1 | // Copyright 2012-2015 The Rust Project Developers. See the COPYRIGHT |
2 | // file at the top-level directory of this distribution and at |
3 | // http://rust-lang.org/COPYRIGHT. |
4 | // |
5 | // Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or |
6 | // http://www.apache.org/licenses/LICENSE-2.0> or the MIT license |
7 | // <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your |
8 | // option. This file may not be copied, modified, or distributed |
9 | // except according to those terms. |
10 | use core::fmt::{self, Write}; |
11 | use core::iter::Fuse; |
12 | use core::ops::Range; |
13 | use tinyvec::TinyVec; |
14 | |
15 | #[derive (Clone)] |
16 | enum DecompositionType { |
17 | Canonical, |
18 | Compatible, |
19 | } |
20 | |
21 | /// External iterator for a string decomposition's characters. |
22 | #[derive (Clone)] |
23 | pub struct Decompositions<I> { |
24 | kind: DecompositionType, |
25 | iter: Fuse<I>, |
26 | |
27 | // This buffer stores pairs of (canonical combining class, character), |
28 | // pushed onto the end in text order. |
29 | // |
30 | // It's divided into up to three sections: |
31 | // 1) A prefix that is free space; |
32 | // 2) "Ready" characters which are sorted and ready to emit on demand; |
33 | // 3) A "pending" block which stills needs more characters for us to be able |
34 | // to sort in canonical order and is not safe to emit. |
35 | buffer: TinyVec<[(u8, char); 4]>, |
36 | ready: Range<usize>, |
37 | } |
38 | |
39 | #[inline ] |
40 | pub fn new_canonical<I: Iterator<Item = char>>(iter: I) -> Decompositions<I> { |
41 | Decompositions { |
42 | kind: self::DecompositionType::Canonical, |
43 | iter: iter.fuse(), |
44 | buffer: TinyVec::new(), |
45 | ready: 0..0, |
46 | } |
47 | } |
48 | |
49 | #[inline ] |
50 | pub fn new_compatible<I: Iterator<Item = char>>(iter: I) -> Decompositions<I> { |
51 | Decompositions { |
52 | kind: self::DecompositionType::Compatible, |
53 | iter: iter.fuse(), |
54 | buffer: TinyVec::new(), |
55 | ready: 0..0, |
56 | } |
57 | } |
58 | |
59 | impl<I> Decompositions<I> { |
60 | #[inline ] |
61 | fn push_back(&mut self, ch: char) { |
62 | let class = super::char::canonical_combining_class(ch); |
63 | |
64 | if class == 0 { |
65 | self.sort_pending(); |
66 | self.buffer.push((class, ch)); |
67 | self.ready.end = self.buffer.len(); |
68 | } else { |
69 | self.buffer.push((class, ch)); |
70 | } |
71 | } |
72 | |
73 | #[inline ] |
74 | fn sort_pending(&mut self) { |
75 | // NB: `sort_by_key` is stable, so it will preserve the original text's |
76 | // order within a combining class. |
77 | self.buffer[self.ready.end..].sort_by_key(|k| k.0); |
78 | } |
79 | |
80 | #[inline ] |
81 | fn reset_buffer(&mut self) { |
82 | // Equivalent to `self.buffer.drain(0..self.ready.end)` |
83 | // but faster than drain() if the buffer is a SmallVec or TinyVec |
84 | let pending = self.buffer.len() - self.ready.end; |
85 | for i in 0..pending { |
86 | self.buffer[i] = self.buffer[i + self.ready.end]; |
87 | } |
88 | self.buffer.truncate(pending); |
89 | self.ready = 0..0; |
90 | } |
91 | |
92 | #[inline ] |
93 | fn increment_next_ready(&mut self) { |
94 | let next = self.ready.start + 1; |
95 | if next == self.ready.end { |
96 | self.reset_buffer(); |
97 | } else { |
98 | self.ready.start = next; |
99 | } |
100 | } |
101 | } |
102 | |
103 | impl<I: Iterator<Item = char>> Iterator for Decompositions<I> { |
104 | type Item = char; |
105 | |
106 | #[inline ] |
107 | fn next(&mut self) -> Option<char> { |
108 | while self.ready.end == 0 { |
109 | match (self.iter.next(), &self.kind) { |
110 | (Some(ch), &DecompositionType::Canonical) => { |
111 | super::char::decompose_canonical(ch, |d| self.push_back(d)); |
112 | } |
113 | (Some(ch), &DecompositionType::Compatible) => { |
114 | super::char::decompose_compatible(ch, |d| self.push_back(d)); |
115 | } |
116 | (None, _) => { |
117 | if self.buffer.is_empty() { |
118 | return None; |
119 | } else { |
120 | self.sort_pending(); |
121 | self.ready.end = self.buffer.len(); |
122 | |
123 | // This implementation means that we can call `next` |
124 | // on an exhausted iterator; the last outer `next` call |
125 | // will result in an inner `next` call. To make this |
126 | // safe, we use `fuse`. |
127 | break; |
128 | } |
129 | } |
130 | } |
131 | } |
132 | |
133 | // We can assume here that, if `self.ready.end` is greater than zero, |
134 | // it's also greater than `self.ready.start`. That's because we only |
135 | // increment `self.ready.start` inside `increment_next_ready`, and |
136 | // whenever it reaches equality with `self.ready.end`, we reset both |
137 | // to zero, maintaining the invariant that: |
138 | // self.ready.start < self.ready.end || self.ready.end == self.ready.start == 0 |
139 | // |
140 | // This less-than-obviously-safe implementation is chosen for performance, |
141 | // minimizing the number & complexity of branches in `next` in the common |
142 | // case of buffering then unbuffering a single character with each call. |
143 | let (_, ch) = self.buffer[self.ready.start]; |
144 | self.increment_next_ready(); |
145 | Some(ch) |
146 | } |
147 | |
148 | fn size_hint(&self) -> (usize, Option<usize>) { |
149 | let (lower, _) = self.iter.size_hint(); |
150 | (lower, None) |
151 | } |
152 | } |
153 | |
154 | impl<I: Iterator<Item = char> + Clone> fmt::Display for Decompositions<I> { |
155 | fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result { |
156 | for c: char in self.clone() { |
157 | f.write_char(c)?; |
158 | } |
159 | Ok(()) |
160 | } |
161 | } |
162 | |