]> git.proxmox.com Git - rustc.git/blob - src/doc/trpl/macros.md
Imported Upstream version 1.0.0+dfsg1
[rustc.git] / src / doc / trpl / macros.md
1 % Macros
2
3 By now you’ve learned about many of the tools Rust provides for abstracting and
4 reusing code. These units of code reuse have a rich semantic structure. For
5 example, functions have a type signature, type parameters have trait bounds,
6 and overloaded functions must belong to a particular trait.
7
8 This structure means that Rust’s core abstractions have powerful compile-time
9 correctness checking. But this comes at the price of reduced flexibility. If
10 you visually identify a pattern of repeated code, you may find it’s difficult
11 or cumbersome to express that pattern as a generic function, a trait, or
12 anything else within Rust’s semantics.
13
14 Macros allow us to abstract at a syntactic level. A macro invocation is
15 shorthand for an "expanded" syntactic form. This expansion happens early in
16 compilation, before any static checking. As a result, macros can capture many
17 patterns of code reuse that Rust’s core abstractions cannot.
18
19 The drawback is that macro-based code can be harder to understand, because
20 fewer of the built-in rules apply. Like an ordinary function, a well-behaved
21 macro can be used without understanding its implementation. However, it can be
22 difficult to design a well-behaved macro! Additionally, compiler errors in
23 macro code are harder to interpret, because they describe problems in the
24 expanded code, not the source-level form that developers use.
25
26 These drawbacks make macros something of a "feature of last resort". That’s not
27 to say that macros are bad; they are part of Rust because sometimes they’re
28 needed for truly concise, well-abstracted code. Just keep this tradeoff in
29 mind.
30
31 # Defining a macro
32
33 You may have seen the `vec!` macro, used to initialize a [vector][] with any
34 number of elements.
35
36 [vector]: vectors.html
37
38 ```rust
39 let x: Vec<u32> = vec![1, 2, 3];
40 # assert_eq!(x, [1, 2, 3]);
41 ```
42
43 This can’t be an ordinary function, because it takes any number of arguments.
44 But we can imagine it as syntactic shorthand for
45
46 ```rust
47 let x: Vec<u32> = {
48 let mut temp_vec = Vec::new();
49 temp_vec.push(1);
50 temp_vec.push(2);
51 temp_vec.push(3);
52 temp_vec
53 };
54 # assert_eq!(x, [1, 2, 3]);
55 ```
56
57 We can implement this shorthand, using a macro: [^actual]
58
59 [^actual]: The actual definition of `vec!` in libcollections differs from the
60 one presented here, for reasons of efficiency and reusability.
61
62 ```rust
63 macro_rules! vec {
64 ( $( $x:expr ),* ) => {
65 {
66 let mut temp_vec = Vec::new();
67 $(
68 temp_vec.push($x);
69 )*
70 temp_vec
71 }
72 };
73 }
74 # fn main() {
75 # assert_eq!(vec![1,2,3], [1, 2, 3]);
76 # }
77 ```
78
79 Whoa, that’s a lot of new syntax! Let’s break it down.
80
81 ```ignore
82 macro_rules! vec { ... }
83 ```
84
85 This says we’re defining a macro named `vec`, much as `fn vec` would define a
86 function named `vec`. In prose, we informally write a macro’s name with an
87 exclamation point, e.g. `vec!`. The exclamation point is part of the invocation
88 syntax and serves to distinguish a macro from an ordinary function.
89
90 ## Matching
91
92 The macro is defined through a series of rules, which are pattern-matching
93 cases. Above, we had
94
95 ```ignore
96 ( $( $x:expr ),* ) => { ... };
97 ```
98
99 This is like a `match` expression arm, but the matching happens on Rust syntax
100 trees, at compile time. The semicolon is optional on the last (here, only)
101 case. The "pattern" on the left-hand side of `=>` is known as a ‘matcher’.
102 These have [their own little grammar] within the language.
103
104 [their own little grammar]: ../reference.html#macros
105
106 The matcher `$x:expr` will match any Rust expression, binding that syntax tree
107 to the ‘metavariable’ `$x`. The identifier `expr` is a ‘fragment specifier’;
108 the full possibilities are enumerated later in this chapter.
109 Surrounding the matcher with `$(...),*` will match zero or more expressions,
110 separated by commas.
