Alternately, the compiler may infer that a lambda expression should copy or
2995 move values (depending on their type) from the environment into the lambda
2996 expression's captured environment.
2998 In this example, we define a function `ten_times` that takes a higher-order
2999 function argument, and we then call it with a lambda expression as an argument:
3002 fn ten_times<F>(f: F) where F: Fn(i32) {
3003 for index in 0..10 {
3008 ten_times(|j| println!("hello, {}", j));
3013 A `loop` expression denotes an infinite loop.
3015 A `loop` expression may optionally have a _label_. The label is written as
3016 a lifetime preceding the loop expression, as in `'foo: loop{ }`. If a
3017 label is present, then labeled `break` and `continue` expressions nested
3018 within this loop may exit out of this loop or return control to its head.
3019 See [break expressions](#break-expressions) and [continue
3020 expressions](#continue-expressions).
3022 ### `break` expressions
3024 A `break` expression has an optional _label_. If the label is absent, then
3025 executing a `break` expression immediately terminates the innermost loop
3026 enclosing it. It is only permitted in the body of a loop. If the label is
3027 present, then `break 'foo` terminates the loop with label `'foo`, which need not
3028 be the innermost label enclosing the `break` expression, but must enclose it.
3030 ### `continue` expressions
3032 A `continue` expression has an optional _label_. If the label is absent, then
3033 executing a `continue` expression immediately terminates the current iteration
3034 of the innermost loop enclosing it, returning control to the loop *head*. In
3035 the case of a `while` loop, the head is the conditional expression controlling
3036 the loop. In the case of a `for` loop, the head is the call-expression
3037 controlling the loop. If the label is present, then `continue 'foo` returns
3038 control to the head of the loop with label `'foo`, which need not be the
3039 innermost label enclosing the `continue` expression, but must enclose it.
3041 A `continue` expression is only permitted in the body of a loop.
3045 A `while` loop begins by evaluating the boolean loop conditional expression.
3046 If the loop conditional expression evaluates to `true`, the loop body block
3047 executes and control returns to the loop conditional expression. If the loop
3048 conditional expression evaluates to `false`, the `while` expression completes.
3061 Like `loop` expressions, `while` loops can be controlled with `break` or
3062 `continue`, and may optionally have a _label_. See [infinite
3063 loops](#infinite-loops), [break expressions](#break-expressions), and
3064 [continue expressions](#continue-expressions) for more information.
3066 ### `for` expressions
3068 A `for` expression is a syntactic construct for looping over elements provided
3069 by an implementation of `std::iter::IntoIterator`.
3071 An example of a `for` loop over the contents of an array:
3075 # fn bar(f: &Foo) { }
3080 let v: &[Foo] = &[a, b, c];
3087 An example of a for loop over a series of integers:
3090 # fn bar(b:usize) { }
3096 Like `loop` expressions, `for` loops can be controlled with `break` or
3097 `continue`, and may optionally have a _label_. See [infinite
3098 loops](#infinite-loops), [break expressions](#break-expressions), and
3099 [continue expressions](#continue-expressions) for more information.
3101 ### `if` expressions
3103 An `if` expression is a conditional branch in program control. The form of an
3104 `if` expression is a condition expression, followed by a consequent block, any
3105 number of `else if` conditions and blocks, and an optional trailing `else`
3106 block. The condition expressions must have type `bool`. If a condition
3107 expression evaluates to `true`, the consequent block is executed and any
3108 subsequent `else if` or `else` block is skipped. If a condition expression
3109 evaluates to `false`, the consequent block is skipped and any subsequent `else
3110 if` condition is evaluated. If all `if` and `else if` conditions evaluate to
3111 `false` then any `else` block is executed.
3113 ### `match` expressions
3115 A `match` expression branches on a *pattern*. The exact form of matching that
3116 occurs depends on the pattern. Patterns consist of some combination of
3117 literals, destructured arrays or enum constructors, structs and tuples,
3118 variable binding specifications, wildcards (`..`), and placeholders (`_`). A
3119 `match` expression has a *head expression*, which is the value to compare to
3120 the patterns. The type of the patterns must equal the type of the head
3123 In a pattern whose head expression has an `enum` type, a placeholder (`_`)
3124 stands for a *single* data field, whereas a wildcard `..` stands for *all* the
3125 fields of a particular variant.
3127 A `match` behaves differently depending on whether or not the head expression
3128 is an [lvalue or an rvalue](#lvalues-rvalues-and-temporaries). If the head
3129 expression is an rvalue, it is first evaluated into a temporary location, and
3130 the resulting value is sequentially compared to the patterns in the arms until
3131 a match is found. The first arm with a matching pattern is chosen as the branch
3132 target of the `match`, any variables bound by the pattern are assigned to local
3133 variables in the arm's block, and control enters the block.
