rust/ripgrep
1
0
Fork 0
mirror of https://github.com/BurntSushi/ripgrep.git synced 2025-06-14 22:15:13 +02:00
Code Issues Releases Activity

repo: move all source code in crates directory

Browse Source 
The top-level listing was just getting a bit too long for my taste. So
put all of the code in one directory and shrink the large top-level mess
to a small top-level mess.

NOTE: This commit only contains renames. The subsequent commit will
actually make ripgrep build again. We do it this way with the naive hope
that this will make it easier for git history to track the renames.
Sigh.
This commit is contained in:
Andrew Gallant
2020-02-17 18:19:19 -05:00
parent 0bc4f0447b
commit fdd8510fdd
113 changed files with 0 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

263
crates/regex/src/ast.rs Normal file
View File

@ -0,0 +1,263 @@
use regex_syntax::ast::parse::Parser;
use regex_syntax::ast::{self, Ast};
/// The results of analyzing AST of a regular expression (e.g., for supporting
/// smart case).
#[derive(Clone, Debug)]
pub struct AstAnalysis {
/// True if and only if a literal uppercase character occurs in the regex.
any_uppercase: bool,
/// True if and only if the regex contains any literal at all.
any_literal: bool,
/// True if and only if the regex consists entirely of a literal and no
/// other special regex characters.
all_verbatim_literal: bool,
}
impl AstAnalysis {
/// Returns a `AstAnalysis` value by doing analysis on the AST of `pattern`.
///
/// If `pattern` is not a valid regular expression, then `None` is
/// returned.
#[allow(dead_code)]
pub fn from_pattern(pattern: &str) -> Option<AstAnalysis> {
Parser::new()
.parse(pattern)
.map(|ast| AstAnalysis::from_ast(&ast))
.ok()
}
/// Perform an AST analysis given the AST.
pub fn from_ast(ast: &Ast) -> AstAnalysis {
let mut analysis = AstAnalysis::new();
analysis.from_ast_impl(ast);
analysis
}
/// Returns true if and only if a literal uppercase character occurs in
/// the pattern.
///
/// For example, a pattern like `\pL` contains no uppercase literals,
/// even though `L` is uppercase and the `\pL` class contains uppercase
/// characters.
pub fn any_uppercase(&self) -> bool {
self.any_uppercase
}
/// Returns true if and only if the regex contains any literal at all.
///
/// For example, a pattern like `\pL` reports `false`, but a pattern like
/// `\pLfoo` reports `true`.
pub fn any_literal(&self) -> bool {
self.any_literal
}
/// Returns true if and only if the entire pattern is a verbatim literal
/// with no special meta characters.
///
/// When this is true, then the pattern satisfies the following law:
/// `escape(pattern) == pattern`. Notable examples where this returns
/// `false` include patterns like `a\u0061` even though `\u0061` is just
/// a literal `a`.
///
/// The purpose of this flag is to determine whether the patterns can be
/// given to non-regex substring search algorithms as-is.
#[allow(dead_code)]
pub fn all_verbatim_literal(&self) -> bool {
self.all_verbatim_literal
}
/// Creates a new `AstAnalysis` value with an initial configuration.
fn new() -> AstAnalysis {
AstAnalysis {
any_uppercase: false,
any_literal: false,
all_verbatim_literal: true,
}
}
fn from_ast_impl(&mut self, ast: &Ast) {
if self.done() {
return;
}
match *ast {
Ast::Empty(_) => {}
Ast::Flags(_)
| Ast::Dot(_)
| Ast::Assertion(_)
| Ast::Class(ast::Class::Unicode(_))
| Ast::Class(ast::Class::Perl(_)) => {
self.all_verbatim_literal = false;
}
Ast::Literal(ref x) => {
self.from_ast_literal(x);
}
Ast::Class(ast::Class::Bracketed(ref x)) => {
self.all_verbatim_literal = false;
self.from_ast_class_set(&x.kind);
}
Ast::Repetition(ref x) => {
self.all_verbatim_literal = false;
self.from_ast_impl(&x.ast);
}
Ast::Group(ref x) => {
self.all_verbatim_literal = false;
self.from_ast_impl(&x.ast);
}
Ast::Alternation(ref alt) => {
self.all_verbatim_literal = false;
for x in &alt.asts {
self.from_ast_impl(x);
}
}
Ast::Concat(ref alt) => {
for x in &alt.asts {
self.from_ast_impl(x);
}
}
}
}
fn from_ast_class_set(&mut self, ast: &ast::ClassSet) {
if self.done() {
return;
}
match *ast {
ast::ClassSet::Item(ref item) => {
self.from_ast_class_set_item(item);
}
ast::ClassSet::BinaryOp(ref x) => {
self.from_ast_class_set(&x.lhs);
self.from_ast_class_set(&x.rhs);
}
}
}
fn from_ast_class_set_item(&mut self, ast: &ast::ClassSetItem) {
if self.done() {
return;
}
match *ast {
ast::ClassSetItem::Empty(_)
| ast::ClassSetItem::Ascii(_)
| ast::ClassSetItem::Unicode(_)
| ast::ClassSetItem::Perl(_) => {}
ast::ClassSetItem::Literal(ref x) => {
self.from_ast_literal(x);
}
ast::ClassSetItem::Range(ref x) => {
self.from_ast_literal(&x.start);
self.from_ast_literal(&x.end);
}
ast::ClassSetItem::Bracketed(ref x) => {
self.from_ast_class_set(&x.kind);
}
ast::ClassSetItem::Union(ref union) => {
for x in &union.items {
self.from_ast_class_set_item(x);
}
}
}
}
fn from_ast_literal(&mut self, ast: &ast::Literal) {
if ast.kind != ast::LiteralKind::Verbatim {
self.all_verbatim_literal = false;
}
self.any_literal = true;
self.any_uppercase = self.any_uppercase || ast.c.is_uppercase();
}
/// Returns true if and only if the attributes can never change no matter
/// what other AST it might see.
fn done(&self) -> bool {
self.any_uppercase && self.any_literal && !self.all_verbatim_literal
}
}
#[cfg(test)]
mod tests {
use super::*;
fn analysis(pattern: &str) -> AstAnalysis {
AstAnalysis::from_pattern(pattern).unwrap()
}
#[test]
fn various() {
let x = analysis("");
assert!(!x.any_uppercase);
assert!(!x.any_literal);
assert!(x.all_verbatim_literal);
let x = analysis("foo");
assert!(!x.any_uppercase);
assert!(x.any_literal);
assert!(x.all_verbatim_literal);
let x = analysis("Foo");
assert!(x.any_uppercase);
assert!(x.any_literal);
assert!(x.all_verbatim_literal);
let x = analysis("foO");
assert!(x.any_uppercase);
assert!(x.any_literal);
assert!(x.all_verbatim_literal);
let x = analysis(r"foo\\");
assert!(!x.any_uppercase);
assert!(x.any_literal);
assert!(!x.all_verbatim_literal);
let x = analysis(r"foo\w");
assert!(!x.any_uppercase);
assert!(x.any_literal);
assert!(!x.all_verbatim_literal);
let x = analysis(r"foo\S");
assert!(!x.any_uppercase);
assert!(x.any_literal);
assert!(!x.all_verbatim_literal);
let x = analysis(r"foo\p{Ll}");
assert!(!x.any_uppercase);
assert!(x.any_literal);
assert!(!x.all_verbatim_literal);
let x = analysis(r"foo[a-z]");
assert!(!x.any_uppercase);
assert!(x.any_literal);
assert!(!x.all_verbatim_literal);
let x = analysis(r"foo[A-Z]");
assert!(x.any_uppercase);
assert!(x.any_literal);
assert!(!x.all_verbatim_literal);
let x = analysis(r"foo[\S\t]");
assert!(!x.any_uppercase);
assert!(x.any_literal);
assert!(!x.all_verbatim_literal);
let x = analysis(r"foo\\S");
assert!(x.any_uppercase);
assert!(x.any_literal);
assert!(!x.all_verbatim_literal);
let x = analysis(r"\p{Ll}");
assert!(!x.any_uppercase);
assert!(!x.any_literal);
assert!(!x.all_verbatim_literal);
let x = analysis(r"aBc\w");
assert!(x.any_uppercase);
assert!(x.any_literal);
assert!(!x.all_verbatim_literal);
let x = analysis(r"a\u0061");
assert!(!x.any_uppercase);
assert!(x.any_literal);
assert!(!x.all_verbatim_literal);
}
}

