/rust/registry/src/index.crates.io-6f17d22bba15001f/base64-0.13.1/src/write/encoder.rs
Line | Count | Source (jump to first uncovered line) |
1 | | use crate::encode::encode_to_slice; |
2 | | use crate::{encode_config_slice, Config}; |
3 | | use std::{ |
4 | | cmp, fmt, |
5 | | io::{ErrorKind, Result, Write}, |
6 | | }; |
7 | | |
8 | | pub(crate) const BUF_SIZE: usize = 1024; |
9 | | /// The most bytes whose encoding will fit in `BUF_SIZE` |
10 | | const MAX_INPUT_LEN: usize = BUF_SIZE / 4 * 3; |
11 | | // 3 bytes of input = 4 bytes of base64, always (because we don't allow line wrapping) |
12 | | const MIN_ENCODE_CHUNK_SIZE: usize = 3; |
13 | | |
14 | | /// A `Write` implementation that base64 encodes data before delegating to the wrapped writer. |
15 | | /// |
16 | | /// Because base64 has special handling for the end of the input data (padding, etc), there's a |
17 | | /// `finish()` method on this type that encodes any leftover input bytes and adds padding if |
18 | | /// appropriate. It's called automatically when deallocated (see the `Drop` implementation), but |
19 | | /// any error that occurs when invoking the underlying writer will be suppressed. If you want to |
20 | | /// handle such errors, call `finish()` yourself. |
21 | | /// |
22 | | /// # Examples |
23 | | /// |
24 | | /// ``` |
25 | | /// use std::io::Write; |
26 | | /// |
27 | | /// // use a vec as the simplest possible `Write` -- in real code this is probably a file, etc. |
28 | | /// let mut enc = base64::write::EncoderWriter::new(Vec::new(), base64::STANDARD); |
29 | | /// |
30 | | /// // handle errors as you normally would |
31 | | /// enc.write_all(b"asdf").unwrap(); |
32 | | /// |
33 | | /// // could leave this out to be called by Drop, if you don't care |
34 | | /// // about handling errors or getting the delegate writer back |
35 | | /// let delegate = enc.finish().unwrap(); |
36 | | /// |
37 | | /// // base64 was written to the writer |
38 | | /// assert_eq!(b"YXNkZg==", &delegate[..]); |
39 | | /// |
40 | | /// ``` |
41 | | /// |
42 | | /// # Panics |
43 | | /// |
44 | | /// Calling `write()` (or related methods) or `finish()` after `finish()` has completed without |
45 | | /// error is invalid and will panic. |
46 | | /// |
47 | | /// # Errors |
48 | | /// |
49 | | /// Base64 encoding itself does not generate errors, but errors from the wrapped writer will be |
50 | | /// returned as per the contract of `Write`. |
51 | | /// |
52 | | /// # Performance |
53 | | /// |
54 | | /// It has some minor performance loss compared to encoding slices (a couple percent). |
55 | | /// It does not do any heap allocation. |
56 | | pub struct EncoderWriter<W: Write> { |
57 | | config: Config, |
58 | | /// Where encoded data is written to. It's an Option as it's None immediately before Drop is |
59 | | /// called so that finish() can return the underlying writer. None implies that finish() has |
60 | | /// been called successfully. |
61 | | delegate: Option<W>, |
62 | | /// Holds a partial chunk, if any, after the last `write()`, so that we may then fill the chunk |
63 | | /// with the next `write()`, encode it, then proceed with the rest of the input normally. |
64 | | extra_input: [u8; MIN_ENCODE_CHUNK_SIZE], |
65 | | /// How much of `extra` is occupied, in `[0, MIN_ENCODE_CHUNK_SIZE]`. |
66 | | extra_input_occupied_len: usize, |
67 | | /// Buffer to encode into. May hold leftover encoded bytes from a previous write call that the underlying writer |
68 | | /// did not write last time. |