3135 When the head expression is an lvalue, the match does not allocate a temporary
3136 location (however, a by-value binding may copy or move from the lvalue). When
3137 possible, it is preferable to match on lvalues, as the lifetime of these
3138 matches inherits the lifetime of the lvalue, rather than being restricted to
3139 the inside of the match.
3141 An example of a `match` expression:
3147 1 => println!("one"),
3148 2 => println!("two"),
3149 3 => println!("three"),
3150 4 => println!("four"),
3151 5 => println!("five"),
3152 _ => println!("something else"),
3156 Patterns that bind variables default to binding to a copy or move of the
3157 matched value (depending on the matched value's type). This can be changed to
3158 bind to a reference by using the `ref` keyword, or to a mutable reference using
3161 Subpatterns can also be bound to variables by the use of the syntax `variable @
3162 subpattern`. For example:
3168 e @ 1 ... 5 => println!("got a range element {}", e),
3169 _ => println!("anything"),
3173 Patterns can also dereference pointers by using the `&`, `&mut` and `box`
3174 symbols, as appropriate. For example, these two matches on `x: &i32` are
3179 let y = match *x { 0 => "zero", _ => "some" };
3180 let z = match x { &0 => "zero", _ => "some" };
3185 Multiple match patterns may be joined with the `|` operator. A range of values
3186 may be specified with `...`. For example:
3191 let message = match x {
3192 0 | 1 => "not many",
3198 Range patterns only work on scalar types (like integers and characters; not
3199 like arrays and structs, which have sub-components). A range pattern may not
3200 be a sub-range of another range pattern inside the same `match`.
3202 Finally, match patterns can accept *pattern guards* to further refine the
3203 criteria for matching a case. Pattern guards appear after the pattern and
3204 consist of a bool-typed expression following the `if` keyword. A pattern guard
3205 may refer to the variables bound within the pattern they follow.
3208 # let maybe_digit = Some(0);
3209 # fn process_digit(i: i32) { }
3210 # fn process_other(i: i32) { }
3212 let message = match maybe_digit {
3213 Some(x) if x < 10 => process_digit(x),
3214 Some(x) => process_other(x),
3219 ### `if let` expressions
3221 An `if let` expression is semantically identical to an `if` expression but in
3222 place of a condition expression it expects a `let` statement with a refutable
3223 pattern. If the value of the expression on the right hand side of the `let`
3224 statement matches the pattern, the corresponding block will execute, otherwise
3225 flow proceeds to the first `else` block that follows.
3228 let dish = ("Ham", "Eggs");
3230 // this body will be skipped because the pattern is refuted
3231 if let ("Bacon", b) = dish {
3232 println!("Bacon is served with {}", b);
3235 // this body will execute
3236 if let ("Ham", b) = dish {
3237 println!("Ham is served with {}", b);
3241 ### `while let` loops
3243 A `while let` loop is semantically identical to a `while` loop but in place of
3244 a condition expression it expects `let` statement with a refutable pattern. If
3245 the value of the expression on the right hand side of the `let` statement
3246 matches the pattern, the loop body block executes and control returns to the
3247 pattern matching statement. Otherwise, the while expression completes.
3249 ### `return` expressions
3251 Return expressions are denoted with the keyword `return`. Evaluating a `return`
3252 expression moves its argument into the designated output location for the
3253 current function call, destroys the current function activation frame, and
3254 transfers control to the caller frame.
3256 An example of a `return` expression:
3259 fn max(a: i32, b: i32) -> i32 {
3271 Every variable, item and value in a Rust program has a type. The _type_ of a
3272 *value* defines the interpretation of the memory holding it.
3274 Built-in types and type-constructors are tightly integrated into the language,
3275 in nontrivial ways that are not possible to emulate in user-defined types.
3276 User-defined types have limited capabilities.
3280 The primitive types are the following:
3282 * The boolean type `bool` with values `true` and `false`.
3283 * The machine types (integer and floating-point).
3284 * The machine-dependent integer types.
3288 The machine types are the following:
3290 * The unsigned word types `u8`, `u16`, `u32` and `u64`, with values drawn from
3291 the integer intervals [0, 2^8 - 1], [0, 2^16 - 1], [0, 2^32 - 1] and
3292 [0, 2^64 - 1] respectively.
3294 * The signed two's complement word types `i8`, `i16`, `i32` and `i64`, with
3295 values drawn from the integer intervals [-(2^(7)), 2^7 - 1],
3296 [-(2^(15)), 2^15 - 1], [-(2^(31)), 2^31 - 1], [-(2^(63)), 2^63 - 1]
3299 * The IEEE 754-2008 `binary32` and `binary64` floating-point types: `f32` and
3300 `f64`, respectively.
3302 #### Machine-dependent integer types
3304 The `usize` type is an unsigned integer type with the same number of bits as the
3305 platform's pointer type. It can represent every memory address in the process.