111
112 Aside from the special matcher syntax, any Rust tokens that appear in a matcher
113 must match exactly. For example,
114
115 ```rust
116 macro_rules! foo {
117 (x => $e:expr) => (println!("mode X: {}", $e));
118 (y => $e:expr) => (println!("mode Y: {}", $e));
119 }
120
121 fn main() {
122 foo!(y => 3);
123 }
124 ```
125
126 will print
127
128 ```text
129 mode Y: 3
130 ```
131
132 With
133
134 ```rust,ignore
135 foo!(z => 3);
136 ```
137
138 we get the compiler error
139
140 ```text
141 error: no rules expected the token `z`
142 ```
143
144 ## Expansion
145
146 The right-hand side of a macro rule is ordinary Rust syntax, for the most part.
147 But we can splice in bits of syntax captured by the matcher. From the original
148 example:
149
150 ```ignore
151 $(
152 temp_vec.push($x);
153 )*
154 ```
155
156 Each matched expression `$x` will produce a single `push` statement in the
157 macro expansion. The repetition in the expansion proceeds in "lockstep" with
158 repetition in the matcher (more on this in a moment).
159
160 Because `$x` was already declared as matching an expression, we don’t repeat
161 `:expr` on the right-hand side. Also, we don’t include a separating comma as
162 part of the repetition operator. Instead, we have a terminating semicolon
163 within the repeated block.
164
165 Another detail: the `vec!` macro has *two* pairs of braces on the right-hand
166 side. They are often combined like so:
167
168 ```ignore
169 macro_rules! foo {
170 () => {{
171 ...
172 }}
173 }
174 ```
175
176 The outer braces are part of the syntax of `macro_rules!`. In fact, you can use
177 `()` or `[]` instead. They simply delimit the right-hand side as a whole.
178
179 The inner braces are part of the expanded syntax. Remember, the `vec!` macro is
180 used in an expression context. To write an expression with multiple statements,
181 including `let`-bindings, we use a block. If your macro expands to a single
182 expression, you don’t need this extra layer of braces.
183
184 Note that we never *declared* that the macro produces an expression. In fact,
185 this is not determined until we use the macro as an expression. With care, you
186 can write a macro whose expansion works in several contexts. For example,
187 shorthand for a data type could be valid as either an expression or a pattern.
188
189 ## Repetition
190
191 The repetition operator follows two principal rules:
192
193 1. `$(...)*` walks through one "layer" of repetitions, for all of the `$name`s
194 it contains, in lockstep, and
195 2. each `$name` must be under at least as many `$(...)*`s as it was matched
196 against. If it is under more, it’ll be duplicated, as appropriate.
197
198 This baroque macro illustrates the duplication of variables from outer
199 repetition levels.
200
201 ```rust
202 macro_rules! o_O {
203 (
204 $(
205 $x:expr; [ $( $y:expr ),* ]
206 );*
207 ) => {
208 &[ $($( $x + $y ),*),* ]
209 }
210 }
211
212 fn main() {
213 let a: &[i32]
214 = o_O!(10; [1, 2, 3];
215 20; [4, 5, 6]);
216
217 assert_eq!(a, [11, 12, 13, 24, 25, 26]);
218 }
219 ```
220
221 That’s most of the matcher syntax. These examples use `$(...)*`, which is a
222 "zero or more" match. Alternatively you can write `$(...)+` for a "one or
223 more" match. Both forms optionally include a separator, which can be any token
224 except `+` or `*`.
225
226 This system is based on
227 "[Macro-by-Example](http://www.cs.indiana.edu/ftp/techreports/TR206.pdf)"
228 (PDF link).
229
230 # Hygiene
231
232 Some languages implement macros using simple text substitution, which leads to
233 various problems. For example, this C program prints `13` instead of the
234 expected `25`.
235
236 ```text
237 #define FIVE_TIMES(x) 5 * x
238
239 int main() {
240 printf("%d\n", FIVE_TIMES(2 + 3));
241 return 0;
242 }
243 ```
244
245 After expansion we have `5 * 2 + 3`, and multiplication has greater precedence
246 than addition. If you’ve used C macros a lot, you probably know the standard
247 idioms for avoiding this problem, as well as five or six others. In Rust, we
248 don’t have to worry about it.
