1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
//! Contains utilities for stdout and stderr.
use crate::io::AsyncWrite;
use std::pin::Pin;
use std::task::{Context, Poll};
/// # Windows
/// [`AsyncWrite`] adapter that finds last char boundary in given buffer and does not write the rest,
/// if buffer contents seems to be `utf8`. Otherwise it only trims buffer down to `DEFAULT_MAX_BUF_SIZE`.
/// That's why, wrapped writer will always receive well-formed utf-8 bytes.
/// # Other platforms
/// Passes data to `inner` as is.
#[derive(Debug)]
pub(crate) struct SplitByUtf8BoundaryIfWindows<W> {
    inner: W,
}

impl<W> SplitByUtf8BoundaryIfWindows<W> {
    pub(crate) fn new(inner: W) -> Self {
        Self { inner }
    }
}

// this constant is defined by Unicode standard.
const MAX_BYTES_PER_CHAR: usize = 4;

// Subject for tweaking here
const MAGIC_CONST: usize = 8;

impl<W> crate::io::AsyncWrite for SplitByUtf8BoundaryIfWindows<W>
where
    W: AsyncWrite + Unpin,
{
    fn poll_write(
        mut self: Pin<&mut Self>,
        cx: &mut Context<'_>,
        mut buf: &[u8],
    ) -> Poll<Result<usize, std::io::Error>> {
        // just a closure to avoid repetitive code
        let mut call_inner = move |buf| Pin::new(&mut self.inner).poll_write(cx, buf);

        // 1. Only windows stdio can suffer from non-utf8.
        // We also check for `test` so that we can write some tests
        // for further code. Since `AsyncWrite` can always shrink
        // buffer at its discretion, excessive (i.e. in tests) shrinking
        // does not break correctness.
        // 2. If buffer is small, it will not be shrunk.
        // That's why, it's "textness" will not change, so we don't have
        // to fixup it.
        if cfg!(not(any(target_os = "windows", test)))
            || buf.len() <= crate::io::blocking::DEFAULT_MAX_BUF_SIZE
        {
            return call_inner(buf);
        }

        buf = &buf[..crate::io::blocking::DEFAULT_MAX_BUF_SIZE];

        // Now there are two possibilities.
        // If caller gave is binary buffer, we **should not** shrink it
        // anymore, because excessive shrinking hits performance.
        // If caller gave as binary buffer, we  **must** additionally
        // shrink it to strip incomplete char at the end of buffer.
        // that's why check we will perform now is allowed to have
        // false-positive.

        // Now let's look at the first MAX_BYTES_PER_CHAR * MAGIC_CONST bytes.
        // if they are (possibly incomplete) utf8, then we can be quite sure
        // that input buffer was utf8.

        let have_to_fix_up = match std::str::from_utf8(&buf[..MAX_BYTES_PER_CHAR * MAGIC_CONST]) {
            Ok(_) => true,
            Err(err) => {
                let incomplete_bytes = MAX_BYTES_PER_CHAR * MAGIC_CONST - err.valid_up_to();
                incomplete_bytes < MAX_BYTES_PER_CHAR
            }
        };

        if have_to_fix_up {
            // We must pop several bytes at the end which form incomplete
            // character. To achieve it, we exploit UTF8 encoding:
            // for any code point, all bytes except first start with 0b10 prefix.
            // see https://en.wikipedia.org/wiki/UTF-8#Encoding for details
            let trailing_incomplete_char_size = buf
                .iter()
                .rev()
                .take(MAX_BYTES_PER_CHAR)
                .position(|byte| *byte < 0b1000_0000 || *byte >= 0b1100_0000)
                .unwrap_or(0)
                + 1;
            buf = &buf[..buf.len() - trailing_incomplete_char_size];
        }

        call_inner(buf)
    }

    fn poll_flush(
        mut self: Pin<&mut Self>,
        cx: &mut Context<'_>,
    ) -> Poll<Result<(), std::io::Error>> {
        Pin::new(&mut self.inner).poll_flush(cx)
    }

    fn poll_shutdown(
        mut self: Pin<&mut Self>,
        cx: &mut Context<'_>,
    ) -> Poll<Result<(), std::io::Error>> {
        Pin::new(&mut self.inner).poll_shutdown(cx)
    }
}