3307 The `isize` type is a signed integer type with the same number of bits as the
3308 platform's pointer type. The theoretical upper bound on object and array size
3309 is the maximum `isize` value. This ensures that `isize` can be used to calculate
3310 differences between pointers into an object or array and can address every byte
3311 within an object along with one byte past the end.
3315 The types `char` and `str` hold textual data.
3317 A value of type `char` is a [Unicode scalar value](
3318 http://www.unicode.org/glossary/#unicode_scalar_value) (i.e. a code point that
3319 is not a surrogate), represented as a 32-bit unsigned word in the 0x0000 to
3320 0xD7FF or 0xE000 to 0x10FFFF range. A `[char]` array is effectively an UCS-4 /
3323 A value of type `str` is a Unicode string, represented as an array of 8-bit
3324 unsigned bytes holding a sequence of UTF-8 code points. Since `str` is of
3325 unknown size, it is not a _first-class_ type, but can only be instantiated
3326 through a pointer type, such as `&str`.
3330 A tuple *type* is a heterogeneous product of other types, called the *elements*
3331 of the tuple. It has no nominal name and is instead structurally typed.
3333 Tuple types and values are denoted by listing the types or values of their
3334 elements, respectively, in a parenthesized, comma-separated list.
3336 Because tuple elements don't have a name, they can only be accessed by
3337 pattern-matching or by using `N` directly as a field to access the
3340 An example of a tuple type and its use:
3343 type Pair<'a> = (i32, &'a str);
3344 let p: Pair<'static> = (10, "ten");
3348 assert_eq!(b, "ten");
3349 assert_eq!(p.0, 10);
3350 assert_eq!(p.1, "ten");
3353 For historical reasons and convenience, the tuple type with no elements (`()`)
3354 is often called ‘unit’ or ‘the unit type’.
3356 ### Array, and Slice types
3358 Rust has two different types for a list of items:
3360 * `[T; N]`, an 'array'
3363 An array has a fixed size, and can be allocated on either the stack or the
3366 A slice is a 'view' into an array. It doesn't own the data it points
3372 // A stack-allocated array
3373 let array: [i32; 3] = [1, 2, 3];
3375 // A heap-allocated array
3376 let vector: Vec<i32> = vec![1, 2, 3];
3378 // A slice into an array
3379 let slice: &[i32] = &vector[..];
3382 As you can see, the `vec!` macro allows you to create a `Vec<T>` easily. The
3383 `vec!` macro is also part of the standard library, rather than the language.
3385 All in-bounds elements of arrays and slices are always initialized, and access
3386 to an array or slice is always bounds-checked.
3390 A `struct` *type* is a heterogeneous product of other types, called the
3391 *fields* of the type.[^structtype]
3393 [^structtype]: `struct` types are analogous to `struct` types in C,
3394 the *record* types of the ML family,
3395 or the *struct* types of the Lisp family.
3397 New instances of a `struct` can be constructed with a [struct
3398 expression](#struct-expressions).
3400 The memory layout of a `struct` is undefined by default to allow for compiler
3401 optimizations like field reordering, but it can be fixed with the
3402 `#[repr(...)]` attribute. In either case, fields may be given in any order in
3403 a corresponding struct *expression*; the resulting `struct` value will always
3404 have the same memory layout.
3406 The fields of a `struct` may be qualified by [visibility
3407 modifiers](#visibility-and-privacy), to allow access to data in a
3408 struct outside a module.
3410 A _tuple struct_ type is just like a struct type, except that the fields are
3413 A _unit-like struct_ type is like a struct type, except that it has no
3414 fields. The one value constructed by the associated [struct
3415 expression](#struct-expressions) is the only value that inhabits such a
3418 ### Enumerated types
3420 An *enumerated type* is a nominal, heterogeneous disjoint union type, denoted
3421 by the name of an [`enum` item](#enumerations). [^enumtype]
3423 [^enumtype]: The `enum` type is analogous to a `data` constructor declaration in
3424 ML, or a *pick ADT* in Limbo.
3426 An [`enum` item](#enumerations) declares both the type and a number of *variant
3427 constructors*, each of which is independently named and takes an optional tuple
3430 New instances of an `enum` can be constructed by calling one of the variant
3431 constructors, in a [call expression](#call-expressions).
3433 Any `enum` value consumes as much memory as the largest variant constructor for
3434 its corresponding `enum` type.
3436 Enum types cannot be denoted *structurally* as types, but must be denoted by
3437 named reference to an [`enum` item](#enumerations).
3441 Nominal types — [enumerations](#enumerated-types) and
3442 [structs](#struct-types) — may be recursive. That is, each `enum`
3443 constructor or `struct` field may refer, directly or indirectly, to the
3444 enclosing `enum` or `struct` type itself. Such recursion has restrictions:
3446 * Recursive types must include a nominal type in the recursion
3447 (not mere [type definitions](grammar.html#type-definitions),
3448 or other structural types such as [arrays](#array-and-slice-types) or [tuples](#tuple-types)).