288
crates/regex/src/config.rs Normal file
View File

@ -0,0 +1,288 @@
use grep_matcher::{ByteSet, LineTerminator};
use regex::bytes::{Regex, RegexBuilder};
use regex_syntax::ast::{self, Ast};
use regex_syntax::hir::{self, Hir};
use ast::AstAnalysis;
use crlf::crlfify;
use error::Error;
use literal::LiteralSets;
use multi::alternation_literals;
use non_matching::non_matching_bytes;
use strip::strip_from_match;
/// Config represents the configuration of a regex matcher in this crate.
/// The configuration is itself a rough combination of the knobs found in
/// the `regex` crate itself, along with additional `grep-matcher` specific
/// options.
///
/// The configuration can be used to build a "configured" HIR expression. A
/// configured HIR expression is an HIR expression that is aware of the
/// configuration which generated it, and provides transformation on that HIR
/// such that the configuration is preserved.
#[derive(Clone, Debug)]
pub struct Config {
pub case_insensitive: bool,
pub case_smart: bool,
pub multi_line: bool,
pub dot_matches_new_line: bool,
pub swap_greed: bool,
pub ignore_whitespace: bool,
pub unicode: bool,
pub octal: bool,
pub size_limit: usize,
pub dfa_size_limit: usize,
pub nest_limit: u32,
pub line_terminator: Option<LineTerminator>,
pub crlf: bool,
pub word: bool,
}
impl Default for Config {
fn default() -> Config {
Config {
case_insensitive: false,
case_smart: false,
multi_line: false,
dot_matches_new_line: false,
swap_greed: false,
ignore_whitespace: false,
unicode: true,
octal: false,
// These size limits are much bigger than what's in the regex
// crate.
size_limit: 100 * (1 << 20),
dfa_size_limit: 1000 * (1 << 20),
nest_limit: 250,
line_terminator: None,
crlf: false,
word: false,
}
}
}
impl Config {
/// Parse the given pattern and returned its HIR expression along with
/// the current configuration.
///
/// If there was a problem parsing the given expression then an error
/// is returned.
pub fn hir(&self, pattern: &str) -> Result<ConfiguredHIR, Error> {
let ast = self.ast(pattern)?;
let analysis = self.analysis(&ast)?;
let expr = hir::translate::TranslatorBuilder::new()
.allow_invalid_utf8(true)
.case_insensitive(self.is_case_insensitive(&analysis))
.multi_line(self.multi_line)
.dot_matches_new_line(self.dot_matches_new_line)
.swap_greed(self.swap_greed)
.unicode(self.unicode)
.build()
.translate(pattern, &ast)
.map_err(Error::regex)?;
let expr = match self.line_terminator {
None => expr,
Some(line_term) => strip_from_match(expr, line_term)?,
};
Ok(ConfiguredHIR {
original: pattern.to_string(),
config: self.clone(),
analysis: analysis,
// If CRLF mode is enabled, replace `$` with `(?:\r?$)`.
expr: if self.crlf { crlfify(expr) } else { expr },
})
}
/// Accounting for the `smart_case` config knob, return true if and only if
/// this pattern should be matched case insensitively.
fn is_case_insensitive(&self, analysis: &AstAnalysis) -> bool {
if self.case_insensitive {
return true;
}
if !self.case_smart {
return false;
}
analysis.any_literal() && !analysis.any_uppercase()
}
/// Returns true if and only if this config is simple enough such that
/// if the pattern is a simple alternation of literals, then it can be
/// constructed via a plain Aho-Corasick automaton.
///
/// Note that it is OK to return true even when settings like `multi_line`
/// are enabled, since if multi-line can impact the match semantics of a
/// regex, then it is by definition not a simple alternation of literals.
pub fn can_plain_aho_corasick(&self) -> bool {
!self.word && !self.case_insensitive && !self.case_smart
}
/// Perform analysis on the AST of this pattern.
///
/// This returns an error if the given pattern failed to parse.
fn analysis(&self, ast: &Ast) -> Result<AstAnalysis, Error> {
Ok(AstAnalysis::from_ast(ast))
}
/// Parse the given pattern into its abstract syntax.
///
/// This returns an error if the given pattern failed to parse.
fn ast(&self, pattern: &str) -> Result<Ast, Error> {
ast::parse::ParserBuilder::new()
.nest_limit(self.nest_limit)
.octal(self.octal)
.ignore_whitespace(self.ignore_whitespace)
.build()
.parse(pattern)
.map_err(Error::regex)
}
}
/// A "configured" HIR expression, which is aware of the configuration which
/// produced this HIR.
///
/// Since the configuration is tracked, values with this type can be
/// transformed into other HIR expressions (or regular expressions) in a way
/// that preserves the configuration. For example, the `fast_line_regex`
/// method will apply literal extraction to the inner HIR and use that to build
/// a new regex that matches the extracted literals in a way that is
/// consistent with the configuration that produced this HIR. For example, the
/// size limits set on the configured HIR will be propagated out to any
/// subsequently constructed HIR or regular expression.
#[derive(Clone, Debug)]
pub struct ConfiguredHIR {
original: String,
config: Config,
analysis: AstAnalysis,
expr: Hir,
}
impl ConfiguredHIR {
/// Return the configuration for this HIR expression.
pub fn config(&self) -> &Config {
&self.config
}
/// Compute the set of non-matching bytes for this HIR expression.
pub fn non_matching_bytes(&self) -> ByteSet {
non_matching_bytes(&self.expr)
}
/// Returns true if and only if this regex needs to have its match offsets
/// tweaked because of CRLF support. Specifically, this occurs when the
/// CRLF hack is enabled and the regex is line anchored at the end. In
/// this case, matches that end with a `\r` have the `\r` stripped.
pub fn needs_crlf_stripped(&self) -> bool {
self.config.crlf && self.expr.is_line_anchored_end()
}
/// Builds a regular expression from this HIR expression.
pub fn regex(&self) -> Result<Regex, Error> {
self.pattern_to_regex(&self.expr.to_string())
}
/// If this HIR corresponds to an alternation of literals with no
/// capturing groups, then this returns those literals.
pub fn alternation_literals(&self) -> Option<Vec<Vec<u8>>> {
if !self.config.can_plain_aho_corasick() {
return None;
}
alternation_literals(&self.expr)
}
/// Applies the given function to the concrete syntax of this HIR and then
/// generates a new HIR based on the result of the function in a way that
/// preserves the configuration.
///
/// For example, this can be used to wrap a user provided regular
/// expression with additional semantics. e.g., See the `WordMatcher`.
pub fn with_pattern<F: FnMut(&str) -> String>(
&self,
mut f: F,
) -> Result<ConfiguredHIR, Error> {
self.pattern_to_hir(&f(&self.expr.to_string()))
}
/// If the current configuration has a line terminator set and if useful
/// literals could be extracted, then a regular expression matching those
/// literals is returned. If no line terminator is set, then `None` is
/// returned.
///
/// If compiling the resulting regular expression failed, then an error
/// is returned.
///
/// This method only returns something when a line terminator is set
/// because matches from this regex are generally candidates that must be
/// confirmed before reporting a match. When performing a line oriented
/// search, confirmation is easy: just extend the candidate match to its
/// respective line boundaries and then re-search that line for a full
/// match. This only works when the line terminator is set because the line
/// terminator setting guarantees that the regex itself can never match
/// through the line terminator byte.
pub fn fast_line_regex(&self) -> Result<Option<Regex>, Error> {
if self.config.line_terminator.is_none() {
return Ok(None);
}
match LiteralSets::new(&self.expr).one_regex(self.config.word) {
None => Ok(None),
Some(pattern) => self.pattern_to_regex(&pattern).map(Some),
}
}
/// Create a regex from the given pattern using this HIR's configuration.
fn pattern_to_regex(&self, pattern: &str) -> Result<Regex, Error> {
// The settings we explicitly set here are intentionally a subset
// of the settings we have. The key point here is that our HIR
// expression is computed with the settings in mind, such that setting
// them here could actually lead to unintended behavior. For example,
// consider the pattern `(?U)a+`. This will get folded into the HIR
// as a non-greedy repetition operator which will in turn get printed
// to the concrete syntax as `a+?`, which is correct. But if we
// set the `swap_greed` option again, then we'll wind up with `(?U)a+?`
// which is equal to `a+` which is not the same as what we were given.
//
// We also don't need to apply `case_insensitive` since this gets
// folded into the HIR and would just cause us to do redundant work.
//
// Finally, we don't need to set `ignore_whitespace` since the concrete
// syntax emitted by the HIR printer never needs it.
//
// We set the rest of the options. Some of them are important, such as
// the size limit, and some of them are necessary to preserve the
// intention of the original pattern. For example, the Unicode flag
// will impact how the WordMatcher functions, namely, whether its
// word boundaries are Unicode aware or not.
RegexBuilder::new(&pattern)
.nest_limit(self.config.nest_limit)
.octal(self.config.octal)
.multi_line(self.config.multi_line)
.dot_matches_new_line(self.config.dot_matches_new_line)
.unicode(self.config.unicode)
.size_limit(self.config.size_limit)
.dfa_size_limit(self.config.dfa_size_limit)
.build()
.map_err(Error::regex)
}
/// Create an HIR expression from the given pattern using this HIR's
/// configuration.
fn pattern_to_hir(&self, pattern: &str) -> Result<ConfiguredHIR, Error> {
// See `pattern_to_regex` comment for explanation of why we only set
// a subset of knobs here. e.g., `swap_greed` is explicitly left out.
let expr = ::regex_syntax::ParserBuilder::new()
.nest_limit(self.config.nest_limit)
.octal(self.config.octal)
.allow_invalid_utf8(true)
.multi_line(self.config.multi_line)
.dot_matches_new_line(self.config.dot_matches_new_line)
.unicode(self.config.unicode)
.build()
.parse(pattern)
.map_err(Error::regex)?;
Ok(ConfiguredHIR {
original: self.original.clone(),
config: self.config.clone(),
analysis: self.analysis.clone(),
expr: expr,
})
}
}

