Crate pest

source ·
Expand description

§pest. The Elegant Parser

pest is a general purpose parser written in Rust with a focus on accessibility, correctness, and performance. It uses parsing expression grammars (or PEG) as input, which are similar in spirit to regular expressions, but which offer the enhanced expressivity needed to parse complex languages.

§Getting started

The recommended way to start parsing with pest is to read the official book.

Other helpful resources:

§Usage

The core of pest is the trait Parser, which provides an interface to the parsing functionality.

The accompanying crate pest_derive can automatically generate a Parser from a PEG grammar. Using pest_derive is highly encouraged, but it is also possible to implement Parser manually if required.

§.pest files

Grammar definitions reside in custom .pest files located in the crate src directory. Parsers are automatically generated from these files using #[derive(Parser)] and a special #[grammar = "..."] attribute on a dummy struct.

#[derive(Parser)]
#[grammar = "path/to/my_grammar.pest"] // relative to src
struct MyParser;

The syntax of .pest files is documented in the pest_derive crate.

§Inline grammars

Grammars can also be inlined by using the #[grammar_inline = "..."] attribute.

§Grammar

A grammar is a series of rules separated by whitespace, possibly containing comments.

§Comments

Comments start with // and end at the end of the line.

// a comment

§Rules

Rules have the following form:

name = optional_modifier { expression }

The name of the rule is formed from alphanumeric characters or _ with the condition that the first character is not a digit and is used to create token pairs. When the rule starts being parsed, the starting part of the token is being produced, with the ending part being produced when the rule finishes parsing.

The following token pair notation a(b(), c()) denotes the tokens: start a, start b, end b, start c, end c, end a.

§Modifiers

Modifiers are optional and can be one of _, @, $, or !. These modifiers change the behavior of the rules.

  1. Silent (_)

    Silent rules do not create token pairs during parsing, nor are they error-reported.

    a = _{ "a" }
    b =  { a ~ "b" }

    Parsing "ab" produces the token pair b().

  2. Atomic (@)

    Atomic rules do not accept whitespace or comments within their expressions and have a cascading effect on any rule they call. I.e. rules that are not atomic but are called by atomic rules behave atomically.

    Any rules called by atomic rules do not generate token pairs.

    a =  { "a" }
    b = @{ a ~ "b" }
    
    WHITESPACE = _{ " " }

    Parsing "ab" produces the token pair b(), while "a b" produces an error.

  3. Compound-atomic ($)

    Compound-atomic are identical to atomic rules with the exception that rules called by them are not forbidden from generating token pairs.

    a =  { "a" }
    b = ${ a ~ "b" }
    
    WHITESPACE = _{ " " }

    Parsing "ab" produces the token pairs b(a()), while "a b" produces an error.

  4. Non-atomic (!)

    Non-atomic are identical to normal rules with the exception that they stop the cascading effect of atomic and compound-atomic rules.

    a =  { "a" }
    b = !{ a ~ "b" }
    c = @{ b }
    
    WHITESPACE = _{ " " }

    Parsing both "ab" and "a b" produce the token pairs c(a()).

§Expressions

Expressions can be either terminals or non-terminals.

  1. Terminals
TerminalUsage
"a"matches the exact string "a"
^"a"matches the exact string "a" case insensitively (ASCII only)
'a'..'z'matches one character between 'a' and 'z'
amatches rule a

Strings and characters follow Rust’s escape mechanisms, while identifiers can contain alphanumeric characters and underscores (_), as long as they do not start with a digit.

  1. Non-terminals
Non-terminalUsage
(e)matches e
e1 ~ e2matches the sequence e1 e2
e1 | e2matches either e1 or e2
e*matches e zero or more times
e+matches e one or more times
e{n}matches e exactly n times
e{, n}matches e at most n times
e{n,}matches e at least n times
e{m, n}matches e between m and n times inclusively
e?optionally matches e
&ematches e without making progress
!ematches if e doesn’t match without making progress
PUSH(e)matches e and pushes it’s captured string down the stack

where e, e1, and e2 are expressions.

Matching is greedy, without backtracking. Note the difference in behavior for these two rules in matching identifiers that don’t end in an underscore:

// input: ab_bb_b

identifier = @{ "a" ~ ("b"|"_")* ~ "b" }
// matches:      a     b_bb_b       nothing -> error!      

identifier = @{ "a" ~ ("_"* ~ "b")* }
// matches:      a     b, _bb, _b   in three repetitions

Expressions can modify the stack only if they match the input. For example, if e1 in the compound expression e1 | e2 does not match the input, then it does not modify the stack, so e2 sees the stack in the same state as e1 did. Repetitions and optionals (e*, e+, e{, n}, e{n,}, e{m,n}, e?) can modify the stack each time e matches. The !e and &e expressions are a special case; they never modify the stack. Many languages have “keyword” tokens (e.g. if, for, while) as well as general tokens (e.g. identifier) that matches any word. In order to match a keyword, generally, you may need to restrict that is not immediately followed by another letter or digit (otherwise it would be matched as an identifier).