3449 * A recursive `enum` item must have at least one non-recursive constructor
3450 (in order to give the recursion a basis case).
3451 * The size of a recursive type must be finite;
3452 in other words the recursive fields of the type must be [pointer types](#pointer-types).
3453 * Recursive type definitions can cross module boundaries, but not module *visibility* boundaries,
3454 or crate boundaries (in order to simplify the module system and type checker).
3456 An example of a *recursive* type and its use:
3461 Cons(T, Box<List<T>>)
3464 let a: List<i32> = List::Cons(7, Box::new(List::Cons(13, Box::new(List::Nil))));
3469 All pointers in Rust are explicit first-class values. They can be copied,
3470 stored into data structs, and returned from functions. There are two
3471 varieties of pointer in Rust:
3474 : These point to memory _owned by some other value_.
3475 A reference type is written `&type`,
3476 or `&'a type` when you need to specify an explicit lifetime.
3477 Copying a reference is a "shallow" operation:
3478 it involves only copying the pointer itself.
3479 Releasing a reference has no effect on the value it points to,
3480 but a reference of a temporary value will keep it alive during the scope
3481 of the reference itself.
3483 * Raw pointers (`*`)
3484 : Raw pointers are pointers without safety or liveness guarantees.
3485 Raw pointers are written as `*const T` or `*mut T`,
3486 for example `*const i32` means a raw pointer to a 32-bit integer.
3487 Copying or dropping a raw pointer has no effect on the lifecycle of any
3488 other value. Dereferencing a raw pointer or converting it to any other
3489 pointer type is an [`unsafe` operation](#unsafe-functions).
3490 Raw pointers are generally discouraged in Rust code;
3491 they exist to support interoperability with foreign code,
3492 and writing performance-critical or low-level functions.
3494 The standard library contains additional 'smart pointer' types beyond references
3499 The function type constructor `fn` forms new function types. A function type
3500 consists of a possibly-empty set of function-type modifiers (such as `unsafe`
3501 or `extern`), a sequence of input types and an output type.
3503 An example of a `fn` type:
3506 fn add(x: i32, y: i32) -> i32 {
3510 let mut x = add(5,7);
3512 type Binop = fn(i32, i32) -> i32;
3513 let bo: Binop = add;
3517 #### Function types for specific items
3519 Internal to the compiler, there are also function types that are specific to a particular
3520 function item. In the following snippet, for example, the internal types of the functions
3521 `foo` and `bar` are different, despite the fact that they have the same signature:
3528 The types of `foo` and `bar` can both be implicitly coerced to the fn
3529 pointer type `fn()`. There is currently no syntax for unique fn types,
3530 though the compiler will emit a type like `fn() {foo}` in error
3531 messages to indicate "the unique fn type for the function `foo`".
3535 A [lambda expression](#lambda-expressions) produces a closure value with
3536 a unique, anonymous type that cannot be written out.
3538 Depending on the requirements of the closure, its type implements one or
3539 more of the closure traits:
3542 : The closure can be called once. A closure called as `FnOnce`
3543 can move out values from its environment.
3546 : The closure can be called multiple times as mutable. A closure called as
3547 `FnMut` can mutate values from its environment. `FnMut` inherits from
3548 `FnOnce` (i.e. anything implementing `FnMut` also implements `FnOnce`).
3551 : The closure can be called multiple times through a shared reference.
3552 A closure called as `Fn` can neither move out from nor mutate values
3553 from its environment. `Fn` inherits from `FnMut`, which itself
3554 inherits from `FnOnce`.
3559 In Rust, a type like `&SomeTrait` or `Box<SomeTrait>` is called a _trait object_.
3560 Each instance of a trait object includes:
3562 - a pointer to an instance of a type `T` that implements `SomeTrait`
3563 - a _virtual method table_, often just called a _vtable_, which contains, for
3564 each method of `SomeTrait` that `T` implements, a pointer to `T`'s
3565 implementation (i.e. a function pointer).
3567 The purpose of trait objects is to permit "late binding" of methods. Calling a
3568 method on a trait object results in virtual dispatch at runtime: that is, a
3569 function pointer is loaded from the trait object vtable and invoked indirectly.
3570 The actual implementation for each vtable entry can vary on an object-by-object
3573 Note that for a trait object to be instantiated, the trait must be
3574 _object-safe_. Object safety rules are defined in [RFC 255].
3576 [RFC 255]: https://github.com/rust-lang/rfcs/blob/master/text/0255-object-safety.md
3578 Given a pointer-typed expression `E` of type `&T` or `Box<T>`, where `T`
3579 implements trait `R`, casting `E` to the corresponding pointer type `&R` or
3580 `Box<R>` results in a value of the _trait object_ `R`. This result is
3581 represented as a pair of pointers: the vtable pointer for the `T`
3582 implementation of `R`, and the pointer value of `E`.