189
crates/regex/src/crlf.rs Normal file
View File

@ -0,0 +1,189 @@
use std::collections::HashMap;
use grep_matcher::{Match, Matcher, NoError};
use regex::bytes::Regex;
use regex_syntax::hir::{self, Hir, HirKind};
use config::ConfiguredHIR;
use error::Error;
use matcher::RegexCaptures;
/// A matcher for implementing "word match" semantics.
#[derive(Clone, Debug)]
pub struct CRLFMatcher {
/// The regex.
regex: Regex,
/// A map from capture group name to capture group index.
names: HashMap<String, usize>,
}
impl CRLFMatcher {
/// Create a new matcher from the given pattern that strips `\r` from the
/// end of every match.
///
/// This panics if the given expression doesn't need its CRLF stripped.
pub fn new(expr: &ConfiguredHIR) -> Result<CRLFMatcher, Error> {
assert!(expr.needs_crlf_stripped());
let regex = expr.regex()?;
let mut names = HashMap::new();
for (i, optional_name) in regex.capture_names().enumerate() {
if let Some(name) = optional_name {
names.insert(name.to_string(), i.checked_sub(1).unwrap());
}
}
Ok(CRLFMatcher { regex, names })
}
/// Return the underlying regex used by this matcher.
pub fn regex(&self) -> &Regex {
&self.regex
}
}
impl Matcher for CRLFMatcher {
type Captures = RegexCaptures;
type Error = NoError;
fn find_at(
&self,
haystack: &[u8],
at: usize,
) -> Result<Option<Match>, NoError> {
let m = match self.regex.find_at(haystack, at) {
None => return Ok(None),
Some(m) => Match::new(m.start(), m.end()),
};
Ok(Some(adjust_match(haystack, m)))
}
fn new_captures(&self) -> Result<RegexCaptures, NoError> {
Ok(RegexCaptures::new(self.regex.capture_locations()))
}
fn capture_count(&self) -> usize {
self.regex.captures_len().checked_sub(1).unwrap()
}
fn capture_index(&self, name: &str) -> Option<usize> {
self.names.get(name).map(|i| *i)
}
fn captures_at(
&self,
haystack: &[u8],
at: usize,
caps: &mut RegexCaptures,
) -> Result<bool, NoError> {
caps.strip_crlf(false);
let r =
self.regex.captures_read_at(caps.locations_mut(), haystack, at);
if !r.is_some() {
return Ok(false);
}
// If the end of our match includes a `\r`, then strip it from all
// capture groups ending at the same location.
let end = caps.locations().get(0).unwrap().1;
if end > 0 && haystack.get(end - 1) == Some(&b'\r') {
caps.strip_crlf(true);
}
Ok(true)
}
// We specifically do not implement other methods like find_iter or
// captures_iter. Namely, the iter methods are guaranteed to be correct
// by virtue of implementing find_at and captures_at above.
}
/// If the given match ends with a `\r`, then return a new match that ends
/// immediately before the `\r`.
pub fn adjust_match(haystack: &[u8], m: Match) -> Match {
if m.end() > 0 && haystack.get(m.end() - 1) == Some(&b'\r') {
m.with_end(m.end() - 1)
} else {
m
}
}
/// Substitutes all occurrences of multi-line enabled `$` with `(?:\r?$)`.
///
/// This does not preserve the exact semantics of the given expression,
/// however, it does have the useful property that anything that matched the
/// given expression will also match the returned expression. The difference is
/// that the returned expression can match possibly other things as well.
///
/// The principle reason why we do this is because the underlying regex engine
/// doesn't support CRLF aware `$` look-around. It's planned to fix it at that
/// level, but we perform this kludge in the mean time.
///
/// Note that while the match preserving semantics are nice and neat, the
/// match position semantics are quite a bit messier. Namely, `$` only ever
/// matches the position between characters where as `\r??` can match a
/// character and change the offset. This is regretable, but works out pretty
/// nicely in most cases, especially when a match is limited to a single line.
pub fn crlfify(expr: Hir) -> Hir {
match expr.into_kind() {
HirKind::Anchor(hir::Anchor::EndLine) => {
let concat = Hir::concat(vec![
Hir::repetition(hir::Repetition {
kind: hir::RepetitionKind::ZeroOrOne,
greedy: false,
hir: Box::new(Hir::literal(hir::Literal::Unicode('\r'))),
}),
Hir::anchor(hir::Anchor::EndLine),
]);
Hir::group(hir::Group {
kind: hir::GroupKind::NonCapturing,
hir: Box::new(concat),
})
}
HirKind::Empty => Hir::empty(),
HirKind::Literal(x) => Hir::literal(x),
HirKind::Class(x) => Hir::class(x),
HirKind::Anchor(x) => Hir::anchor(x),
HirKind::WordBoundary(x) => Hir::word_boundary(x),
HirKind::Repetition(mut x) => {
x.hir = Box::new(crlfify(*x.hir));
Hir::repetition(x)
}
HirKind::Group(mut x) => {
x.hir = Box::new(crlfify(*x.hir));
Hir::group(x)
}
HirKind::Concat(xs) => {
Hir::concat(xs.into_iter().map(crlfify).collect())
}
HirKind::Alternation(xs) => {
Hir::alternation(xs.into_iter().map(crlfify).collect())
}
}
}
#[cfg(test)]
mod tests {
use super::crlfify;
use regex_syntax::Parser;
fn roundtrip(pattern: &str) -> String {
let expr1 = Parser::new().parse(pattern).unwrap();
let expr2 = crlfify(expr1);
expr2.to_string()
}
#[test]
fn various() {
assert_eq!(roundtrip(r"(?m)$"), "(?:\r??(?m:$))");
assert_eq!(roundtrip(r"(?m)$$"), "(?:\r??(?m:$))(?:\r??(?m:$))");
assert_eq!(
roundtrip(r"(?m)(?:foo$|bar$)"),
"(?:foo(?:\r??(?m:$))|bar(?:\r??(?m:$)))"
);
assert_eq!(roundtrip(r"(?m)$a"), "(?:\r??(?m:$))a");
// Not a multiline `$`, so no crlfifying occurs.
assert_eq!(roundtrip(r"$"), "\\z");
// It's a literal, derp.
assert_eq!(roundtrip(r"\$"), "\\$");
}
}