§Special rules

Special rules can be called within the grammar. They are:

  • WHITESPACE - runs between rules and sub-rules
  • COMMENT - runs between rules and sub-rules
  • ANY - matches exactly one char
  • SOI - (start-of-input) matches only when a Parser is still at the starting position
  • EOI - (end-of-input) matches only when a Parser has reached its end
  • POP - pops a string from the stack and matches it
  • POP_ALL - pops the entire state of the stack and matches it
  • PEEK - peeks a string from the stack and matches it
  • PEEK[a..b] - peeks part of the stack and matches it
  • PEEK_ALL - peeks the entire state of the stack and matches it
  • DROP - drops the top of the stack (fails to match if the stack is empty)

WHITESPACE and COMMENT should be defined manually if needed. All other rules cannot be overridden.

§WHITESPACE and COMMENT

When defined, these rules get matched automatically in sequences (~) and repetitions (*, +) between expressions. Atomic rules and those rules called by atomic rules are exempt from this behavior.

These rules should be defined so as to match one whitespace character and one comment only since they are run in repetitions.

If both WHITESPACE and COMMENT are defined, this grammar:

a = { b ~ c }

is effectively transformed into this one behind the scenes:

a = { b ~ WHITESPACE* ~ (COMMENT ~ WHITESPACE*)* ~ c }

§PUSH, POP, DROP, and PEEK

PUSH(e) simply pushes the captured string of the expression e down a stack. This stack can then later be used to match grammar based on its content with POP and PEEK.

PEEK always matches the string at the top of stack. So, if the stack contains ["b", "a"] ("a" being on top), this grammar:

a = { PEEK }

is effectively transformed into at parse time:

a = { "a" }

POP works the same way with the exception that it pops the string off of the stack if the match worked. With the stack from above, if POP matches "a", the stack will be mutated to ["b"].

DROP makes it possible to remove the string at the top of the stack without matching it. If the stack is nonempty, DROP drops the top of the stack. If the stack is empty, then DROP fails to match.

§Advanced peeking

PEEK[start..end] and PEEK_ALL allow to peek deeper into the stack. The syntax works exactly like Rust’s exclusive slice syntax. Additionally, negative indices can be used to indicate an offset from the top. If the end lies before or at the start, the expression matches (as does a PEEK_ALL on an empty stack). With the stack ["c", "b", "a"] ("a" on top):

fill = PUSH("c") ~ PUSH("b") ~ PUSH("a")
v = { PEEK_ALL } = { "a" ~ "b" ~ "c" }  // top to bottom
w = { PEEK[..] } = { "c" ~ "b" ~ "a" }  // bottom to top
x = { PEEK[1..2] } = { PEEK[1..-1] } = { "b" }
y = { PEEK[..-2] } = { PEEK[0..1] } = { "a" }
z = { PEEK[1..] } = { PEEK[-2..3] } = { "c" ~ "b" }
n = { PEEK[2..-2] } = { PEEK[2..1] } = { "" }

For historical reasons, PEEK_ALL matches from top to bottom, while PEEK[start..end] matches from bottom to top. There is currently no syntax to match a slice of the stack top to bottom.

§Rule

All rules defined or used in the grammar populate a generated enum called Rule. This implements pest’s RuleType and can be used throughout the API.

§Built-in rules

Pest also comes with a number of built-in rules for convenience. They are:

  • ASCII_DIGIT - matches a numeric character from 0..9
  • ASCII_NONZERO_DIGIT - matches a numeric character from 1..9
  • ASCII_BIN_DIGIT - matches a numeric character from 0..1
  • ASCII_OCT_DIGIT - matches a numeric character from 0..7
  • ASCII_HEX_DIGIT - matches a numeric character from 0..9 or a..f or A..F
  • ASCII_ALPHA_LOWER - matches a character from a..z
  • ASCII_ALPHA_UPPER - matches a character from A..Z
  • ASCII_ALPHA - matches a character from a..z or A..Z
  • ASCII_ALPHANUMERIC - matches a character from a..z or A..Z or 0..9
  • ASCII - matches a character from \x00..\x7f
  • NEWLINE - matches either “\n” or “\r\n” or “\r”

Modules§

  • Types for different kinds of parsing failures.
  • Types and iterators for parser output.
  • Constructs useful in prefix, postfix, and infix operator parsing with the Pratt parsing method.
  • prec_climberDeprecated
    Constructs useful in infix operator parsing with the precedence climbing method.

Macros§

  • Testing tool that compares produced errors.
  • Testing tool that compares produced tokens.

Structs§

Enums§

Traits§

  • A trait with a single method that parses strings.
  • A trait which parser rules must implement.

Functions§

  • Merges two spans into one.
  • Sets the maximum call limit for the parser state to prevent stack overflows or excessive execution times in some grammars. If set, the calls are tracked as a running total over all non-terminal rules that can nest closures (which are passed to transform the parser state).
  • Sets whether information for more error details should be collected. This is useful for debugging parser errors (as it leads to more comprehensive error messages), but it has a higher performance cost. (hence, it’s off by default)
  • Creates a ParserState from a &str, supplying it to a closure f.

Type Aliases§

  • Type alias to simplify specifying the return value of chained closures.