3584 An example of a trait object:
3588 fn stringify(&self) -> String;
3591 impl Printable for i32 {
3592 fn stringify(&self) -> String { self.to_string() }
3595 fn print(a: Box<Printable>) {
3596 println!("{}", a.stringify());
3600 print(Box::new(10) as Box<Printable>);
3604 In this example, the trait `Printable` occurs as a trait object in both the
3605 type signature of `print`, and the cast expression in `main`.
3609 Within the body of an item that has type parameter declarations, the names of
3610 its type parameters are types:
3613 fn to_vec<A: Clone>(xs: &[A]) -> Vec<A> {
3617 let first: A = xs[0].clone();
3618 let mut rest: Vec<A> = to_vec(&xs[1..]);
3619 rest.insert(0, first);
3624 Here, `first` has type `A`, referring to `to_vec`'s `A` type parameter; and `rest`
3625 has type `Vec<A>`, a vector with element type `A`.
3629 The special type `Self` has a meaning within traits and impls. In a trait definition, it refers
3630 to an implicit type parameter representing the "implementing" type. In an impl,
3631 it is an alias for the implementing type. For example, in:
3635 fn make_string(&self) -> String;
3638 impl Printable for String {
3639 fn make_string(&self) -> String {
3645 The notation `&self` is a shorthand for `self: &Self`. In this case,
3646 in the impl, `Self` refers to the value of type `String` that is the
3647 receiver for a call to the method `make_string`.
3651 Subtyping is implicit and can occur at any stage in type checking or
3652 inference. Subtyping in Rust is very restricted and occurs only due to
3653 variance with respect to lifetimes and between types with higher ranked
3654 lifetimes. If we were to erase lifetimes from types, then the only subtyping
3655 would be due to type equality.
3657 Consider the following example: string literals always have `'static`
3658 lifetime. Nevertheless, we can assign `s` to `t`:
3662 let s: &'static str = "hi";
3666 Since `'static` "lives longer" than `'a`, `&'static str` is a subtype of
3671 Coercions are defined in [RFC401]. A coercion is implicit and has no syntax.
3673 [RFC401]: https://github.com/rust-lang/rfcs/blob/master/text/0401-coercions.md
3677 A coercion can only occur at certain coercion sites in a program; these are
3678 typically places where the desired type is explicit or can be derived by
3679 propagation from explicit types (without type inference). Possible coercion
3682 * `let` statements where an explicit type is given.
3684 For example, `42` is coerced to have type `i8` in the following:
3690 * `static` and `const` statements (similar to `let` statements).
3692 * Arguments for function calls
3694 The value being coerced is the actual parameter, and it is coerced to
3695 the type of the formal parameter.
3697 For example, `42` is coerced to have type `i8` in the following:
3707 * Instantiations of struct or variant fields
3709 For example, `42` is coerced to have type `i8` in the following:
3712 struct Foo { x: i8 }
3719 * Function results, either the final line of a block if it is not
3720 semicolon-terminated or any expression in a `return` statement
3722 For example, `42` is coerced to have type `i8` in the following:
3730 If the expression in one of these coercion sites is a coercion-propagating
3731 expression, then the relevant sub-expressions in that expression are also
3732 coercion sites. Propagation recurses from these new coercion sites.
3733 Propagating expressions and their relevant sub-expressions are:
3735 * Array literals, where the array has type `[U; n]`. Each sub-expression in
3736 the array literal is a coercion site for coercion to type `U`.
3738 * Array literals with repeating syntax, where the array has type `[U; n]`. The
3739 repeated sub-expression is a coercion site for coercion to type `U`.
3741 * Tuples, where a tuple is a coercion site to type `(U_0, U_1, ..., U_n)`.
3742 Each sub-expression is a coercion site to the respective type, e.g. the
3743 zeroth sub-expression is a coercion site to type `U_0`.
3745 * Parenthesized sub-expressions (`(e)`): if the expression has type `U`, then
3746 the sub-expression is a coercion site to `U`.
3748 * Blocks: if a block has type `U`, then the last expression in the block (if
3749 it is not semicolon-terminated) is a coercion site to `U`. This includes
3750 blocks which are part of control flow statements, such as `if`/`else`, if
3751 the block has a known type.
3755 Coercion is allowed between the following types:
3757 * `T` to `U` if `T` is a subtype of `U` (*reflexive case*)
3759 * `T_1` to `T_3` where `T_1` coerces to `T_2` and `T_2` coerces to `T_3`
3762 Note that this is not fully supported yet
3766 * `*mut T` to `*const T`
3768 * `&T` to `*const T`
3770 * `&mut T` to `*mut T`
3772 * `&T` to `&U` if `T` implements `Deref<Target = U>`. For example:
3775 use std::ops::Deref;
3777 struct CharContainer {
3781 impl Deref for CharContainer {
3784 fn deref<'a>(&'a self) -> &'a char {
3789 fn foo(arg: &char) {}
3792 let x = &mut CharContainer { value: 'y' };
3793 foo(x); //&mut CharContainer is coerced to &char.