249
250 ```rust
251 macro_rules! five_times {
252 ($x:expr) => (5 * $x);
253 }
254
255 fn main() {
256 assert_eq!(25, five_times!(2 + 3));
257 }
258 ```
259
260 The metavariable `$x` is parsed as a single expression node, and keeps its
261 place in the syntax tree even after substitution.
262
263 Another common problem in macro systems is ‘variable capture’. Here’s a C
264 macro, using [a GNU C extension] to emulate Rust’s expression blocks.
265
266 [a GNU C extension]: https://gcc.gnu.org/onlinedocs/gcc/Statement-Exprs.html
267
268 ```text
269 #define LOG(msg) ({ \
270 int state = get_log_state(); \
271 if (state > 0) { \
272 printf("log(%d): %s\n", state, msg); \
273 } \
274 })
275 ```
276
277 Here’s a simple use case that goes terribly wrong:
278
279 ```text
280 const char *state = "reticulating splines";
281 LOG(state)
282 ```
283
284 This expands to
285
286 ```text
287 const char *state = "reticulating splines";
288 int state = get_log_state();
289 if (state > 0) {
290 printf("log(%d): %s\n", state, state);
291 }
292 ```
293
294 The second variable named `state` shadows the first one. This is a problem
295 because the print statement should refer to both of them.
296
297 The equivalent Rust macro has the desired behavior.
298
299 ```rust
300 # fn get_log_state() -> i32 { 3 }
301 macro_rules! log {
302 ($msg:expr) => {{
303 let state: i32 = get_log_state();
304 if state > 0 {
305 println!("log({}): {}", state, $msg);
306 }
307 }};
308 }
309
310 fn main() {
311 let state: &str = "reticulating splines";
312 log!(state);
313 }
314 ```
315
316 This works because Rust has a [hygienic macro system][]. Each macro expansion
317 happens in a distinct ‘syntax context’, and each variable is tagged with the
318 syntax context where it was introduced. It’s as though the variable `state`
319 inside `main` is painted a different "color" from the variable `state` inside
320 the macro, and therefore they don’t conflict.
321
322 [hygienic macro system]: http://en.wikipedia.org/wiki/Hygienic_macro
323
324 This also restricts the ability of macros to introduce new bindings at the
325 invocation site. Code such as the following will not work:
326
327 ```rust,ignore
328 macro_rules! foo {
329 () => (let x = 3);
330 }
331
332 fn main() {
333 foo!();
334 println!("{}", x);
335 }
336 ```
337
338 Instead you need to pass the variable name into the invocation, so it’s tagged
339 with the right syntax context.
340
341 ```rust
342 macro_rules! foo {
343 ($v:ident) => (let $v = 3);
344 }
345
346 fn main() {
347 foo!(x);
348 println!("{}", x);
349 }
350 ```
351
352 This holds for `let` bindings and loop labels, but not for [items][].
353 So the following code does compile:
354
355 ```rust
356 macro_rules! foo {
357 () => (fn x() { });
358 }
359
360 fn main() {
361 foo!();
362 x();
363 }
364 ```
365
366 [items]: ../reference.html#items
367
368 # Recursive macros
369
370 A macro’s expansion can include more macro invocations, including invocations
371 of the very same macro being expanded. These recursive macros are useful for
372 processing tree-structured input, as illustrated by this (simplistic) HTML
373 shorthand:
374
375 ```rust
376 # #![allow(unused_must_use)]
377 macro_rules! write_html {
378 ($w:expr, ) => (());
379
380 ($w:expr, $e:tt) => (write!($w, "{}", $e));
381
382 ($w:expr, $tag:ident [ $($inner:tt)* ] $($rest:tt)*) => {{
383 write!($w, "<{}>", stringify!($tag));
384 write_html!($w, $($inner)*);
385 write!($w, "</{}>", stringify!($tag));
386 write_html!($w, $($rest)*);
387 }};
388 }
389
390 fn main() {
391 # // FIXME(#21826)
392 use std::fmt::Write;
393 let mut out = String::new();
394
395 write_html!(&mut out,
396 html[
397 head[title["Macros guide"]]
398 body[h1["Macros are the best!"]]
