1 | use std::convert::TryInto; |
2 | |
3 | use simd_adler32::Adler32; |
4 | |
5 | use crate::tables::{ |
6 | self, CLCL_ORDER, DIST_SYM_TO_DIST_BASE, DIST_SYM_TO_DIST_EXTRA, FDEFLATE_DIST_DECODE_TABLE, |
7 | FDEFLATE_LITLEN_DECODE_TABLE, FIXED_CODE_LENGTHS, LEN_SYM_TO_LEN_BASE, LEN_SYM_TO_LEN_EXTRA, |
8 | }; |
9 | |
10 | /// An error encountered while decompressing a deflate stream. |
11 | #[derive (Debug, PartialEq)] |
12 | pub enum DecompressionError { |
13 | /// The zlib header is corrupt. |
14 | BadZlibHeader, |
15 | /// All input was consumed, but the end of the stream hasn't been reached. |
16 | InsufficientInput, |
17 | /// A block header specifies an invalid block type. |
18 | InvalidBlockType, |
19 | /// An uncompressed block's NLEN value is invalid. |
20 | InvalidUncompressedBlockLength, |
21 | /// Too many literals were specified. |
22 | InvalidHlit, |
23 | /// Too many distance codes were specified. |
24 | InvalidHdist, |
25 | /// Attempted to repeat a previous code before reading any codes, or past the end of the code |
26 | /// lengths. |
27 | InvalidCodeLengthRepeat, |
28 | /// The stream doesn't specify a valid huffman tree. |
29 | BadCodeLengthHuffmanTree, |
30 | /// The stream doesn't specify a valid huffman tree. |
31 | BadLiteralLengthHuffmanTree, |
32 | /// The stream doesn't specify a valid huffman tree. |
33 | BadDistanceHuffmanTree, |
34 | /// The stream contains a literal/length code that was not allowed by the header. |
35 | InvalidLiteralLengthCode, |
36 | /// The stream contains a distance code that was not allowed by the header. |
37 | InvalidDistanceCode, |
38 | /// The stream contains contains back-reference as the first symbol. |
39 | InputStartsWithRun, |
40 | /// The stream contains a back-reference that is too far back. |
41 | DistanceTooFarBack, |
42 | /// The deflate stream checksum is incorrect. |
43 | WrongChecksum, |
44 | /// Extra input data. |
45 | ExtraInput, |
46 | } |
47 | |
48 | struct BlockHeader { |
49 | hlit: usize, |
50 | hdist: usize, |
51 | hclen: usize, |
52 | num_lengths_read: usize, |
53 | |
54 | /// Low 3-bits are code length code length, high 5-bits are code length code. |
55 | table: [u8; 128], |
56 | code_lengths: [u8; 320], |
57 | } |
58 | |
59 | const LITERAL_ENTRY: u32 = 0x8000; |
60 | const EXCEPTIONAL_ENTRY: u32 = 0x4000; |
61 | const SECONDARY_TABLE_ENTRY: u32 = 0x2000; |
62 | |
63 | /// The Decompressor state for a compressed block. |
64 | /// |
65 | /// The main litlen_table uses a 12-bit input to lookup the meaning of the symbol. The table is |
66 | /// split into 4 sections: |
67 | /// |
68 | /// aaaaaaaa_bbbbbbbb_1000yyyy_0000xxxx x = input_advance_bits, y = output_advance_bytes (literal) |
69 | /// 0000000z_zzzzzzzz_00000yyy_0000xxxx x = input_advance_bits, y = extra_bits, z = distance_base (length) |
70 | /// 00000000_00000000_01000000_0000xxxx x = input_advance_bits (EOF) |
71 | /// 0000xxxx_xxxxxxxx_01100000_00000000 x = secondary_table_index |
72 | /// 00000000_00000000_01000000_00000000 invalid code |
73 | /// |
74 | /// The distance table is a 512-entry table that maps 9 bits of distance symbols to their meaning. |
75 | /// |
76 | /// 00000000_00000000_00000000_00000000 symbol is more than 9 bits |
77 | /// zzzzzzzz_zzzzzzzz_0000yyyy_0000xxxx x = input_advance_bits, y = extra_bits, z = distance_base |
78 | #[repr (align(64))] |
79 | #[derive (Eq, PartialEq, Debug)] |
80 | struct CompressedBlock { |
81 | litlen_table: [u32; 4096], |
82 | dist_table: [u32; 512], |
83 | |
84 | dist_symbol_lengths: [u8; 30], |
85 | dist_symbol_masks: [u16; 30], |
86 | dist_symbol_codes: [u16; 30], |
87 | |
88 | secondary_table: Vec<u16>, |
89 | eof_code: u16, |
90 | eof_mask: u16, |
91 | eof_bits: u8, |
92 | } |
93 | |
94 | const FDEFLATE_COMPRESSED_BLOCK: CompressedBlock = CompressedBlock { |
95 | litlen_table: FDEFLATE_LITLEN_DECODE_TABLE, |
96 | dist_table: FDEFLATE_DIST_DECODE_TABLE, |
97 | dist_symbol_lengths: [ |
98 | 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, |
99 | ], |
100 | dist_symbol_masks: [ |
101 | 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, |
102 | ], |
103 | dist_symbol_codes: [ |
104 | 0, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, |
105 | 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, |
106 | 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, |
107 | ], |