3797 * `&mut T` to `&mut U` if `T` implements `DerefMut<Target = U>`.
3799 * TyCtor(`T`) to TyCtor(coerce_inner(`T`)), where TyCtor(`T`) is one of
3807 - coerce_inner(`[T, ..n]`) = `[T]`
3808 - coerce_inner(`T`) = `U` where `T` is a concrete type which implements the
3811 In the future, coerce_inner will be recursively extended to tuples and
3812 structs. In addition, coercions from sub-traits to super-traits will be
3813 added. See [RFC401] for more details.
3817 Several traits define special evaluation behavior.
3821 The `Copy` trait changes the semantics of a type implementing it. Values whose
3822 type implements `Copy` are copied rather than moved upon assignment.
3824 ## The `Sized` trait
3826 The `Sized` trait indicates that the size of this type is known at compile-time.
3830 The `Drop` trait provides a destructor, to be run whenever a value of this type
3833 ## The `Deref` trait
3835 The `Deref<Target = U>` trait allows a type to implicitly implement all the methods
3836 of the type `U`. When attempting to resolve a method call, the compiler will search
3837 the top-level type for the implementation of the called method. If no such method is
3838 found, `.deref()` is called and the compiler continues to search for the method
3839 implementation in the returned type `U`.
3843 A Rust program's memory consists of a static set of *items* and a *heap*.
3844 Immutable portions of the heap may be safely shared between threads, mutable
3845 portions may not be safely shared, but several mechanisms for effectively-safe
3846 sharing of mutable values, built on unsafe code but enforcing a safe locking
3847 discipline, exist in the standard library.
3849 Allocations in the stack consist of *variables*, and allocations in the heap
3852 ### Memory allocation and lifetime
3854 The _items_ of a program are those functions, modules and types that have their
3855 value calculated at compile-time and stored uniquely in the memory image of the
3856 rust process. Items are neither dynamically allocated nor freed.
3858 The _heap_ is a general term that describes boxes. The lifetime of an
3859 allocation in the heap depends on the lifetime of the box values pointing to
3860 it. Since box values may themselves be passed in and out of frames, or stored
3861 in the heap, heap allocations may outlive the frame they are allocated within.
3863 ### Memory ownership
3865 When a stack frame is exited, its local allocations are all released, and its
3866 references to boxes are dropped.
3870 A _variable_ is a component of a stack frame, either a named function parameter,
3871 an anonymous [temporary](#lvalues-rvalues-and-temporaries), or a named local
3874 A _local variable_ (or *stack-local* allocation) holds a value directly,
3875 allocated within the stack's memory. The value is a part of the stack frame.
3877 Local variables are immutable unless declared otherwise like: `let mut x = ...`.
3879 Function parameters are immutable unless declared with `mut`. The `mut` keyword
3880 applies only to the following parameter (so `|mut x, y|` and `fn f(mut x:
3881 Box<i32>, y: Box<i32>)` declare one mutable variable `x` and one immutable
3884 Methods that take either `self` or `Box<Self>` can optionally place them in a
3885 mutable variable by prefixing them with `mut` (similar to regular arguments):
3889 fn change(mut self) -> Self;
3890 fn modify(mut self: Box<Self>) -> Box<Self>;
3894 Local variables are not initialized when allocated; the entire frame worth of
3895 local variables are allocated at once, on frame-entry, in an uninitialized
3896 state. Subsequent statements within a function may or may not initialize the
3897 local variables. Local variables can be used only after they have been
3898 initialized; this is enforced by the compiler.
3902 The Rust compiler supports various methods to link crates together both
3903 statically and dynamically. This section will explore the various methods to
3904 link Rust crates together, and more information about native libraries can be
3905 found in the [FFI section of the book][ffi].
3907 In one session of compilation, the compiler can generate multiple artifacts
3908 through the usage of either command line flags or the `crate_type` attribute.
3909 If one or more command line flags are specified, all `crate_type` attributes will
3910 be ignored in favor of only building the artifacts specified by command line.
3912 * `--crate-type=bin`, `#[crate_type = "bin"]` - A runnable executable will be
3913 produced. This requires that there is a `main` function in the crate which
3914 will be run when the program begins executing. This will link in all Rust and
3915 native dependencies, producing a distributable binary.
3917 * `--crate-type=lib`, `#[crate_type = "lib"]` - A Rust library will be produced.
3918 This is an ambiguous concept as to what exactly is produced because a library
3919 can manifest itself in several forms. The purpose of this generic `lib` option
3920 is to generate the "compiler recommended" style of library. The output library
3921 will always be usable by rustc, but the actual type of library may change from
3922 time-to-time. The remaining output types are all different flavors of
3923 libraries, and the `lib` type can be seen as an alias for one of them (but the
3924 actual one is compiler-defined).
