Syntax of pest grammars

pest grammars are lists of rules. Rules are defined like this:

//! Grammar doc
my_rule = { ... }

/// Rule doc
another_rule = {        // comments are preceded by two slashes
    ...                 // whitespace goes anywhere
}

Since rule names are translated into Rust enum variants, they are not allowed to be Rust keywords.

The left curly bracket { defining a rule can be preceded by symbols that affect its operation:

silent_rule = _{ ... }
atomic_rule = @{ ... }

Expressions

Grammar rules are built from expressions (hence "parsing expression grammar"). These expressions are a terse, formal description of how to parse an input string.

Expressions are composable: they can be built out of other expressions and nested inside of each other to produce arbitrarily complex rules (although you should break very complicated expressions into multiple rules to make them easier to manage).

PEG expressions are suitable for both high-level meaning, like "a function signature, followed by a function body", and low-level meaning, like "a semicolon, followed by a line feed". The combining form "followed by", the sequence operator, is the same in either case.

Terminals

The most basic rule is a literal string in double quotes: "text".

A string can be case-insensitive (for ASCII characters only) if preceded by a caret: ^"text".

A single character in a range is written as two single-quoted characters, separated by two dots: '0'..'9'.

You can match any single character at all with the special rule ANY. This is equivalent to '\u{00}'..'\u{10FFFF}', any single Unicode character.

"a literal string"
^"ASCII case-insensitive string"
'a'..'z'
ANY

Finally, you can refer to other rules by writing their names directly, and even use rules recursively:

my_rule = { "slithy " ~ other_rule }
other_rule = { "toves" }
recursive_rule = { "mimsy " ~ recursive_rule }

Sequence

The sequence operator is written as a tilde ~.

first ~ and_then

("abc") ~ (^"def") ~ ('g'..'z')        // matches "abcDEFr"

When matching a sequence expression, first is attempted. If first matches successfully, and_then is attempted next. However, if first fails, the entire expression fails.

A list of expressions can be chained together with sequences, which indicates that all of the components must occur, in the specified order.

Ordered choice

The choice operator is written as a vertical line |.

first | or_else

("abc") | (^"def") | ('g'..'z')        // matches "DEF"

When matching a choice expression, first is attempted. If first matches successfully, the entire expression succeeds immediately. However, if first fails, or_else is attempted next.

Note that first and or_else are always attempted at the same position, even if first matched some input before it failed. When encountering a parse failure, the engine will try the next ordered choice as though no input had been matched. Failed parses never consume any input.

start = { "Beware " ~ creature }
creature = {
    ("the " ~ "Jabberwock")
    | ("the " ~ "Jubjub bird")
}
"Beware the Jubjub bird"
 ^ (start) Parses via the second choice of `creature`,
           even though the first choice matched "the " successfully.

It is somewhat tempting to borrow terminology and think of this operation as "alternation" or simply "OR", but this is misleading. The word "choice" is used specifically because the operation is not merely logical "OR".

Repetition

There are two repetition operators: the asterisk * and plus sign +. They are placed after an expression. The asterisk * indicates that the preceding expression can occur zero or more times. The plus sign + indicates that the preceding expression can occur one or more times (it must occur at least once).

The question mark operator ? is similar, except it indicates that the expression is optional — it can occur zero or one times.

("zero" ~ "or" ~ "more")*
 ("one" | "or" | "more")+
           (^"optional")?

Note that expr* and expr? will always succeed, because they are allowed to match zero times. For example, "a"* ~ "b"? will succeed even on an empty input string.

Other numbers of repetitions can be indicated using curly brackets:

expr{n}           // exactly n repetitions
expr{m, n}        // between m and n repetitions, inclusive

expr{, n}         // at most n repetitions
expr{m, }         // at least m repetitions

Thus expr* is equivalent to expr{0, }; expr+ is equivalent to expr{1, }; and expr? is equivalent to expr{0, 1}.

Predicates

Preceding an expression with an ampersand & or exclamation mark ! turns it into a predicate that never consumes any input. You might know these operators as "lookahead" or "non-progressing".

The positive predicate, written as an ampersand &, attempts to match its inner expression. If the inner expression succeeds, parsing continues, but at the same position as the predicate — &foo ~ bar is thus a kind of "AND" statement: "the input string must match foo AND bar". If the inner expression fails, the whole expression fails too.

The negative predicate, written as an exclamation mark !, attempts to match its inner expression. If the inner expression fails, the predicate succeeds and parsing continues at the same position as the predicate. If the inner expression succeeds, the predicate fails!foo ~ bar is thus a kind of "NOT" statement: "the input string must match bar but NOT foo".