69 | | output: [u8; BUF_SIZE], |
70 | | /// How much of `output` is occupied with encoded data that couldn't be written last time |
71 | | output_occupied_len: usize, |
72 | | /// panic safety: don't write again in destructor if writer panicked while we were writing to it |
73 | | panicked: bool, |
74 | | } |
75 | | |
76 | | impl<W: Write> fmt::Debug for EncoderWriter<W> { |
77 | 0 | fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result { |
78 | 0 | write!( |
79 | 0 | f, |
80 | 0 | "extra_input: {:?} extra_input_occupied_len:{:?} output[..5]: {:?} output_occupied_len: {:?}", |
81 | 0 | self.extra_input, |
82 | 0 | self.extra_input_occupied_len, |
83 | 0 | &self.output[0..5], |
84 | 0 | self.output_occupied_len |
85 | 0 | ) |
86 | 0 | } |
87 | | } |
88 | | |
89 | | impl<W: Write> EncoderWriter<W> { |
90 | | /// Create a new encoder that will write to the provided delegate writer `w`. |
91 | 0 | pub fn new(w: W, config: Config) -> EncoderWriter<W> { |
92 | 0 | EncoderWriter { |
93 | 0 | config, |
94 | 0 | delegate: Some(w), |
95 | 0 | extra_input: [0u8; MIN_ENCODE_CHUNK_SIZE], |
96 | 0 | extra_input_occupied_len: 0, |
97 | 0 | output: [0u8; BUF_SIZE], |
98 | 0 | output_occupied_len: 0, |
99 | 0 | panicked: false, |
100 | 0 | } |
101 | 0 | } |
102 | | |
103 | | /// Encode all remaining buffered data and write it, including any trailing incomplete input |
104 | | /// triples and associated padding. |
105 | | /// |
106 | | /// Once this succeeds, no further writes or calls to this method are allowed. |
107 | | /// |
108 | | /// This may write to the delegate writer multiple times if the delegate writer does not accept |
109 | | /// all input provided to its `write` each invocation. |
110 | | /// |
111 | | /// If you don't care about error handling, it is not necessary to call this function, as the |
112 | | /// equivalent finalization is done by the Drop impl. |
113 | | /// |
114 | | /// Returns the writer that this was constructed around. |
115 | | /// |
116 | | /// # Errors |
117 | | /// |
118 | | /// The first error that is not of `ErrorKind::Interrupted` will be returned. |
119 | 0 | pub fn finish(&mut self) -> Result<W> { |
120 | 0 | // If we could consume self in finish(), we wouldn't have to worry about this case, but |
121 | 0 | // finish() is retryable in the face of I/O errors, so we can't consume here. |
122 | 0 | if self.delegate.is_none() { |
123 | 0 | panic!("Encoder has already had finish() called") |
124 | 0 | }; |
125 | 0 |
|
126 | 0 | self.write_final_leftovers()?; |
127 | | |
128 | 0 | let writer = self.delegate.take().expect("Writer must be present"); |
129 | 0 |
|
130 | 0 | Ok(writer) |
131 | 0 | } |
132 | | |
133 | | /// Write any remaining buffered data to the delegate writer. |
134 | 0 | fn write_final_leftovers(&mut self) -> Result<()> { |
135 | 0 | if self.delegate.is_none() { |
136 | | // finish() has already successfully called this, and we are now in drop() with a None |
137 | | // writer, so just no-op |
138 | 0 | return Ok(()); |
139 | 0 | } |
140 | 0 |
|
141 | 0 | self.write_all_encoded_output()?; |
142 | | |
143 | 0 | if self.extra_input_occupied_len > 0 { |
144 | 0 | let encoded_len = encode_config_slice( |
145 | 0 | &self.extra_input[..self.extra_input_occupied_len], |
146 | 0 | self.config, |
147 | 0 | &mut self.output[..], |
148 | 0 | ); |
149 | 0 |
|
150 | 0 | self.output_occupied_len = encoded_len; |
151 | 0 |
|
152 | 0 | self.write_all_encoded_output()?; |
153 | | |