3926 * `--crate-type=dylib`, `#[crate_type = "dylib"]` - A dynamic Rust library will
3927 be produced. This is different from the `lib` output type in that this forces
3928 dynamic library generation. The resulting dynamic library can be used as a
3929 dependency for other libraries and/or executables. This output type will
3930 create `*.so` files on linux, `*.dylib` files on osx, and `*.dll` files on
3933 * `--crate-type=staticlib`, `#[crate_type = "staticlib"]` - A static system
3934 library will be produced. This is different from other library outputs in that
3935 the Rust compiler will never attempt to link to `staticlib` outputs. The
3936 purpose of this output type is to create a static library containing all of
3937 the local crate's code along with all upstream dependencies. The static
3938 library is actually a `*.a` archive on linux and osx and a `*.lib` file on
3939 windows. This format is recommended for use in situations such as linking
3940 Rust code into an existing non-Rust application because it will not have
3941 dynamic dependencies on other Rust code.
3943 * `--crate-type=rlib`, `#[crate_type = "rlib"]` - A "Rust library" file will be
3944 produced. This is used as an intermediate artifact and can be thought of as a
3945 "static Rust library". These `rlib` files, unlike `staticlib` files, are
3946 interpreted by the Rust compiler in future linkage. This essentially means
3947 that `rustc` will look for metadata in `rlib` files like it looks for metadata
3948 in dynamic libraries. This form of output is used to produce statically linked
3949 executables as well as `staticlib` outputs.
3951 Note that these outputs are stackable in the sense that if multiple are
3952 specified, then the compiler will produce each form of output at once without
3953 having to recompile. However, this only applies for outputs specified by the
3954 same method. If only `crate_type` attributes are specified, then they will all
3955 be built, but if one or more `--crate-type` command line flags are specified,
3956 then only those outputs will be built.
3958 With all these different kinds of outputs, if crate A depends on crate B, then
3959 the compiler could find B in various different forms throughout the system. The
3960 only forms looked for by the compiler, however, are the `rlib` format and the
3961 dynamic library format. With these two options for a dependent library, the
3962 compiler must at some point make a choice between these two formats. With this
3963 in mind, the compiler follows these rules when determining what format of
3964 dependencies will be used:
3966 1. If a static library is being produced, all upstream dependencies are
3967 required to be available in `rlib` formats. This requirement stems from the
3968 reason that a dynamic library cannot be converted into a static format.
3970 Note that it is impossible to link in native dynamic dependencies to a static
3971 library, and in this case warnings will be printed about all unlinked native
3972 dynamic dependencies.
3974 2. If an `rlib` file is being produced, then there are no restrictions on what
3975 format the upstream dependencies are available in. It is simply required that
3976 all upstream dependencies be available for reading metadata from.
3978 The reason for this is that `rlib` files do not contain any of their upstream
3979 dependencies. It wouldn't be very efficient for all `rlib` files to contain a
3980 copy of `libstd.rlib`!
3982 3. If an executable is being produced and the `-C prefer-dynamic` flag is not
3983 specified, then dependencies are first attempted to be found in the `rlib`
3984 format. If some dependencies are not available in an rlib format, then
3985 dynamic linking is attempted (see below).
3987 4. If a dynamic library or an executable that is being dynamically linked is
3988 being produced, then the compiler will attempt to reconcile the available
3989 dependencies in either the rlib or dylib format to create a final product.
3991 A major goal of the compiler is to ensure that a library never appears more
3992 than once in any artifact. For example, if dynamic libraries B and C were
3993 each statically linked to library A, then a crate could not link to B and C
3994 together because there would be two copies of A. The compiler allows mixing
3995 the rlib and dylib formats, but this restriction must be satisfied.
3997 The compiler currently implements no method of hinting what format a library
3998 should be linked with. When dynamically linking, the compiler will attempt to
3999 maximize dynamic dependencies while still allowing some dependencies to be
4000 linked in via an rlib.
4002 For most situations, having all libraries available as a dylib is recommended
4003 if dynamically linking. For other situations, the compiler will emit a
4004 warning if it is unable to determine which formats to link each library with.
4006 In general, `--crate-type=bin` or `--crate-type=lib` should be sufficient for
4007 all compilation needs, and the other options are just available if more
4008 fine-grained control is desired over the output format of a Rust crate.
4012 Unsafe operations are those that potentially violate the memory-safety
4013 guarantees of Rust's static semantics.
4015 The following language level features cannot be used in the safe subset of
4018 - Dereferencing a [raw pointer](#pointer-types).
4019 - Reading or writing a [mutable static variable](#mutable-statics).
4020 - Calling an unsafe function (including an intrinsic or foreign function).