399 ]);
400
401 assert_eq!(out,
402 "<html><head><title>Macros guide</title></head>\
403 <body><h1>Macros are the best!</h1></body></html>");
404 }
405 ```
406
407 # Debugging macro code
408
409 To see the results of expanding macros, run `rustc --pretty expanded`. The
410 output represents a whole crate, so you can also feed it back in to `rustc`,
411 which will sometimes produce better error messages than the original
412 compilation. Note that the `--pretty expanded` output may have a different
413 meaning if multiple variables of the same name (but different syntax contexts)
414 are in play in the same scope. In this case `--pretty expanded,hygiene` will
415 tell you about the syntax contexts.
416
417 `rustc` provides two syntax extensions that help with macro debugging. For now,
418 they are unstable and require feature gates.
419
420 * `log_syntax!(...)` will print its arguments to standard output, at compile
421 time, and "expand" to nothing.
422
423 * `trace_macros!(true)` will enable a compiler message every time a macro is
424 expanded. Use `trace_macros!(false)` later in expansion to turn it off.
425
426 # Syntactic requirements
427
428 Even when Rust code contains un-expanded macros, it can be parsed as a full
429 [syntax tree][ast]. This property can be very useful for editors and other
430 tools that process code. It also has a few consequences for the design of
431 Rust’s macro system.
432
433 [ast]: glossary.html#abstract-syntax-tree
434
435 One consequence is that Rust must determine, when it parses a macro invocation,
436 whether the macro stands in for
437
438 * zero or more items,
439 * zero or more methods,
440 * an expression,
441 * a statement, or
442 * a pattern.
443
444 A macro invocation within a block could stand for some items, or for an
445 expression / statement. Rust uses a simple rule to resolve this ambiguity. A
446 macro invocation that stands for items must be either
447
448 * delimited by curly braces, e.g. `foo! { ... }`, or
449 * terminated by a semicolon, e.g. `foo!(...);`
450
451 Another consequence of pre-expansion parsing is that the macro invocation must
452 consist of valid Rust tokens. Furthermore, parentheses, brackets, and braces
453 must be balanced within a macro invocation. For example, `foo!([)` is
454 forbidden. This allows Rust to know where the macro invocation ends.
455
456 More formally, the macro invocation body must be a sequence of ‘token trees’.
457 A token tree is defined recursively as either
458
459 * a sequence of token trees surrounded by matching `()`, `[]`, or `{}`, or
460 * any other single token.
461
462 Within a matcher, each metavariable has a ‘fragment specifier’, identifying
463 which syntactic form it matches.
464
465 * `ident`: an identifier. Examples: `x`; `foo`.
466 * `path`: a qualified name. Example: `T::SpecialA`.
467 * `expr`: an expression. Examples: `2 + 2`; `if true then { 1 } else { 2 }`; `f(42)`.
468 * `ty`: a type. Examples: `i32`; `Vec<(char, String)>`; `&T`.
469 * `pat`: a pattern. Examples: `Some(t)`; `(17, 'a')`; `_`.
470 * `stmt`: a single statement. Example: `let x = 3`.
471 * `block`: a brace-delimited sequence of statements. Example:
472 `{ log(error, "hi"); return 12; }`.
473 * `item`: an [item][]. Examples: `fn foo() { }`; `struct Bar;`.
474 * `meta`: a "meta item", as found in attributes. Example: `cfg(target_os = "windows")`.
475 * `tt`: a single token tree.
476
477 There are additional rules regarding the next token after a metavariable:
478
479 * `expr` variables must be followed by one of: `=> , ;`
480 * `ty` and `path` variables must be followed by one of: `=> , : = > as`
481 * `pat` variables must be followed by one of: `=> , =`
482 * Other variables may be followed by any token.
483
484 These rules provide some flexibility for Rust’s syntax to evolve without
485 breaking existing macros.
486
487 The macro system does not deal with parse ambiguity at all. For example, the
488 grammar `$($t:ty)* $e:expr` will always fail to parse, because the parser would
489 be forced to choose between parsing `$t` and parsing `$e`. Changing the
490 invocation syntax to put a distinctive token in front can solve the problem. In
491 this case, you can write `$(T $t:ty)* E $e:exp`.
492
493 [item]: ../reference.html#items
494
495 # Scoping and macro import/export
496
497 Macros are expanded at an early stage in compilation, before name resolution.
498 One downside is that scoping works differently for macros, compared to other
499 constructs in the language.
