README.md
1regex-syntax
2============
3This crate provides a robust regular expression parser.
4
5[](https://travis-ci.com/rust-lang/regex)
6[](https://ci.appveyor.com/project/rust-lang-libs/regex)
7[](https://crates.io/crates/regex-syntax)
8[](https://github.com/rust-lang/regex)
9
10
11### Documentation
12
13https://docs.rs/regex-syntax
14
15
16### Overview
17
18There are two primary types exported by this crate: `Ast` and `Hir`. The former
19is a faithful abstract syntax of a regular expression, and can convert regular
20expressions back to their concrete syntax while mostly preserving its original
21form. The latter type is a high level intermediate representation of a regular
22expression that is amenable to analysis and compilation into byte codes or
23automata. An `Hir` achieves this by drastically simplifying the syntactic
24structure of the regular expression. While an `Hir` can be converted back to
25its equivalent concrete syntax, the result is unlikely to resemble the original
26concrete syntax that produced the `Hir`.
27
28
29### Example
30
31This example shows how to parse a pattern string into its HIR:
32
33```rust
34use regex_syntax::Parser;
35use regex_syntax::hir::{self, Hir};
36
37let hir = Parser::new().parse("a|b").unwrap();
38assert_eq!(hir, Hir::alternation(vec![
39 Hir::literal(hir::Literal::Unicode('a')),
40 Hir::literal(hir::Literal::Unicode('b')),
41]));
42```
43
44
45### Safety
46
47This crate has no `unsafe` code and sets `forbid(unsafe_code)`. While it's
48possible this crate could use `unsafe` code in the future, the standard
49for doing so is extremely high. In general, most code in this crate is not
50performance critical, since it tends to be dwarfed by the time it takes to
51compile a regular expression into an automaton. Therefore, there is little need
52for extreme optimization, and therefore, use of `unsafe`.
53
54The standard for using `unsafe` in this crate is extremely high because this
55crate is intended to be reasonably safe to use with user supplied regular
56expressions. Therefore, while their may be bugs in the regex parser itself,
57they should _never_ result in memory unsafety unless there is either a bug
58in the compiler or the standard library. (Since `regex-syntax` has zero
59dependencies.)
60
61
62### Crate features
63
64By default, this crate bundles a fairly large amount of Unicode data tables
65(a source size of ~750KB). Because of their large size, one can disable some
66or all of these data tables. If a regular expression attempts to use Unicode
67data that is not available, then an error will occur when translating the `Ast`
68to the `Hir`.
69
70The full set of features one can disable are
71[in the "Crate features" section of the documentation](https://docs.rs/regex-syntax/*/#crate-features).
72
73
74### Testing
75
76Simply running `cargo test` will give you very good coverage. However, because
77of the large number of features exposed by this crate, a `test` script is
78included in this directory which will test several feature combinations. This
79is the same script that is run in CI.
80
81
82### Motivation
83
84The primary purpose of this crate is to provide the parser used by `regex`.
85Specifically, this crate is treated as an implementation detail of the `regex`,
86and is primarily developed for the needs of `regex`.
87
88Since this crate is an implementation detail of `regex`, it may experience
89breaking change releases at a different cadence from `regex`. This is only
90possible because this crate is _not_ a public dependency of `regex`.
91
92Another consequence of this de-coupling is that there is no direct way to
93compile a `regex::Regex` from a `regex_syntax::hir::Hir`. Instead, one must
94first convert the `Hir` to a string (via its `std::fmt::Display`) and then
95compile that via `Regex::new`. While this does repeat some work, compilation
96typically takes much longer than parsing.
97
98Stated differently, the coupling between `regex` and `regex-syntax` exists only
99at the level of the concrete syntax.
100
){kind=link}
){kind=link}