88
crates/regex/src/error.rs Normal file
View File

@ -0,0 +1,88 @@
use std::error;
use std::fmt;
use util;
/// An error that can occur in this crate.
///
/// Generally, this error corresponds to problems building a regular
/// expression, whether it's in parsing, compilation or a problem with
/// guaranteeing a configured optimization.
#[derive(Clone, Debug)]
pub struct Error {
kind: ErrorKind,
}
impl Error {
pub(crate) fn new(kind: ErrorKind) -> Error {
Error { kind }
}
pub(crate) fn regex<E: error::Error>(err: E) -> Error {
Error { kind: ErrorKind::Regex(err.to_string()) }
}
/// Return the kind of this error.
pub fn kind(&self) -> &ErrorKind {
&self.kind
}
}
/// The kind of an error that can occur.
#[derive(Clone, Debug)]
pub enum ErrorKind {
/// An error that occurred as a result of parsing a regular expression.
/// This can be a syntax error or an error that results from attempting to
/// compile a regular expression that is too big.
///
/// The string here is the underlying error converted to a string.
Regex(String),
/// An error that occurs when a building a regex that isn't permitted to
/// match a line terminator. In general, building the regex will do its
/// best to make matching a line terminator impossible (e.g., by removing
/// `\n` from the `\s` character class), but if the regex contains a
/// `\n` literal, then there is no reasonable choice that can be made and
/// therefore an error is reported.
///
/// The string is the literal sequence found in the regex that is not
/// allowed.
NotAllowed(String),
/// This error occurs when a non-ASCII line terminator was provided.
///
/// The invalid byte is included in this error.
InvalidLineTerminator(u8),
/// Hints that destructuring should not be exhaustive.
///
/// This enum may grow additional variants, so this makes sure clients
/// don't count on exhaustive matching. (Otherwise, adding a new variant
/// could break existing code.)
#[doc(hidden)]
__Nonexhaustive,
}
impl error::Error for Error {
fn description(&self) -> &str {
match self.kind {
ErrorKind::Regex(_) => "regex error",
ErrorKind::NotAllowed(_) => "literal not allowed",
ErrorKind::InvalidLineTerminator(_) => "invalid line terminator",
ErrorKind::__Nonexhaustive => unreachable!(),
}
}
}
impl fmt::Display for Error {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
match self.kind {
ErrorKind::Regex(ref s) => write!(f, "{}", s),
ErrorKind::NotAllowed(ref lit) => {
write!(f, "the literal '{:?}' is not allowed in a regex", lit)
}
ErrorKind::InvalidLineTerminator(byte) => {
let x = util::show_bytes(&[byte]);
write!(f, "line terminators must be ASCII, but '{}' is not", x)
}
ErrorKind::__Nonexhaustive => unreachable!(),
}
}
}

28
crates/regex/src/lib.rs Normal file
View File

@ -0,0 +1,28 @@
/*!
An implementation of `grep-matcher`'s `Matcher` trait for Rust's regex engine.
*/
#![deny(missing_docs)]
extern crate aho_corasick;
extern crate grep_matcher;
#[macro_use]
extern crate log;
extern crate regex;
extern crate regex_syntax;
extern crate thread_local;
pub use error::{Error, ErrorKind};
pub use matcher::{RegexCaptures, RegexMatcher, RegexMatcherBuilder};
mod ast;
mod config;
mod crlf;
mod error;
mod literal;
mod matcher;
mod multi;
mod non_matching;
mod strip;
mod util;
mod word;