154 | | // write succeeded, do not write the encoding of extra again if finish() is retried |
155 | 0 | self.extra_input_occupied_len = 0; |
156 | 0 | } |
157 | | |
158 | 0 | Ok(()) |
159 | 0 | } |
160 | | |
161 | | /// Write as much of the encoded output to the delegate writer as it will accept, and store the |
162 | | /// leftovers to be attempted at the next write() call. Updates `self.output_occupied_len`. |
163 | | /// |
164 | | /// # Errors |
165 | | /// |
166 | | /// Errors from the delegate writer are returned. In the case of an error, |
167 | | /// `self.output_occupied_len` will not be updated, as errors from `write` are specified to mean |
168 | | /// that no write took place. |
169 | 0 | fn write_to_delegate(&mut self, current_output_len: usize) -> Result<()> { |
170 | 0 | self.panicked = true; |
171 | 0 | let res = self |
172 | 0 | .delegate |
173 | 0 | .as_mut() |
174 | 0 | .expect("Writer must be present") |
175 | 0 | .write(&self.output[..current_output_len]); |
176 | 0 | self.panicked = false; |
177 | 0 |
|
178 | 0 | res.map(|consumed| { |
179 | 0 | debug_assert!(consumed <= current_output_len); |
180 | | |
181 | 0 | if consumed < current_output_len { |
182 | 0 | self.output_occupied_len = current_output_len.checked_sub(consumed).unwrap(); |
183 | 0 | // If we're blocking on I/O, the minor inefficiency of copying bytes to the |
184 | 0 | // start of the buffer is the least of our concerns... |
185 | 0 | // Rotate moves more than we need to, but copy_within isn't stabilized yet. |
186 | 0 | self.output.rotate_left(consumed); |
187 | 0 | } else { |
188 | 0 | self.output_occupied_len = 0; |
189 | 0 | } |
190 | 0 | }) |
191 | 0 | } |
192 | | |
193 | | /// Write all buffered encoded output. If this returns `Ok`, `self.output_occupied_len` is `0`. |
194 | | /// |
195 | | /// This is basically write_all for the remaining buffered data but without the undesirable |
196 | | /// abort-on-`Ok(0)` behavior. |
197 | | /// |
198 | | /// # Errors |
199 | | /// |
200 | | /// Any error emitted by the delegate writer abort the write loop and is returned, unless it's |
201 | | /// `Interrupted`, in which case the error is ignored and writes will continue. |
202 | 0 | fn write_all_encoded_output(&mut self) -> Result<()> { |
203 | 0 | while self.output_occupied_len > 0 { |
204 | 0 | let remaining_len = self.output_occupied_len; |
205 | 0 | match self.write_to_delegate(remaining_len) { |
206 | | // try again on interrupts ala write_all |
207 | 0 | Err(ref e) if e.kind() == ErrorKind::Interrupted => {} |
208 | | // other errors return |
209 | 0 | Err(e) => return Err(e), |
210 | | // success no-ops because remaining length is already updated |
211 | 0 | Ok(_) => {} |
212 | | }; |
213 | | } |
214 | | |
215 | 0 | debug_assert_eq!(0, self.output_occupied_len); |
216 | 0 | Ok(()) |
217 | 0 | } |
218 | | } |
219 | | |
220 | | impl<W: Write> Write for EncoderWriter<W> { |
221 | | /// Encode input and then write to the delegate writer. |
222 | | /// |
223 | | /// Under non-error circumstances, this returns `Ok` with the value being the number of bytes |
224 | | /// of `input` consumed. The value may be `0`, which interacts poorly with `write_all`, which |
225 | | /// interprets `Ok(0)` as an error, despite it being allowed by the contract of `write`. See |
226 | | /// <https://github.com/rust-lang/rust/issues/56889> for more on that. |
227 | | /// |
228 | | /// If the previous call to `write` provided more (encoded) data than the delegate writer could |