#[cfg(test)]
#[cfg(not(loom))]
mod tests {
    use crate::io::blocking::DEFAULT_MAX_BUF_SIZE;
    use crate::io::AsyncWriteExt;
    use std::io;
    use std::pin::Pin;
    use std::task::Context;
    use std::task::Poll;

    struct TextMockWriter;

    impl crate::io::AsyncWrite for TextMockWriter {
        fn poll_write(
            self: Pin<&mut Self>,
            _cx: &mut Context<'_>,
            buf: &[u8],
        ) -> Poll<Result<usize, io::Error>> {
            assert!(buf.len() <= DEFAULT_MAX_BUF_SIZE);
            assert!(std::str::from_utf8(buf).is_ok());
            Poll::Ready(Ok(buf.len()))
        }

        fn poll_flush(self: Pin<&mut Self>, _cx: &mut Context<'_>) -> Poll<Result<(), io::Error>> {
            Poll::Ready(Ok(()))
        }

        fn poll_shutdown(
            self: Pin<&mut Self>,
            _cx: &mut Context<'_>,
        ) -> Poll<Result<(), io::Error>> {
            Poll::Ready(Ok(()))
        }
    }

    struct LoggingMockWriter {
        write_history: Vec<usize>,
    }

    impl LoggingMockWriter {
        fn new() -> Self {
            LoggingMockWriter {
                write_history: Vec::new(),
            }
        }
    }

    impl crate::io::AsyncWrite for LoggingMockWriter {
        fn poll_write(
            mut self: Pin<&mut Self>,
            _cx: &mut Context<'_>,
            buf: &[u8],
        ) -> Poll<Result<usize, io::Error>> {
            assert!(buf.len() <= DEFAULT_MAX_BUF_SIZE);
            self.write_history.push(buf.len());
            Poll::Ready(Ok(buf.len()))
        }

        fn poll_flush(self: Pin<&mut Self>, _cx: &mut Context<'_>) -> Poll<Result<(), io::Error>> {
            Poll::Ready(Ok(()))
        }

        fn poll_shutdown(
            self: Pin<&mut Self>,
            _cx: &mut Context<'_>,
        ) -> Poll<Result<(), io::Error>> {
            Poll::Ready(Ok(()))
        }
    }

    #[test]
    #[cfg_attr(miri, ignore)]
    fn test_splitter() {
        let data = str::repeat("█", DEFAULT_MAX_BUF_SIZE);
        let mut wr = super::SplitByUtf8BoundaryIfWindows::new(TextMockWriter);
        let fut = async move {
            wr.write_all(data.as_bytes()).await.unwrap();
        };
        crate::runtime::Builder::new_current_thread()
            .build()
            .unwrap()
            .block_on(fut);
    }

    #[test]
    #[cfg_attr(miri, ignore)]
    fn test_pseudo_text() {
        // In this test we write a piece of binary data, whose beginning is
        // text though. We then validate that even in this corner case buffer
        // was not shrunk too much.
        let checked_count = super::MAGIC_CONST * super::MAX_BYTES_PER_CHAR;
        let mut data: Vec<u8> = str::repeat("a", checked_count).into();
        data.extend(std::iter::repeat(0b1010_1010).take(DEFAULT_MAX_BUF_SIZE - checked_count + 1));
        let mut writer = LoggingMockWriter::new();
        let mut splitter = super::SplitByUtf8BoundaryIfWindows::new(&mut writer);
        crate::runtime::Builder::new_current_thread()
            .build()
            .unwrap()
            .block_on(async {
                splitter.write_all(&data).await.unwrap();
            });
        // Check that at most two writes were performed
        assert!(writer.write_history.len() <= 2);
        // Check that all has been written
        assert_eq!(
            writer.write_history.iter().copied().sum::<usize>(),
            data.len()
        );
        // Check that at most MAX_BYTES_PER_CHAR + 1 (i.e. 5) bytes were shrunk
        // from the buffer: one because it was outside of DEFAULT_MAX_BUF_SIZE boundary, and
        // up to one "utf8 code point".
        assert!(data.len() - writer.write_history[0] <= super::MAX_BYTES_PER_CHAR + 1);
    }
}