391
crates/regex/src/literal.rs Normal file
View File

@ -0,0 +1,391 @@
/*
This module is responsible for extracting *inner* literals out of the AST of a
regular expression. Normally this is the job of the regex engine itself, but
the regex engine doesn't look for inner literals. Since we're doing line based
searching, we can use them, so we need to do it ourselves.
*/
use regex_syntax::hir::literal::{Literal, Literals};
use regex_syntax::hir::{self, Hir, HirKind};
use util;
/// Represents prefix, suffix and inner "required" literals for a regular
/// expression.
///
/// Prefixes and suffixes are detected using regex-syntax. The inner required
/// literals are detected using something custom (but based on the code in
/// regex-syntax).
#[derive(Clone, Debug)]
pub struct LiteralSets {
/// A set of prefix literals.
prefixes: Literals,
/// A set of suffix literals.
suffixes: Literals,
/// A set of literals such that at least one of them must appear in every
/// match. A literal in this set may be neither a prefix nor a suffix.
required: Literals,
}
impl LiteralSets {
/// Create a set of literals from the given HIR expression.
pub fn new(expr: &Hir) -> LiteralSets {
let mut required = Literals::empty();
union_required(expr, &mut required);
LiteralSets {
prefixes: Literals::prefixes(expr),
suffixes: Literals::suffixes(expr),
required: required,
}
}
/// If it is deemed advantageuous to do so (via various suspicious
/// heuristics), this will return a single regular expression pattern that
/// matches a subset of the language matched by the regular expression that
/// generated these literal sets. The idea here is that the pattern
/// returned by this method is much cheaper to search for. i.e., It is
/// usually a single literal or an alternation of literals.
pub fn one_regex(&self, word: bool) -> Option<String> {
// TODO: The logic in this function is basically inscrutable. It grew
// organically in the old grep 0.1 crate. Ideally, it would be
// re-worked. In fact, the entire inner literal extraction should be
// re-worked. Actually, most of regex-syntax's literal extraction
// should also be re-worked. Alas... only so much time in the day.
if !word {
if self.prefixes.all_complete() && !self.prefixes.is_empty() {
debug!("literal prefixes detected: {:?}", self.prefixes);
// When this is true, the regex engine will do a literal scan,
// so we don't need to return anything. But we only do this
// if we aren't doing a word regex, since a word regex adds
// a `(?:\W|^)` to the beginning of the regex, thereby
// defeating the regex engine's literal detection.
return None;
}
}
// Out of inner required literals, prefixes and suffixes, which one
// is the longest? We pick the longest to do fast literal scan under
// the assumption that a longer literal will have a lower false
// positive rate.
let pre_lcp = self.prefixes.longest_common_prefix();
let pre_lcs = self.prefixes.longest_common_suffix();
let suf_lcp = self.suffixes.longest_common_prefix();
let suf_lcs = self.suffixes.longest_common_suffix();
let req_lits = self.required.literals();
let req = match req_lits.iter().max_by_key(|lit| lit.len()) {
None => &[],
Some(req) => &***req,
};
let mut lit = pre_lcp;
if pre_lcs.len() > lit.len() {
lit = pre_lcs;
}
if suf_lcp.len() > lit.len() {
lit = suf_lcp;
}
if suf_lcs.len() > lit.len() {
lit = suf_lcs;
}
if req_lits.len() == 1 && req.len() > lit.len() {
lit = req;
}
// Special case: if we detected an alternation of inner required
// literals and its longest literal is bigger than the longest
// prefix/suffix, then choose the alternation. In practice, this
// helps with case insensitive matching, which can generate lots of
// inner required literals.
let any_empty = req_lits.iter().any(|lit| lit.is_empty());
if req.len() > lit.len() && req_lits.len() > 1 && !any_empty {
debug!("required literals found: {:?}", req_lits);
let alts: Vec<String> = req_lits
.into_iter()
.map(|x| util::bytes_to_regex(x))
.collect();
// We're matching raw bytes, so disable Unicode mode.
Some(format!("(?-u:{})", alts.join("|")))
} else if lit.is_empty() {
// If we're here, then we have no LCP. No LCS. And no detected
// inner required literals. In theory this shouldn't happen, but
// the inner literal detector isn't as nice as we hope and doens't
// actually support returning a set of alternating required
// literals. (Instead, it only returns a set where EVERY literal
// in it is required. It cannot currently express "either P or Q
// is required.")
//
// In this case, it is possible that we still have meaningful
// prefixes or suffixes to use. So we look for the set of literals
// with the highest minimum length and use that to build our "fast"
// regex.
//
// This manifest in fairly common scenarios. e.g.,
//
// rg -w 'foo|bar|baz|quux'
//
// Normally, without the `-w`, the regex engine itself would
// detect the prefix correctly. Unfortunately, the `-w` option
// turns the regex into something like this:
//
// rg '(^|\W)(foo|bar|baz|quux)($|\W)'
//
// Which will defeat all prefix and suffix literal optimizations.
// (Not in theory---it could be better. But the current
// implementation isn't good enough.) ... So we make up for it
// here.
let p_min_len = self.prefixes.min_len();
let s_min_len = self.suffixes.min_len();
let lits = match (p_min_len, s_min_len) {
(None, None) => return None,
(Some(_), None) => {
debug!("prefix literals found");
self.prefixes.literals()
}
(None, Some(_)) => {
debug!("suffix literals found");
self.suffixes.literals()
}
(Some(p), Some(s)) => {
if p >= s {
debug!("prefix literals found");
self.prefixes.literals()
} else {
debug!("suffix literals found");
self.suffixes.literals()
}
}
};
debug!("prefix/suffix literals found: {:?}", lits);
let alts: Vec<String> =
lits.into_iter().map(|x| util::bytes_to_regex(x)).collect();
// We're matching raw bytes, so disable Unicode mode.
Some(format!("(?-u:{})", alts.join("|")))
} else {
debug!("required literal found: {:?}", util::show_bytes(lit));
Some(format!("(?-u:{})", util::bytes_to_regex(&lit)))
}
}
}
fn union_required(expr: &Hir, lits: &mut Literals) {
match *expr.kind() {
HirKind::Literal(hir::Literal::Unicode(c)) => {
let mut buf = [0u8; 4];
lits.cross_add(c.encode_utf8(&mut buf).as_bytes());
}
HirKind::Literal(hir::Literal::Byte(b)) => {
lits.cross_add(&[b]);
}
HirKind::Class(hir::Class::Unicode(ref cls)) => {
if count_unicode_class(cls) >= 5 || !lits.add_char_class(cls) {
lits.cut();
}
}
HirKind::Class(hir::Class::Bytes(ref cls)) => {
if count_byte_class(cls) >= 5 || !lits.add_byte_class(cls) {
lits.cut();
}
}
HirKind::Group(hir::Group { ref hir, .. }) => {
union_required(&**hir, lits);
}
HirKind::Repetition(ref x) => match x.kind {
hir::RepetitionKind::ZeroOrOne => lits.cut(),
hir::RepetitionKind::ZeroOrMore => lits.cut(),
hir::RepetitionKind::OneOrMore => {
union_required(&x.hir, lits);
}
hir::RepetitionKind::Range(ref rng) => {
let (min, max) = match *rng {
hir::RepetitionRange::Exactly(m) => (m, Some(m)),
hir::RepetitionRange::AtLeast(m) => (m, None),
hir::RepetitionRange::Bounded(m, n) => (m, Some(n)),
};
repeat_range_literals(
&x.hir,
min,
max,
x.greedy,
lits,
union_required,
);
}
},
HirKind::Concat(ref es) if es.is_empty() => {}
HirKind::Concat(ref es) if es.len() == 1 => {
union_required(&es[0], lits)
}
HirKind::Concat(ref es) => {
for e in es {
let mut lits2 = lits.to_empty();
union_required(e, &mut lits2);
if lits2.is_empty() {
lits.cut();
continue;
}
if lits2.contains_empty() || !is_simple(&e) {
lits.cut();
}
if !lits.cross_product(&lits2) || !lits2.any_complete() {
// If this expression couldn't yield any literal that
// could be extended, then we need to quit. Since we're
// short-circuiting, we also need to freeze every member.
lits.cut();
break;
}
}
}
HirKind::Alternation(ref es) => {
alternate_literals(es, lits, union_required);
}
_ => lits.cut(),
}
}
fn repeat_range_literals<F: FnMut(&Hir, &mut Literals)>(
e: &Hir,
min: u32,
_max: Option<u32>,
_greedy: bool,
lits: &mut Literals,
mut f: F,
) {
if min == 0 {
// This is a bit conservative. If `max` is set, then we could
// treat this as a finite set of alternations. For now, we
// just treat it as `e*`.
lits.cut();
} else {
// We only extract literals from a single repetition, even though
// we could do more. e.g., `a{3}` will have `a` extracted instead of
// `aaa`. The reason is that inner literal extraction can't be unioned
// across repetitions. e.g., extracting `foofoofoo` from `(\w+foo){3}`
// is wrong.
f(e, lits);
lits.cut();
}
}
fn alternate_literals<F: FnMut(&Hir, &mut Literals)>(
es: &[Hir],
lits: &mut Literals,
mut f: F,
) {
let mut lits2 = lits.to_empty();
for e in es {
let mut lits3 = lits.to_empty();
lits3.set_limit_size(lits.limit_size() / 5);
f(e, &mut lits3);
if lits3.is_empty() || !lits2.union(lits3) {
// If we couldn't find suffixes for *any* of the
// alternates, then the entire alternation has to be thrown
// away and any existing members must be frozen. Similarly,
// if the union couldn't complete, stop and freeze.
lits.cut();
return;
}
}
// All we do at the moment is look for prefixes and suffixes. If both
// are empty, then we report nothing. We should be able to do better than
// this, but we'll need something more expressive than just a "set of
// literals."
let lcp = lits2.longest_common_prefix();
let lcs = lits2.longest_common_suffix();
if !lcp.is_empty() {
lits.cross_add(lcp);
}
lits.cut();
if !lcs.is_empty() {
lits.add(Literal::empty());
lits.add(Literal::new(lcs.to_vec()));
}
}
fn is_simple(expr: &Hir) -> bool {
match *expr.kind() {
HirKind::Empty
| HirKind::Literal(_)
| HirKind::Class(_)
| HirKind::Repetition(_)
| HirKind::Concat(_)
| HirKind::Alternation(_) => true,
HirKind::Anchor(_) | HirKind::WordBoundary(_) | HirKind::Group(_) => {
false
}
}
}
/// Return the number of characters in the given class.
fn count_unicode_class(cls: &hir::ClassUnicode) -> u32 {
cls.iter().map(|r| 1 + (r.end() as u32 - r.start() as u32)).sum()
}
/// Return the number of bytes in the given class.
fn count_byte_class(cls: &hir::ClassBytes) -> u32 {
cls.iter().map(|r| 1 + (r.end() as u32 - r.start() as u32)).sum()
}
#[cfg(test)]
mod tests {
use super::LiteralSets;
use regex_syntax::Parser;
fn sets(pattern: &str) -> LiteralSets {
let hir = Parser::new().parse(pattern).unwrap();
LiteralSets::new(&hir)
}
fn one_regex(pattern: &str) -> Option<String> {
sets(pattern).one_regex(false)
}
// Put a pattern into the same format as the one returned by `one_regex`.
fn pat(pattern: &str) -> Option<String> {
Some(format!("(?-u:{})", pattern))
}
#[test]
fn various() {
// Obviously no literals.
assert!(one_regex(r"\w").is_none());
assert!(one_regex(r"\pL").is_none());
// Tantalizingly close.
assert!(one_regex(r"\w|foo").is_none());
// There's a literal, but it's better if the regex engine handles it
// internally.
assert!(one_regex(r"abc").is_none());
// Core use cases.
assert_eq!(one_regex(r"\wabc\w"), pat("abc"));
assert_eq!(one_regex(r"abc\w"), pat("abc"));
// TODO: Make these pass. We're missing some potentially big wins
// without these.
// assert_eq!(one_regex(r"\w(foo|bar|baz)"), pat("foo|bar|baz"));
// assert_eq!(one_regex(r"\w(foo|bar|baz)\w"), pat("foo|bar|baz"));
}
#[test]
fn regression_1064() {
// Regression from:
// https://github.com/BurntSushi/ripgrep/issues/1064
// assert_eq!(one_regex(r"a.*c"), pat("a"));
assert_eq!(one_regex(r"a(.*c)"), pat("a"));
}
#[test]
fn regression_1319() {
// Regression from:
// https://github.com/BurntSushi/ripgrep/issues/1319
assert_eq!(
one_regex(r"TTGAGTCCAGGAG[ATCG]{2}C"),
pat("TTGAGTCCAGGAGA|TTGAGTCCAGGAGC|\
TTGAGTCCAGGAGG|TTGAGTCCAGGAGT")
);
}
}