229 | | /// accept in a single call to its `write`, the remaining data is buffered. As long as buffered |
230 | | /// data is present, subsequent calls to `write` will try to write the remaining buffered data |
231 | | /// to the delegate and return either `Ok(0)` -- and therefore not consume any of `input` -- or |
232 | | /// an error. |
233 | | /// |
234 | | /// # Errors |
235 | | /// |
236 | | /// Any errors emitted by the delegate writer are returned. |
237 | 0 | fn write(&mut self, input: &[u8]) -> Result<usize> { |
238 | 0 | if self.delegate.is_none() { |
239 | 0 | panic!("Cannot write more after calling finish()"); |
240 | 0 | } |
241 | 0 |
|
242 | 0 | if input.is_empty() { |
243 | 0 | return Ok(0); |
244 | 0 | } |
245 | 0 |
|
246 | 0 | // The contract of `Write::write` places some constraints on this implementation: |
247 | 0 | // - a call to `write()` represents at most one call to a wrapped `Write`, so we can't |
248 | 0 | // iterate over the input and encode multiple chunks. |
249 | 0 | // - Errors mean that "no bytes were written to this writer", so we need to reset the |
250 | 0 | // internal state to what it was before the error occurred |
251 | 0 |
|
252 | 0 | // before reading any input, write any leftover encoded output from last time |
253 | 0 | if self.output_occupied_len > 0 { |
254 | 0 | let current_len = self.output_occupied_len; |
255 | 0 | return self |
256 | 0 | .write_to_delegate(current_len) |
257 | 0 | // did not read any input |
258 | 0 | .map(|_| 0); |
259 | 0 | } |
260 | 0 |
|
261 | 0 | debug_assert_eq!(0, self.output_occupied_len); |
262 | | |
263 | | // how many bytes, if any, were read into `extra` to create a triple to encode |
264 | 0 | let mut extra_input_read_len = 0; |
265 | 0 | let mut input = input; |
266 | 0 |
|
267 | 0 | let orig_extra_len = self.extra_input_occupied_len; |
268 | 0 |
|
269 | 0 | let mut encoded_size = 0; |
270 | 0 | // always a multiple of MIN_ENCODE_CHUNK_SIZE |
271 | 0 | let mut max_input_len = MAX_INPUT_LEN; |
272 | 0 |
|
273 | 0 | // process leftover un-encoded input from last write |
274 | 0 | if self.extra_input_occupied_len > 0 { |
275 | 0 | debug_assert!(self.extra_input_occupied_len < 3); |
276 | 0 | if input.len() + self.extra_input_occupied_len >= MIN_ENCODE_CHUNK_SIZE { |
277 | | // Fill up `extra`, encode that into `output`, and consume as much of the rest of |
278 | | // `input` as possible. |
279 | | // We could write just the encoding of `extra` by itself but then we'd have to |
280 | | // return after writing only 4 bytes, which is inefficient if the underlying writer |
281 | | // would make a syscall. |
282 | 0 | extra_input_read_len = MIN_ENCODE_CHUNK_SIZE - self.extra_input_occupied_len; |
283 | 0 | debug_assert!(extra_input_read_len > 0); |
284 | | // overwrite only bytes that weren't already used. If we need to rollback extra_len |
285 | | // (when the subsequent write errors), the old leading bytes will still be there. |
286 | 0 | self.extra_input[self.extra_input_occupied_len..MIN_ENCODE_CHUNK_SIZE] |
287 | 0 | .copy_from_slice(&input[0..extra_input_read_len]); |
288 | 0 |
|
289 | 0 | let len = encode_to_slice( |
290 | 0 | &self.extra_input[0..MIN_ENCODE_CHUNK_SIZE], |
291 | 0 | &mut self.output[..], |
292 | 0 | self.config.char_set.encode_table(), |
293 | 0 | ); |
294 | 0 | debug_assert_eq!(4, len); |
295 | | |
296 | 0 | input = &input[extra_input_read_len..]; |
297 | 0 |
|
298 | 0 | // consider extra to be used up, since we encoded it |
299 | 0 | self.extra_input_occupied_len = 0; |