4024 Unsafe functions are functions that are not safe in all contexts and/or for all
4025 possible inputs. Such a function must be prefixed with the keyword `unsafe` and
4026 can only be called from an `unsafe` block or another `unsafe` function.
4030 A block of code can be prefixed with the `unsafe` keyword, to permit calling
4031 `unsafe` functions or dereferencing raw pointers within a safe function.
4033 When a programmer has sufficient conviction that a sequence of potentially
4034 unsafe operations is actually safe, they can encapsulate that sequence (taken
4035 as a whole) within an `unsafe` block. The compiler will consider uses of such
4036 code safe, in the surrounding context.
4038 Unsafe blocks are used to wrap foreign libraries, make direct use of hardware
4039 or implement features not directly present in the language. For example, Rust
4040 provides the language features necessary to implement memory-safe concurrency
4041 in the language but the implementation of threads and message passing is in the
4044 Rust's type system is a conservative approximation of the dynamic safety
4045 requirements, so in some cases there is a performance cost to using safe code.
4046 For example, a doubly-linked list is not a tree structure and can only be
4047 represented with reference-counted pointers in safe code. By using `unsafe`
4048 blocks to represent the reverse links as raw pointers, it can be implemented
4051 ## Behavior considered undefined
4053 The following is a list of behavior which is forbidden in all Rust code,
4054 including within `unsafe` blocks and `unsafe` functions. Type checking provides
4055 the guarantee that these issues are never caused by safe code.
4058 * Dereferencing a null/dangling raw pointer
4059 * Reads of [undef](http://llvm.org/docs/LangRef.html#undefined-values)
4060 (uninitialized) memory
4061 * Breaking the [pointer aliasing
4062 rules](http://llvm.org/docs/LangRef.html#pointer-aliasing-rules)
4063 with raw pointers (a subset of the rules used by C)
4064 * `&mut` and `&` follow LLVM’s scoped [noalias] model, except if the `&T`
4065 contains an `UnsafeCell<U>`. Unsafe code must not violate these aliasing
4067 * Mutating non-mutable data (that is, data reached through a shared reference or
4068 data owned by a `let` binding), unless that data is contained within an `UnsafeCell<U>`.
4069 * Invoking undefined behavior via compiler intrinsics:
4070 * Indexing outside of the bounds of an object with `std::ptr::offset`
4071 (`offset` intrinsic), with
4072 the exception of one byte past the end which is permitted.
4073 * Using `std::ptr::copy_nonoverlapping_memory` (`memcpy32`/`memcpy64`
4074 intrinsics) on overlapping buffers
4075 * Invalid values in primitive types, even in private fields/locals:
4076 * Dangling/null references or boxes
4077 * A value other than `false` (0) or `true` (1) in a `bool`
4078 * A discriminant in an `enum` not included in the type definition
4079 * A value in a `char` which is a surrogate or above `char::MAX`
4080 * Non-UTF-8 byte sequences in a `str`
4081 * Unwinding into Rust from foreign code or unwinding from Rust into foreign
4082 code. Rust's failure system is not compatible with exception handling in
4083 other languages. Unwinding must be caught and handled at FFI boundaries.
4085 [noalias]: http://llvm.org/docs/LangRef.html#noalias
4087 ## Behavior not considered unsafe
4089 This is a list of behavior not considered *unsafe* in Rust terms, but that may
4093 * Leaks of memory and other resources
4094 * Exiting without calling destructors
4096 - Overflow is considered "unexpected" behavior and is always user-error,
4097 unless the `wrapping` primitives are used. In non-optimized builds, the compiler
4098 will insert debug checks that panic on overflow, but in optimized builds overflow
4099 instead results in wrapped values. See [RFC 560] for the rationale and more details.
4101 [RFC 560]: https://github.com/rust-lang/rfcs/blob/master/text/0560-integer-overflow.md
4103 # Appendix: Influences
4105 Rust is not a particularly original language, with design elements coming from
4106 a wide range of sources. Some of these are listed below (including elements
4107 that have since been removed):
4109 * SML, OCaml: algebraic data types, pattern matching, type inference,
4110 semicolon statement separation
4111 * C++: references, RAII, smart pointers, move semantics, monomorphization,
4113 * ML Kit, Cyclone: region based memory management
4114 * Haskell (GHC): typeclasses, type families
4115 * Newsqueak, Alef, Limbo: channels, concurrency
4116 * Erlang: message passing, thread failure, ~~linked thread failure~~,
4117 ~~lightweight concurrency~~
4118 * Swift: optional bindings
4119 * Scheme: hygienic macros
4121 * Ruby: ~~block syntax~~
4122 * NIL, Hermes: ~~typestate~~
4123 * [Unicode Annex #31](http://www.unicode.org/reports/tr31/): identifier and
4126 [ffi]: book/ffi.html
4127 [plugin]: book/compiler-plugins.html