1066
crates/regex/src/matcher.rs Normal file
View File

File diff suppressed because it is too large Load Diff

125
crates/regex/src/multi.rs Normal file
View File

@ -0,0 +1,125 @@
use aho_corasick::{AhoCorasick, AhoCorasickBuilder, MatchKind};
use grep_matcher::{Match, Matcher, NoError};
use regex_syntax::hir::Hir;
use error::Error;
use matcher::RegexCaptures;
/// A matcher for an alternation of literals.
///
/// Ideally, this optimization would be pushed down into the regex engine, but
/// making this work correctly there would require quite a bit of refactoring.
/// Moreover, doing it one layer above lets us do thing like, "if we
/// specifically only want to search for literals, then don't bother with
/// regex parsing at all."
#[derive(Clone, Debug)]
pub struct MultiLiteralMatcher {
/// The Aho-Corasick automaton.
ac: AhoCorasick,
}
impl MultiLiteralMatcher {
/// Create a new multi-literal matcher from the given literals.
pub fn new<B: AsRef<[u8]>>(
literals: &[B],
) -> Result<MultiLiteralMatcher, Error> {
let ac = AhoCorasickBuilder::new()
.match_kind(MatchKind::LeftmostFirst)
.auto_configure(literals)
.build_with_size::<usize, _, _>(literals)
.map_err(Error::regex)?;
Ok(MultiLiteralMatcher { ac })
}
}
impl Matcher for MultiLiteralMatcher {
type Captures = RegexCaptures;
type Error = NoError;
fn find_at(
&self,
haystack: &[u8],
at: usize,
) -> Result<Option<Match>, NoError> {
match self.ac.find(&haystack[at..]) {
None => Ok(None),
Some(m) => Ok(Some(Match::new(at + m.start(), at + m.end()))),
}
}
fn new_captures(&self) -> Result<RegexCaptures, NoError> {
Ok(RegexCaptures::simple())
}
fn capture_count(&self) -> usize {
1
}
fn capture_index(&self, _: &str) -> Option<usize> {
None
}
fn captures_at(
&self,
haystack: &[u8],
at: usize,
caps: &mut RegexCaptures,
) -> Result<bool, NoError> {
caps.set_simple(None);
let mat = self.find_at(haystack, at)?;
caps.set_simple(mat);
Ok(mat.is_some())
}
// We specifically do not implement other methods like find_iter. Namely,
// the iter methods are guaranteed to be correct by virtue of implementing
// find_at above.
}
/// Alternation literals checks if the given HIR is a simple alternation of
/// literals, and if so, returns them. Otherwise, this returns None.
pub fn alternation_literals(expr: &Hir) -> Option<Vec<Vec<u8>>> {
use regex_syntax::hir::{HirKind, Literal};
// This is pretty hacky, but basically, if `is_alternation_literal` is
// true, then we can make several assumptions about the structure of our
// HIR. This is what justifies the `unreachable!` statements below.
if !expr.is_alternation_literal() {
return None;
}
let alts = match *expr.kind() {
HirKind::Alternation(ref alts) => alts,
_ => return None, // one literal isn't worth it
};
let extendlit = |lit: &Literal, dst: &mut Vec<u8>| match *lit {
Literal::Unicode(c) => {
let mut buf = [0; 4];
dst.extend_from_slice(c.encode_utf8(&mut buf).as_bytes());
}
Literal::Byte(b) => {
dst.push(b);
}
};
let mut lits = vec![];
for alt in alts {
let mut lit = vec![];
match *alt.kind() {
HirKind::Empty => {}
HirKind::Literal(ref x) => extendlit(x, &mut lit),
HirKind::Concat(ref exprs) => {
for e in exprs {
match *e.kind() {
HirKind::Literal(ref x) => extendlit(x, &mut lit),
_ => unreachable!("expected literal, got {:?}", e),
}
}
}
_ => unreachable!("expected literal or concat, got {:?}", alt),
}
lits.push(lit);
}
Some(lits)
}

128
crates/regex/src/non_matching.rs Normal file
View File

@ -0,0 +1,128 @@
use grep_matcher::ByteSet;
use regex_syntax::hir::{self, Hir, HirKind};
use regex_syntax::utf8::Utf8Sequences;
/// Return a confirmed set of non-matching bytes from the given expression.
pub fn non_matching_bytes(expr: &Hir) -> ByteSet {
let mut set = ByteSet::full();
remove_matching_bytes(expr, &mut set);
set
}
/// Remove any bytes from the given set that can occur in a matched produced by
/// the given expression.
fn remove_matching_bytes(expr: &Hir, set: &mut ByteSet) {
match *expr.kind() {
HirKind::Empty | HirKind::Anchor(_) | HirKind::WordBoundary(_) => {}
HirKind::Literal(hir::Literal::Unicode(c)) => {
for &b in c.encode_utf8(&mut [0; 4]).as_bytes() {
set.remove(b);
}
}
HirKind::Literal(hir::Literal::Byte(b)) => {
set.remove(b);
}
HirKind::Class(hir::Class::Unicode(ref cls)) => {
for range in cls.iter() {
// This is presumably faster than encoding every codepoint
// to UTF-8 and then removing those bytes from the set.
for seq in Utf8Sequences::new(range.start(), range.end()) {
for byte_range in seq.as_slice() {
set.remove_all(byte_range.start, byte_range.end);
}
}
}
}
HirKind::Class(hir::Class::Bytes(ref cls)) => {
for range in cls.iter() {
set.remove_all(range.start(), range.end());
}
}
HirKind::Repetition(ref x) => {
remove_matching_bytes(&x.hir, set);
}
HirKind::Group(ref x) => {
remove_matching_bytes(&x.hir, set);
}
HirKind::Concat(ref xs) => {
for x in xs {
remove_matching_bytes(x, set);
}
}
HirKind::Alternation(ref xs) => {
for x in xs {
remove_matching_bytes(x, set);
}
}
}
}
#[cfg(test)]
mod tests {
use grep_matcher::ByteSet;
use regex_syntax::ParserBuilder;
use super::non_matching_bytes;
fn extract(pattern: &str) -> ByteSet {
let expr = ParserBuilder::new()
.allow_invalid_utf8(true)
.build()
.parse(pattern)
.unwrap();
non_matching_bytes(&expr)
}
fn sparse(set: &ByteSet) -> Vec<u8> {
let mut sparse_set = vec![];
for b in (0..256).map(|b| b as u8) {
if set.contains(b) {
sparse_set.push(b);
}
}
sparse_set
}
fn sparse_except(except: &[u8]) -> Vec<u8> {
let mut except_set = vec![false; 256];
for &b in except {
except_set[b as usize] = true;
}
let mut set = vec![];
for b in (0..256).map(|b| b as u8) {
if !except_set[b as usize] {
set.push(b);
}
}
set
}
#[test]
fn dot() {
assert_eq!(
sparse(&extract(".")),
vec![
b'\n', 192, 193, 245, 246, 247, 248, 249, 250, 251, 252, 253,
254, 255,
]
);
assert_eq!(
sparse(&extract("(?s).")),
vec![
192, 193, 245, 246, 247, 248, 249, 250, 251, 252, 253, 254,
255,
]
);
assert_eq!(sparse(&extract("(?-u).")), vec![b'\n']);
assert_eq!(sparse(&extract("(?s-u).")), vec![]);
}
#[test]
fn literal() {
assert_eq!(sparse(&extract("a")), sparse_except(&[b'a']));
assert_eq!(sparse(&extract("☃")), sparse_except(&[0xE2, 0x98, 0x83]));
assert_eq!(sparse(&extract(r"\xFF")), sparse_except(&[0xC3, 0xBF]));
assert_eq!(sparse(&extract(r"(?-u)\xFF")), sparse_except(&[0xFF]));
}
}