300 | 0 | // don't clobber where we just encoded to |
301 | 0 | encoded_size = 4; |
302 | 0 | // and don't read more than can be encoded |
303 | 0 | max_input_len = MAX_INPUT_LEN - MIN_ENCODE_CHUNK_SIZE; |
304 | | |
305 | | // fall through to normal encoding |
306 | | } else { |
307 | | // `extra` and `input` are non empty, but `|extra| + |input| < 3`, so there must be |
308 | | // 1 byte in each. |
309 | 0 | debug_assert_eq!(1, input.len()); |
310 | 0 | debug_assert_eq!(1, self.extra_input_occupied_len); |
311 | | |
312 | 0 | self.extra_input[self.extra_input_occupied_len] = input[0]; |
313 | 0 | self.extra_input_occupied_len += 1; |
314 | 0 | return Ok(1); |
315 | | }; |
316 | 0 | } else if input.len() < MIN_ENCODE_CHUNK_SIZE { |
317 | | // `extra` is empty, and `input` fits inside it |
318 | 0 | self.extra_input[0..input.len()].copy_from_slice(input); |
319 | 0 | self.extra_input_occupied_len = input.len(); |
320 | 0 | return Ok(input.len()); |
321 | 0 | }; |
322 | | |
323 | | // either 0 or 1 complete chunks encoded from extra |
324 | 0 | debug_assert!(encoded_size == 0 || encoded_size == 4); |
325 | 0 | debug_assert!( |
326 | | // didn't encode extra input |
327 | 0 | MAX_INPUT_LEN == max_input_len |
328 | | // encoded one triple |
329 | 0 | || MAX_INPUT_LEN == max_input_len + MIN_ENCODE_CHUNK_SIZE |
330 | | ); |
331 | | |
332 | | // encode complete triples only |
333 | 0 | let input_complete_chunks_len = input.len() - (input.len() % MIN_ENCODE_CHUNK_SIZE); |
334 | 0 | let input_chunks_to_encode_len = cmp::min(input_complete_chunks_len, max_input_len); |
335 | 0 | debug_assert_eq!(0, max_input_len % MIN_ENCODE_CHUNK_SIZE); |
336 | 0 | debug_assert_eq!(0, input_chunks_to_encode_len % MIN_ENCODE_CHUNK_SIZE); |
337 | | |
338 | 0 | encoded_size += encode_to_slice( |
339 | 0 | &input[..(input_chunks_to_encode_len)], |
340 | 0 | &mut self.output[encoded_size..], |
341 | 0 | self.config.char_set.encode_table(), |
342 | 0 | ); |
343 | 0 |
|
344 | 0 | // not updating `self.output_occupied_len` here because if the below write fails, it should |
345 | 0 | // "never take place" -- the buffer contents we encoded are ignored and perhaps retried |
346 | 0 | // later, if the consumer chooses. |
347 | 0 |
|
348 | 0 | self.write_to_delegate(encoded_size) |
349 | 0 | // no matter whether we wrote the full encoded buffer or not, we consumed the same |
350 | 0 | // input |
351 | 0 | .map(|_| extra_input_read_len + input_chunks_to_encode_len) |
352 | 0 | .map_err(|e| { |
353 | 0 | // in case we filled and encoded `extra`, reset extra_len |
354 | 0 | self.extra_input_occupied_len = orig_extra_len; |
355 | 0 |
|
356 | 0 | e |
357 | 0 | }) |
358 | 0 | } |
359 | | |
360 | | /// Because this is usually treated as OK to call multiple times, it will *not* flush any |
361 | | /// incomplete chunks of input or write padding. |
362 | | /// # Errors |
363 | | /// |
364 | | /// The first error that is not of [`ErrorKind::Interrupted`] will be returned. |
365 | 0 | fn flush(&mut self) -> Result<()> { |
366 | 0 | self.write_all_encoded_output()?; |
367 | 0 | self.delegate |
368 | 0 | .as_mut() |
369 | 0 | .expect("Writer must be present") |
370 | 0 | .flush() |
371 | 0 | } |
372 | | } |
373 | | |
374 | | impl<W: Write> Drop for EncoderWriter<W> { |
375 | 0 | fn drop(&mut self) { |
376 | 0 | if !self.panicked { |
377 | 0 | // like `BufWriter`, ignore errors during drop |
378 | 0 | let _ = self.write_final_leftovers(); |
379 | 0 | } |
380 | 0 | } |
381 | | } |