This leads to the common idiom meaning "any character but":

not_space_or_tab = {
    !(                // if the following text is not
        " "           //     a space
        | "\t"        //     or a tab
    )
    ~ ANY             // then consume one character
}

triple_quoted_string = {
    "'''"
    ~ triple_quoted_character*
    ~ "'''"
}
triple_quoted_character = {
    !"'''"        // if the following text is not three apostrophes
    ~ ANY         // then consume one character
}

Operator precedence and grouping (WIP)

The repetition operators asterisk *, plus sign +, and question mark ? apply to the immediately preceding expression.

"One " ~ "or " ~ "more. "+
"One " ~ "or " ~ ("more. "+)
    are equivalent and match
"One or more. more. more. more. "

Larger expressions can be repeated by surrounding them with parentheses.

("One " ~ "or " ~ "more. ")+
    matches
"One or more. One or more. "

Repetition operators have the highest precedence, followed by predicate operators, the sequence operator, and finally ordered choice.

my_rule = {
    "a"* ~ "b"?
    | &"b"+ ~ "a"
}

// equivalent to

my_rule = {
      ( ("a"*) ~ ("b"?) )
    | ( (&("b"+)) ~ "a" )
}

Start and end of input

The rules SOI and EOI match the start and end of the input string, respectively. Neither consumes any text. They only indicate whether the parser is currently at one edge of the input.

For example, to ensure that a rule matches the entire input, where any syntax error results in a failed parse (rather than a successful but incomplete parse):

main = {
    SOI
    ~ (...)
    ~ EOI
}

Implicit whitespace

Many languages and text formats allow arbitrary whitespace and comments between logical tokens. For instance, Rust considers 4+5 equivalent to 4 + 5 and 4 /* comment */ + 5.

The optional rules WHITESPACE and COMMENT implement this behaviour. If either (or both) are defined, they will be implicitly inserted at every sequence and between every repetition (except in atomic rules).

expression = { "4" ~ "+" ~ "5" }
WHITESPACE = _{ " " }
COMMENT = _{ "/*" ~ (!"*/" ~ ANY)* ~ "*/" }
"4+5"
"4 + 5"
"4  +     5"
"4 /* comment */ + 5"

As you can see, WHITESPACE and COMMENT are run repeatedly, so they need only match a single whitespace character or a single comment. The grammar above is equivalent to:

expression = {
    "4"   ~ (ws | com)*
    ~ "+" ~ (ws | com)*
    ~ "5"
}
ws = _{ " " }
com = _{ "/*" ~ (!"*/" ~ ANY)* ~ "*/" }

Note that Implicit whitespace is not inserted at the beginning or end of rules — for instance, expression does not match " 4+5 ". If you want to include Implicit whitespace at the beginning and end of a rule, you will need to sandwich it between two empty rules (often SOI and EOI as above):

WHITESPACE = _{ " " }
expression = { "4" ~ "+" ~ "5" }
main = { SOI ~ expression ~ EOI }
"4+5"
"  4 + 5   "

(Be sure to mark the WHITESPACE and COMMENT rules as silent unless you want to see them included inside other rules!)

If you want the WHITESPACE and COMMENT inner rules included, note that you need to mark the rule with an explicit $:

COMMENT = ${ SingleLineComment }
SingleLineComment = { "//" ~ (!"\n" ~ ANY)* }

Silent and atomic rules

Silent

Silent rules are just like normal rules — when run, they function the same way — except they do not produce pairs or tokens. If a rule is silent, it will never appear in a parse result.

To make a silent rule, precede the left curly bracket { with a low line (underscore) _.

silent = _{ ... }

Rules called from a silent rule are not treated as silent unless they are declared to be silent. These rules may produce pairs or tokens and can appear in a parse result.

Atomic

Pest has two kinds of atomic rules: atomic and compound atomic. To make one, write the sigil before the left curly bracket {.

/// Atomic rule start with `@`
atomic = @{ ... }

/// Compound Atomic start with `$`
compound_atomic = ${ ... }

Both kinds of atomic rule prevent implicit whitespace:

  1. Inside an atomic rule, the tilde ~ means "immediately followed by".
  2. Repetition operators (asterisk * and plus sign +) have no implicit separation.

In addition, all other rules called from an atomic rule are also treated as atomic.

The difference between the two is how they produce tokens for inner rules:

  • atomic - In an Atomic rule, interior matching rules are silent.
  • compound atomic - By contrast, compound atomic rules produce inner tokens as normal.

Atomic rules are useful when the text you are parsing ignores whitespace except in a few cases, such as literal strings. In this instance, you can write WHITESPACE or COMMENT rules, then make your string-matching rule be atomic.