153
crates/regex/src/strip.rs Normal file
View File

@ -0,0 +1,153 @@
use grep_matcher::LineTerminator;
use regex_syntax::hir::{self, Hir, HirKind};
use error::{Error, ErrorKind};
/// Return an HIR that is guaranteed to never match the given line terminator,
/// if possible.
///
/// If the transformation isn't possible, then an error is returned.
///
/// In general, if a literal line terminator occurs anywhere in the HIR, then
/// this will return an error. However, if the line terminator occurs within
/// a character class with at least one other character (that isn't also a line
/// terminator), then the line terminator is simply stripped from that class.
///
/// If the given line terminator is not ASCII, then this function returns an
/// error.
pub fn strip_from_match(
expr: Hir,
line_term: LineTerminator,
) -> Result<Hir, Error> {
if line_term.is_crlf() {
let expr1 = strip_from_match_ascii(expr, b'\r')?;
strip_from_match_ascii(expr1, b'\n')
} else {
let b = line_term.as_byte();
if b > 0x7F {
return Err(Error::new(ErrorKind::InvalidLineTerminator(b)));
}
strip_from_match_ascii(expr, b)
}
}
/// The implementation of strip_from_match. The given byte must be ASCII. This
/// function panics otherwise.
fn strip_from_match_ascii(expr: Hir, byte: u8) -> Result<Hir, Error> {
assert!(byte <= 0x7F);
let chr = byte as char;
assert_eq!(chr.len_utf8(), 1);
let invalid = || Err(Error::new(ErrorKind::NotAllowed(chr.to_string())));
Ok(match expr.into_kind() {
HirKind::Empty => Hir::empty(),
HirKind::Literal(hir::Literal::Unicode(c)) => {
if c == chr {
return invalid();
}
Hir::literal(hir::Literal::Unicode(c))
}
HirKind::Literal(hir::Literal::Byte(b)) => {
if b as char == chr {
return invalid();
}
Hir::literal(hir::Literal::Byte(b))
}
HirKind::Class(hir::Class::Unicode(mut cls)) => {
let remove = hir::ClassUnicode::new(Some(
hir::ClassUnicodeRange::new(chr, chr),
));
cls.difference(&remove);
if cls.ranges().is_empty() {
return invalid();
}
Hir::class(hir::Class::Unicode(cls))
}
HirKind::Class(hir::Class::Bytes(mut cls)) => {
let remove = hir::ClassBytes::new(Some(
hir::ClassBytesRange::new(byte, byte),
));
cls.difference(&remove);
if cls.ranges().is_empty() {
return invalid();
}
Hir::class(hir::Class::Bytes(cls))
}
HirKind::Anchor(x) => Hir::anchor(x),
HirKind::WordBoundary(x) => Hir::word_boundary(x),
HirKind::Repetition(mut x) => {
x.hir = Box::new(strip_from_match_ascii(*x.hir, byte)?);
Hir::repetition(x)
}
HirKind::Group(mut x) => {
x.hir = Box::new(strip_from_match_ascii(*x.hir, byte)?);
Hir::group(x)
}
HirKind::Concat(xs) => {
let xs = xs
.into_iter()
.map(|e| strip_from_match_ascii(e, byte))
.collect::<Result<Vec<Hir>, Error>>()?;
Hir::concat(xs)
}
HirKind::Alternation(xs) => {
let xs = xs
.into_iter()
.map(|e| strip_from_match_ascii(e, byte))
.collect::<Result<Vec<Hir>, Error>>()?;
Hir::alternation(xs)
}
})
}
#[cfg(test)]
mod tests {
use regex_syntax::Parser;
use super::{strip_from_match, LineTerminator};
use error::Error;
fn roundtrip(pattern: &str, byte: u8) -> String {
roundtrip_line_term(pattern, LineTerminator::byte(byte)).unwrap()
}
fn roundtrip_crlf(pattern: &str) -> String {
roundtrip_line_term(pattern, LineTerminator::crlf()).unwrap()
}
fn roundtrip_err(pattern: &str, byte: u8) -> Result<String, Error> {
roundtrip_line_term(pattern, LineTerminator::byte(byte))
}
fn roundtrip_line_term(
pattern: &str,
line_term: LineTerminator,
) -> Result<String, Error> {
let expr1 = Parser::new().parse(pattern).unwrap();
let expr2 = strip_from_match(expr1, line_term)?;
Ok(expr2.to_string())
}
#[test]
fn various() {
assert_eq!(roundtrip(r"[a\n]", b'\n'), "[a]");
assert_eq!(roundtrip(r"[a\n]", b'a'), "[\n]");
assert_eq!(roundtrip_crlf(r"[a\n]"), "[a]");
assert_eq!(roundtrip_crlf(r"[a\r]"), "[a]");
assert_eq!(roundtrip_crlf(r"[a\r\n]"), "[a]");
assert_eq!(roundtrip(r"(?-u)\s", b'a'), r"(?-u:[\x09-\x0D\x20])");
assert_eq!(roundtrip(r"(?-u)\s", b'\n'), r"(?-u:[\x09\x0B-\x0D\x20])");
assert!(roundtrip_err(r"\n", b'\n').is_err());
assert!(roundtrip_err(r"abc\n", b'\n').is_err());
assert!(roundtrip_err(r"\nabc", b'\n').is_err());
assert!(roundtrip_err(r"abc\nxyz", b'\n').is_err());
assert!(roundtrip_err(r"\x0A", b'\n').is_err());
assert!(roundtrip_err(r"\u000A", b'\n').is_err());
assert!(roundtrip_err(r"\U0000000A", b'\n').is_err());
assert!(roundtrip_err(r"\u{A}", b'\n').is_err());
assert!(roundtrip_err("\n", b'\n').is_err());
}
}

29
crates/regex/src/util.rs Normal file
View File

@ -0,0 +1,29 @@
/// Converts an arbitrary sequence of bytes to a literal suitable for building
/// a regular expression.
pub fn bytes_to_regex(bs: &[u8]) -> String {
use regex_syntax::is_meta_character;
use std::fmt::Write;
let mut s = String::with_capacity(bs.len());
for &b in bs {
if b <= 0x7F && !is_meta_character(b as char) {
write!(s, r"{}", b as char).unwrap();
} else {
write!(s, r"\x{:02x}", b).unwrap();
}
}
s
}
/// Converts arbitrary bytes to a nice string.
pub fn show_bytes(bs: &[u8]) -> String {
use std::ascii::escape_default;
use std::str;
let mut nice = String::new();
for &b in bs {
let part: Vec<u8> = escape_default(b).collect();
nice.push_str(str::from_utf8(&part).unwrap());
}
nice
}

