• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 //! # Why `winnow`?
2 //!
3 //! To answer this question, it will be useful to contrast this with other approaches to parsing.
4 //!
5 //! <div class="warning">
6 //!
7 //! **Note:** This will focus on principles and priorities. For a deeper and wider wider
8 //! comparison with other Rust parser libraries, see
9 //! [parse-rosetta-rs](https://github.com/rosetta-rs/parse-rosetta-rs).
10 //!
11 //! </div>
12 //!
13 //! ## Hand-written parsers
14 //!
15 //! Typically, a hand-written parser gives you the flexibility to get
16 //! - Fast parse performance
17 //! - Fast compile-time
18 //! - Small binary sizes
19 //! - High quality error message
20 //! - Fewer dependencies to audit
21 //!
22 //! However, this comes at the cost of doing it all yourself, including
23 //! - Optimizing for each of the above characteristics you care about
24 //! - Ensuring the safety of any `unsafe` code (buffer overflows being a common bug with parsers)
25 //! - Being aware of, familiar with, and correctly implement the relevant algorithms.
26 //!   matklad, who has written two rust compile frontends, commented
27 //!   ["I’ve implemented a production-grade Pratt parser once, but I no longer immediately understand that code :-)"](https://matklad.github.io/2020/04/13/simple-but-powerful-pratt-parsing.html)
28 //!
29 //! This approach works well if:
30 //! - Your format is small and is unlikely to change
31 //! - Your format is large but you have people who can focus solely on parsing, like with large
32 //!   programming languages
33 //!
34 //! ## `winnow`
35 //!
36 //! Unlike traditional programming language parsers that use
37 //! [lex](https://en.wikipedia.org/wiki/Lex_(software)) or
38 //! [yacc](https://en.wikipedia.org/wiki/Yacc), you can think of `winnow` as a general version of
39 //! the helpers you would create along the way to writing a hand-written parser.
40 //!
41 //! `winnow` includes support for:
42 //! - Zero-copy parsing
43 //! - [Parse traces][trace] for easier debugging
44 //! - [Streaming parsing][Partial] for network communication or large file
45 //! - [Stateful] parsers
46 //!
47 //! For binary formats, `winnow` includes:
48 //! - [A hexadecimal view][crate::Bytes] in [trace]
49 //! - [TLV](https://en.wikipedia.org/wiki/Type-length-value) (e.g. [`length_take`])
50 //! - Some common parsers to help get started, like numbers
51 //!
52 //! For text formats, `winnow` includes:
53 //! - [Tracking of spans][crate::LocatingSlice]
54 //! - [A textual view when parsing as bytes][crate::BStr] in [trace]
55 //! - Ability to evaluate directly, parse to an AST, or lex and parse the format
56 //!
57 //! This works well for:
58 //! - Prototyping for what will be a hand-written parser
59 //! - When you want to minimize the work to evolve your format
60 //! - When you don't have contributors focused solely on parsing and your grammar is large enough
61 //!   to be unwieldy to hand write.
62 //!
63 //! ## `nom`
64 //!
65 //! `winnow` is a fork of the venerable [`nom`](https://crates.io/crates/nom). The difference
66 //! between them is largely in priorities.  `nom` prioritizes:
67 //! - Lower churn for existing users while `winnow` is trying to find ways to make things better
68 //!   for the parsers yet to be written.
69 //! - Having a small core, relying on external crates like
70 //!   [`nom-locate`](https://crates.io/crates/nom_locate) and
71 //!   [`nom-supreme`](https://crates.io/crates/nom-supreme), encouraging flexibility among users
72 //!   and to not block users on new features being merged while `winnow` aims to include all the
73 //!   fundamentals for parsing to ensure the experience is cohesive and high quality.
74 //!
75 //! For more details, see the [design differences][super::nom#api-differences].
76 //!
77 //! See also our [nom migration guide][super::nom#migrating-from-nom].
78 //!
79 //! ## `chumsky`
80 //!
81 //! [`chumsky`](https://crates.io/crates/chumsky) is an up and coming parser-combinator library
82 //! that includes advanced features like error recovery.
83 //!
84 //! Probably the biggest diverging philosophy is `chumsky`s stance:
85 //!
86 //! > "If you need to implement either `Parser` or `Strategy` by hand, that's a problem that needs fixing".
87 //!
88 //! This is under "batteries included" but it also ties into the feeling that `chumsky` acts more like
89 //! a framework. Instead of composing together helpers, you are expected to do everything through
90 //! their system to the point that it is non-trivial to implement their `Parser` trait and are
91 //! encouraged to use the
92 //! [`custom`](https://docs.rs/chumsky/0.9.0/chumsky/primitive/fn.custom.html) helper. This
93 //! requires re-framing everything to fit within their model and makes the code harder to understand
94 //! and debug as you are working with abstract operations that will eventually be applied
95 //! rather than directly with the parsers.
96 //!
97 //! In contrast, `winnow` is an introspectable toolbox that can easily be customized at any level.
98 //! Probably the biggest thing that `winnow` loses out on is optimizations from ["parse modes" via
99 //! GATs](https://github.com/zesterer/chumsky/pull/82) which allows downstream parsers to tell
100 //! upstream parsers when information will be discarded, allowing bypassing expensive operations,
101 //! like allocations. This requires a lot more complex interaction with parsers that isn't as
102 //! trivial to do with bare functions which would lose out on any of that side-band information.
103 //! Instead, we work around this with things like the [`Accumulate`] trait.
104 
105 #![allow(unused_imports)]
106 use crate::binary::length_take;
107 use crate::combinator::trace;
108 use crate::stream::Accumulate;
109 use crate::stream::Partial;
110 use crate::stream::Stateful;
111