Non-atomic

Sometimes, you'll want to cancel the effects of atomic parsing. For instance, you might want to have string interpolation with an expression inside, where the inside expression can still have whitespace like normal.

#!/bin/env python3
print(f"The answer is {2 + 4}.")

This is where you use a non-atomic rule. Write an exclamation mark ! in front of the defining curly bracket. The rule will run as non-atomic, whether it is called from an atomic rule or not.

fstring = @{ "\"" ~ ... }
expr = !{ ... }

Tags

Sometimes, you may want to attach a label to a part of a rule. This is useful for distinguishing among different types of tokens (of the same expression) or for the ease of extracting information from parse trees (without creating additional rules). To do this, you can use the # symbol to bind a name to a part of a rule:

rule = { #tag = ... }

You can then access tags in your parse tree by using the as_node_tag method on Pair or you can use the helper methods find_first_tagged or find_tagged on Pairs:

let pairs = ExampleParser::parse(Rule::example_rule, example_input).unwrap();
for pair in pairs.clone() {
    if let Some(tag) = pair.as_node_tag() {
        // ...
    }
}
let first = pairs.find_first_tagged("tag");
let all_tagged = pairs.find_tagged("tag");

Note that the tagged part of the rule cannot be silent or the tag will be ignored, as would be the case in the following example.

rule = { #tag = expression }
expression = _{ ... }

This also includes built-in rules, which are always silent. So the following example will also not create a tag.

rule = { #tag = ASCII }

[!WARNING] You need to enable "grammar-extras" feature to use this functionality:

# ...
pest_derive = { version = "2.7", features = ["grammar-extras"] }

The stack (WIP)

pest maintains a stack that can be manipulated directly from the grammar. An expression can be matched and pushed onto the stack with the keyword PUSH, then later matched exactly with the keywords PEEK and POP.

Using the stack allows the exact same text to be matched multiple times, rather than the same pattern.

For example,

same_text = {
    PUSH( "a" | "b" | "c" )
    ~ POP
}
same_pattern = {
    ("a" | "b" | "c")
    ~ ("a" | "b" | "c")
}

In this case, same_pattern will match "ab", while same_text will not.

One practical use is in parsing Rust "raw string literals", which look like this:

const raw_str: &str = r###"
    Some number of number signs # followed by a quotation mark ".

    Quotation marks can be used anywhere inside: """"""""",
    as long as one is not followed by a matching number of number signs,
    which ends the string: "###;

When parsing a raw string, we have to keep track of how many number signs # occurred before the quotation mark. We can do this using the stack:

raw_string = {
    "r" ~ PUSH("#"*) ~ "\""    // push the number signs onto the stack
    ~ raw_string_interior
    ~ "\"" ~ POP               // match a quotation mark and the number signs
}
raw_string_interior = {
    (
        !("\"" ~ PEEK)    // unless the next character is a quotation mark
                          // followed by the correct amount of number signs,
        ~ ANY             // consume one character
    )*
}

Indentation-Sensitive Languages

In conjunction with some extra helpers, the stack can be used to allow parsing indentation-sensitive languages, such as Python.

The general idea is that you store the leading whitespace on the stack with PUSH and then use PEEK_ALL to match all of the whitespace on subsequent lines.

When exiting an indented block, use DROP to remove the stack entry without needing to match it.

An example grammar demonstrating this concept is given here:

Grammar = { SOI ~ NEWLINE* ~ BlockContent* ~ NEWLINE* ~ EOI }

NewBlock = _{
    // The first line in the block
    PEEK_ALL ~ PUSH("  "+ | "\t"+) ~ BlockContent ~
    // Subsequent lines in the block
    (PEEK_ALL ~ BlockContent)* ~
    // Remove the last layer of indentation from the stack when exiting the block
    DROP
}

BlockName = { ASCII_ALPHA+ }

BlockContent = {
    BlockName ~ (NEWLINE | EOI) ~ NewBlock*
}

This matches texts such as the following, whilst preserving indentation structure:

Hello
  This
    Is
    An
  Indentation
    Sensitive
      Language
Demonstration

Cheat sheet

SyntaxMeaningSyntaxMeaning
foo = { ... }regular rulebaz = @{ ... }atomic
bar = _{ ... }silentqux = ${ ... }compound-atomic
#tag = ...tagsplugh = !{ ... }non-atomic
"abc"exact string^"abc"case insensitive
'a'..'z'character rangeANYany character
foo ~ barsequencebaz | quxordered choice
foo*zero or morebar+one or more
baz?optionalqux{n}exactly n
qux{m, n}between m and n (inclusive)
&foopositive predicate!barnegative predicate
PUSH(baz)match and push
POPmatch and popPEEKmatch without pop
DROPpop without matchingPEEK_ALLmatch entire stack