292
crates/regex/src/word.rs Normal file
View File

@ -0,0 +1,292 @@
use std::cell::RefCell;
use std::collections::HashMap;
use std::sync::Arc;
use grep_matcher::{Match, Matcher, NoError};
use regex::bytes::{CaptureLocations, Regex};
use thread_local::CachedThreadLocal;
use config::ConfiguredHIR;
use error::Error;
use matcher::RegexCaptures;
/// A matcher for implementing "word match" semantics.
#[derive(Debug)]
pub struct WordMatcher {
/// The regex which is roughly `(?:^|\W)(<original pattern>)(?:$|\W)`.
regex: Regex,
/// The original regex supplied by the user, which we use in a fast path
/// to try and detect matches before deferring to slower engines.
original: Regex,
/// A map from capture group name to capture group index.
names: HashMap<String, usize>,
/// A reusable buffer for finding the match location of the inner group.
locs: Arc<CachedThreadLocal<RefCell<CaptureLocations>>>,
}
impl Clone for WordMatcher {
fn clone(&self) -> WordMatcher {
// We implement Clone manually so that we get a fresh CachedThreadLocal
// such that it can set its own thread owner. This permits each thread
// usings `locs` to hit the fast path.
WordMatcher {
regex: self.regex.clone(),
original: self.original.clone(),
names: self.names.clone(),
locs: Arc::new(CachedThreadLocal::new()),
}
}
}
impl WordMatcher {
/// Create a new matcher from the given pattern that only produces matches
/// that are considered "words."
///
/// The given options are used to construct the regular expression
/// internally.
pub fn new(expr: &ConfiguredHIR) -> Result<WordMatcher, Error> {
let original =
expr.with_pattern(|pat| format!("^(?:{})$", pat))?.regex()?;
let word_expr = expr.with_pattern(|pat| {
let pat = format!(r"(?:(?-m:^)|\W)({})(?:(?-m:$)|\W)", pat);
debug!("word regex: {:?}", pat);
pat
})?;
let regex = word_expr.regex()?;
let locs = Arc::new(CachedThreadLocal::new());
let mut names = HashMap::new();
for (i, optional_name) in regex.capture_names().enumerate() {
if let Some(name) = optional_name {
names.insert(name.to_string(), i.checked_sub(1).unwrap());
}
}
Ok(WordMatcher { regex, original, names, locs })
}
/// Return the underlying regex used by this matcher.
pub fn regex(&self) -> &Regex {
&self.regex
}
/// Attempt to do a fast confirmation of a word match that covers a subset
/// (but hopefully a big subset) of most cases. Ok(Some(..)) is returned
/// when a match is found. Ok(None) is returned when there is definitively
/// no match. Err(()) is returned when this routine could not detect
/// whether there was a match or not.
fn fast_find(
&self,
haystack: &[u8],
at: usize,
) -> Result<Option<Match>, ()> {
// This is a bit hairy. The whole point here is to avoid running an
// NFA simulation in the regex engine. Remember, our word regex looks
// like this:
//
// (^|\W)(<original regex>)($|\W)
// where ^ and $ have multiline mode DISABLED
//
// What we want are the match offsets of <original regex>. So in the
// easy/common case, the original regex will be sandwiched between
// two codepoints that are in the \W class. So our approach here is to
// look for a match of the overall word regexp, strip the \W ends and
// then check whether the original regex matches what's left. If so,
// then we are guaranteed a correct match.
//
// This only works though if we know that the match is sandwiched
// between two \W codepoints. This only occurs when neither ^ nor $
// match. This in turn only occurs when the match is at either the
// beginning or end of the haystack. In either of those cases, we
// declare defeat and defer to the slower implementation.
//
// The reason why we cannot handle the ^/$ cases here is because we
// can't assume anything about the original pattern. (Try commenting
// out the checks for ^/$ below and run the tests to see examples.)
let mut cand = match self.regex.find_at(haystack, at) {
None => return Ok(None),
Some(m) => Match::new(m.start(), m.end()),
};
if cand.start() == 0 || cand.end() == haystack.len() {
return Err(());
}
let (_, slen) = bstr::decode_utf8(&haystack[cand]);
let (_, elen) = bstr::decode_last_utf8(&haystack[cand]);
cand =
cand.with_start(cand.start() + slen).with_end(cand.end() - elen);
if self.original.is_match(&haystack[cand]) {
Ok(Some(cand))
} else {
Err(())
}
}
}
impl Matcher for WordMatcher {
type Captures = RegexCaptures;
type Error = NoError;
fn find_at(
&self,
haystack: &[u8],
at: usize,
) -> Result<Option<Match>, NoError> {
// To make this easy to get right, we extract captures here instead of
// calling `find_at`. The actual match is at capture group `1` instead
// of `0`. We *could* use `find_at` here and then trim the match after
// the fact, but that's a bit harder to get right, and it's not clear
// if it's worth it.
//
// OK, well, it turns out that it is worth it! But it is quite tricky.
// See `fast_find` for details. Effectively, this lets us skip running
// the NFA simulation in the regex engine in the vast majority of
// cases. However, the NFA simulation is required for full correctness.
match self.fast_find(haystack, at) {
Ok(Some(m)) => return Ok(Some(m)),
Ok(None) => return Ok(None),
Err(()) => {}
}
let cell =
self.locs.get_or(|| RefCell::new(self.regex.capture_locations()));
let mut caps = cell.borrow_mut();
self.regex.captures_read_at(&mut caps, haystack, at);
Ok(caps.get(1).map(|m| Match::new(m.0, m.1)))
}
fn new_captures(&self) -> Result<RegexCaptures, NoError> {
Ok(RegexCaptures::with_offset(self.regex.capture_locations(), 1))
}
fn capture_count(&self) -> usize {
self.regex.captures_len().checked_sub(1).unwrap()
}
fn capture_index(&self, name: &str) -> Option<usize> {
self.names.get(name).map(|i| *i)
}
fn captures_at(
&self,
haystack: &[u8],
at: usize,
caps: &mut RegexCaptures,
) -> Result<bool, NoError> {
let r =
self.regex.captures_read_at(caps.locations_mut(), haystack, at);
Ok(r.is_some())
}
// We specifically do not implement other methods like find_iter or
// captures_iter. Namely, the iter methods are guaranteed to be correct
// by virtue of implementing find_at and captures_at above.
}
#[cfg(test)]
mod tests {
use super::WordMatcher;
use config::Config;
use grep_matcher::{Captures, Match, Matcher};
fn matcher(pattern: &str) -> WordMatcher {
let chir = Config::default().hir(pattern).unwrap();
WordMatcher::new(&chir).unwrap()
}
fn find(pattern: &str, haystack: &str) -> Option<(usize, usize)> {
matcher(pattern)
.find(haystack.as_bytes())
.unwrap()
.map(|m| (m.start(), m.end()))
}
fn find_by_caps(pattern: &str, haystack: &str) -> Option<(usize, usize)> {
let m = matcher(pattern);
let mut caps = m.new_captures().unwrap();
if !m.captures(haystack.as_bytes(), &mut caps).unwrap() {
None
} else {
caps.get(0).map(|m| (m.start(), m.end()))
}
}
// Test that the standard `find` API reports offsets correctly.
#[test]
fn various_find() {
assert_eq!(Some((0, 3)), find(r"foo", "foo"));
assert_eq!(Some((0, 3)), find(r"foo", "foo("));
assert_eq!(Some((1, 4)), find(r"foo", "!foo("));
assert_eq!(None, find(r"foo", "!afoo("));
assert_eq!(Some((0, 3)), find(r"foo", "foo☃"));
assert_eq!(None, find(r"foo", "fooб"));
assert_eq!(Some((0, 4)), find(r"foo5", "foo5"));
assert_eq!(None, find(r"foo", "foo5"));
assert_eq!(Some((1, 4)), find(r"foo", "!foo!"));
assert_eq!(Some((1, 5)), find(r"foo!", "!foo!"));
assert_eq!(Some((0, 5)), find(r"!foo!", "!foo!"));
assert_eq!(Some((0, 3)), find(r"foo", "foo\n"));
assert_eq!(Some((1, 4)), find(r"foo", "!foo!\n"));
assert_eq!(Some((1, 5)), find(r"foo!", "!foo!\n"));
assert_eq!(Some((0, 5)), find(r"!foo!", "!foo!\n"));
assert_eq!(Some((1, 6)), find(r"!?foo!?", "!!foo!!"));
assert_eq!(Some((0, 5)), find(r"!?foo!?", "!foo!"));
assert_eq!(Some((2, 5)), find(r"!?foo!?", "a!foo!a"));
assert_eq!(Some((2, 7)), find(r"!?foo!?", "##!foo!\n"));
assert_eq!(Some((3, 7)), find(r"f?oo!?", "##\nfoo!##"));
assert_eq!(Some((2, 5)), find(r"(?-u)foo[^a]*", "#!foo☃aaa"));
}
// See: https://github.com/BurntSushi/ripgrep/issues/389
#[test]
fn regression_dash() {
assert_eq!(Some((0, 2)), find(r"-2", "-2"));
}
// Test that the captures API also reports offsets correctly, just as
// find does. This exercises a different path in the code since captures
// are handled differently.
#[test]
fn various_captures() {
assert_eq!(Some((0, 3)), find_by_caps(r"foo", "foo"));
assert_eq!(Some((0, 3)), find_by_caps(r"foo", "foo("));
assert_eq!(Some((1, 4)), find_by_caps(r"foo", "!foo("));
assert_eq!(None, find_by_caps(r"foo", "!afoo("));
assert_eq!(Some((0, 3)), find_by_caps(r"foo", "foo☃"));
assert_eq!(None, find_by_caps(r"foo", "fooб"));
// assert_eq!(Some((0, 3)), find_by_caps(r"foo", "fooб"));
// See: https://github.com/BurntSushi/ripgrep/issues/389
assert_eq!(Some((0, 2)), find_by_caps(r"-2", "-2"));
}
// Test that the capture reporting methods work as advertised.
#[test]
fn capture_indexing() {
let m = matcher(r"(a)(?P<foo>b)(c)");
assert_eq!(4, m.capture_count());
assert_eq!(Some(2), m.capture_index("foo"));
let mut caps = m.new_captures().unwrap();
assert_eq!(4, caps.len());
assert!(m.captures(b"abc", &mut caps).unwrap());
assert_eq!(caps.get(0), Some(Match::new(0, 3)));
assert_eq!(caps.get(1), Some(Match::new(0, 1)));
assert_eq!(caps.get(2), Some(Match::new(1, 2)));
assert_eq!(caps.get(3), Some(Match::new(2, 3)));
assert_eq!(caps.get(4), None);
assert!(m.captures(b"#abc#", &mut caps).unwrap());
assert_eq!(caps.get(0), Some(Match::new(1, 4)));
assert_eq!(caps.get(1), Some(Match::new(1, 2)));
assert_eq!(caps.get(2), Some(Match::new(2, 3)));
assert_eq!(caps.get(3), Some(Match::new(3, 4)));
assert_eq!(caps.get(4), None);
}
}