500
501 Definition and expansion of macros both happen in a single depth-first,
502 lexical-order traversal of a crate’s source. So a macro defined at module scope
503 is visible to any subsequent code in the same module, which includes the body
504 of any subsequent child `mod` items.
505
506 A macro defined within the body of a single `fn`, or anywhere else not at
507 module scope, is visible only within that item.
508
509 If a module has the `macro_use` attribute, its macros are also visible in its
510 parent module after the child’s `mod` item. If the parent also has `macro_use`
511 then the macros will be visible in the grandparent after the parent’s `mod`
512 item, and so forth.
513
514 The `macro_use` attribute can also appear on `extern crate`. In this context
515 it controls which macros are loaded from the external crate, e.g.
516
517 ```rust,ignore
518 #[macro_use(foo, bar)]
519 extern crate baz;
520 ```
521
522 If the attribute is given simply as `#[macro_use]`, all macros are loaded. If
523 there is no `#[macro_use]` attribute then no macros are loaded. Only macros
524 defined with the `#[macro_export]` attribute may be loaded.
525
526 To load a crate’s macros without linking it into the output, use `#[no_link]`
527 as well.
528
529 An example:
530
531 ```rust
532 macro_rules! m1 { () => (()) }
533
534 // visible here: m1
535
536 mod foo {
537 // visible here: m1
538
539 #[macro_export]
540 macro_rules! m2 { () => (()) }
541
542 // visible here: m1, m2
543 }
544
545 // visible here: m1
546
547 macro_rules! m3 { () => (()) }
548
549 // visible here: m1, m3
550
551 #[macro_use]
552 mod bar {
553 // visible here: m1, m3
554
555 macro_rules! m4 { () => (()) }
556
557 // visible here: m1, m3, m4
558 }
559
560 // visible here: m1, m3, m4
561 # fn main() { }
562 ```
563
564 When this library is loaded with `#[macro_use] extern crate`, only `m2` will
565 be imported.
566
567 The Rust Reference has a [listing of macro-related
568 attributes](../reference.html#macro-related-attributes).
569
570 # The variable `$crate`
571
572 A further difficulty occurs when a macro is used in multiple crates. Say that
573 `mylib` defines
574
575 ```rust
576 pub fn increment(x: u32) -> u32 {
577 x + 1
578 }
579
580 #[macro_export]
581 macro_rules! inc_a {
582 ($x:expr) => ( ::increment($x) )
583 }
584
585 #[macro_export]
586 macro_rules! inc_b {
587 ($x:expr) => ( ::mylib::increment($x) )
588 }
589 # fn main() { }
590 ```
591
592 `inc_a` only works within `mylib`, while `inc_b` only works outside the
593 library. Furthermore, `inc_b` will break if the user imports `mylib` under
594 another name.
595
596 Rust does not (yet) have a hygiene system for crate references, but it does
597 provide a simple workaround for this problem. Within a macro imported from a
598 crate named `foo`, the special macro variable `$crate` will expand to `::foo`.
599 By contrast, when a macro is defined and then used in the same crate, `$crate`
600 will expand to nothing. This means we can write
601
602 ```rust
603 #[macro_export]
604 macro_rules! inc {
605 ($x:expr) => ( $crate::increment($x) )
606 }
607 # fn main() { }
608 ```
609
610 to define a single macro that works both inside and outside our library. The
611 function name will expand to either `::increment` or `::mylib::increment`.
612
613 To keep this system simple and correct, `#[macro_use] extern crate ...` may
614 only appear at the root of your crate, not inside `mod`. This ensures that
615 `$crate` is a single identifier.
616
617 # The deep end
618
619 The introductory chapter mentioned recursive macros, but it did not give the
620 full story. Recursive macros are useful for another reason: Each recursive
621 invocation gives you another opportunity to pattern-match the macro’s
622 arguments.
623
624 As an extreme example, it is possible, though hardly advisable, to implement
625 the [Bitwise Cyclic Tag](http://esolangs.org/wiki/Bitwise_Cyclic_Tag) automaton
626 within Rust’s macro system.
627
628 ```rust
629 macro_rules! bct {
630 // cmd 0: d ... => ...
631 (0, $($ps:tt),* ; $_d:tt)
632 => (bct!($($ps),*, 0 ; ));
633 (0, $($ps:tt),* ; $_d:tt, $($ds:tt),*)
634 => (bct!($($ps),*, 0 ; $($ds),*));
635
636 // cmd 1p: 1 ... => 1 ... p
637 (1, $p:tt, $($ps:tt),* ; 1)
638 => (bct!($($ps),*, 1, $p ; 1, $p));
639 (1, $p:tt, $($ps:tt),* ; 1, $($ds:tt),*)
640 => (bct!($($ps),*, 1, $p ; 1, $($ds),*, $p));
641
642 // cmd 1p: 0 ... => 0 ...
643 (1, $p:tt, $($ps:tt),* ; $($ds:tt),*)
644 => (bct!($($ps),*, 1, $p ; $($ds),*));
645
646 // halt on empty data string
647 ( $($ps:tt),* ; )
648 => (());
649 }
650 ```
651
652 Exercise: use macros to reduce duplication in the above definition of the
653 `bct!` macro.
654
655 # Common macros
656
657 Here are some common macros you’ll see in Rust code.
658
659 ## panic!
660
661 This macro causes the current thread to panic. You can give it a message
662 to panic with:
663
664 ```rust,no_run
665 panic!("oh no!");
666 ```
667
668 ## vec!
669
670 The `vec!` macro is used throughout the book, so you’ve probably seen it
671 already. It creates `Vec<T>`s with ease:
672
673 ```rust
674 let v = vec![1, 2, 3, 4, 5];
675 ```
676
677 It also lets you make vectors with repeating values. For example, a hundred
678 zeroes:
679
680 ```rust
681 let v = vec![0; 100];
682 ```
683
684 ## assert! and assert_eq!
685
686 These two macros are used in tests. `assert!` takes a boolean, and `assert_eq!`
687 takes two values and compares them. Truth passes, success `panic!`s. Like
688 this:
689
690 ```rust,no_run
691 // A-ok!
692
693 assert!(true);
694 assert_eq!(5, 3 + 2);
695
696 // nope :(
697
698 assert!(5 < 3);
699 assert_eq!(5, 3);
700 ```
701 ## try!
702
703 `try!` is used for error handling. It takes something that can return a
704 `Result<T, E>`, and gives `T` if it’s a `Ok<T>`, and `return`s with the
705 `Err(E)` if it’s that. Like this:
706
707 ```rust,no_run
708 use std::fs::File;
709
710 fn foo() -> std::io::Result<()> {
711 let f = try!(File::create("foo.txt"));
712
713 Ok(())
714 }
715 ```
716
717 This is cleaner than doing this:
718
719 ```rust,no_run
720 use std::fs::File;
721
722 fn foo() -> std::io::Result<()> {
723 let f = File::create("foo.txt");
724
725 let f = match f {
726 Ok(t) => t,
727 Err(e) => return Err(e),
728 };
729
730 Ok(())
731 }
732 ```
733
734 ## unreachable!
735
736 This macro is used when you think some code should never execute:
737
738 ```rust
739 if false {
740 unreachable!();
741 }
742 ```
743
744 Sometimes, the compiler may make you have a different branch that you know
745 will never, ever run. In these cases, use this macro, so that if you end
746 up wrong, you’ll get a `panic!` about it.
747
748 ```rust
749 let x: Option<i32> = None;
750
751 match x {
752 Some(_) => unreachable!(),
753 None => println!("I know x is None!"),
754 }
755 ```
756
757 ## unimplemented!
758
759 The `unimplemented!` macro can be used when you’re trying to get your functions
760 to typecheck, and don’t want to worry about writing out the body of the
761 function. One example of this situation is implementing a trait with multiple
762 required methods, where you want to tackle one at a time. Define the others
763 as `unimplemented!` until you’re ready to write them.
764
765 # Procedural macros
766
767 If Rust’s macro system can’t do what you need, you may want to write a
768 [compiler plugin](compiler-plugins.html) instead. Compared to `macro_rules!`
769 macros, this is significantly more work, the interfaces are much less stable,
770 and bugs can be much harder to track down. In exchange you get the
771 flexibility of running arbitrary Rust code within the compiler. Syntax
772 extension plugins are sometimes called ‘procedural macros’ for this reason.