108 | secondary_table: Vec::new(), |
109 | eof_code: 0x8ff, |
110 | eof_mask: 0xfff, |
111 | eof_bits: 0xc, |
112 | }; |
113 | |
114 | #[derive (Debug, Copy, Clone, Eq, PartialEq)] |
115 | enum State { |
116 | ZlibHeader, |
117 | BlockHeader, |
118 | CodeLengthCodes, |
119 | CodeLengths, |
120 | CompressedData, |
121 | UncompressedData, |
122 | Checksum, |
123 | Done, |
124 | } |
125 | |
126 | /// Decompressor for arbitrary zlib streams. |
127 | pub struct Decompressor { |
128 | /// State for decoding a compressed block. |
129 | compression: CompressedBlock, |
130 | // State for decoding a block header. |
131 | header: BlockHeader, |
132 | // Number of bytes left for uncompressed block. |
133 | uncompressed_bytes_left: u16, |
134 | |
135 | buffer: u64, |
136 | nbits: u8, |
137 | |
138 | queued_rle: Option<(u8, usize)>, |
139 | queued_backref: Option<(usize, usize)>, |
140 | last_block: bool, |
141 | |
142 | state: State, |
143 | checksum: Adler32, |
144 | ignore_adler32: bool, |
145 | } |
146 | |
147 | impl Default for Decompressor { |
148 | fn default() -> Self { |
149 | Self::new() |
150 | } |
151 | } |
152 | |
153 | impl Decompressor { |
154 | /// Create a new decompressor. |
155 | pub fn new() -> Self { |
156 | Self { |
157 | buffer: 0, |
158 | nbits: 0, |
159 | compression: CompressedBlock { |
160 | litlen_table: [0; 4096], |
161 | dist_table: [0; 512], |
162 | secondary_table: Vec::new(), |
163 | dist_symbol_lengths: [0; 30], |
164 | dist_symbol_masks: [0; 30], |
165 | dist_symbol_codes: [0xffff; 30], |
166 | eof_code: 0, |
167 | eof_mask: 0, |
168 | eof_bits: 0, |
169 | }, |
170 | header: BlockHeader { |
171 | hlit: 0, |
172 | hdist: 0, |
173 | hclen: 0, |
174 | table: [0; 128], |
175 | num_lengths_read: 0, |
176 | code_lengths: [0; 320], |
177 | }, |
178 | uncompressed_bytes_left: 0, |
179 | queued_rle: None, |
180 | queued_backref: None, |
181 | checksum: Adler32::new(), |
182 | state: State::ZlibHeader, |
183 | last_block: false, |
184 | ignore_adler32: false, |
185 | } |
186 | } |
187 | |
188 | /// Ignore the checksum at the end of the stream. |
189 | pub fn ignore_adler32(&mut self) { |
190 | self.ignore_adler32 = true; |
191 | } |
192 | |
193 | fn fill_buffer(&mut self, input: &mut &[u8]) { |
194 | if input.len() >= 8 { |
195 | self.buffer |= u64::from_le_bytes(input[..8].try_into().unwrap()) << self.nbits; |
196 | *input = &mut &input[(63 - self.nbits as usize) / 8..]; |
197 | self.nbits |= 56; |
198 | } else { |
199 | let nbytes = input.len().min((63 - self.nbits as usize) / 8); |
200 | let mut input_data = [0; 8]; |
201 | input_data[..nbytes].copy_from_slice(&input[..nbytes]); |
202 | self.buffer |= u64::from_le_bytes(input_data) |
203 | .checked_shl(self.nbits as u32) |
204 | .unwrap_or(0); |
205 | self.nbits += nbytes as u8 * 8; |
206 | *input = &mut &input[nbytes..]; |
207 | } |
208 | } |
209 | |
210 | fn peak_bits(&mut self, nbits: u8) -> u64 { |
211 | debug_assert!(nbits <= 56 && nbits <= self.nbits); |
212 | self.buffer & ((1u64 << nbits) - 1) |
213 | } |
214 | fn consume_bits(&mut self, nbits: u8) { |
215 | debug_assert!(self.nbits >= nbits); |
216 | self.buffer >>= nbits; |
217 | self.nbits -= nbits; |
218 | } |
219 | |
220 | fn read_block_header(&mut self, remaining_input: &mut &[u8]) -> Result<(), DecompressionError> { |
221 | self.fill_buffer(remaining_input); |
222 | if self.nbits < 3 { |
223 | return Ok(()); |
224 | } |
225 | |
226 | let start = self.peak_bits(3); |
227 | self.last_block = start & 1 != 0; |
228 | match start >> 1 { |
229 | 0b00 => { |
230 | let align_bits = (self.nbits - 3) % 8; |
231 | let header_bits = 3 + 32 + align_bits; |
232 | if self.nbits < header_bits { |
233 | return Ok(()); |
234 | } |
235 | |
236 | let len = (self.peak_bits(align_bits + 19) >> (align_bits + 3)) as u16; |
237 | let nlen = (self.peak_bits(header_bits) >> (align_bits + 19)) as u16; |
238 | if nlen != !len { |
239 | return Err(DecompressionError::InvalidUncompressedBlockLength); |
240 | } |
241 | |
242 | self.state = State::UncompressedData; |
243 | self.uncompressed_bytes_left = len; |
244 | self.consume_bits(header_bits); |
245 | Ok(()) |
246 | } |
247 | 0b01 => { |
248 | self.consume_bits(3); |
249 | // TODO: Do this statically rather than every time. |
250 | Self::build_tables(288, &FIXED_CODE_LENGTHS, &mut self.compression, 6)?; |
251 | self.state = State::CompressedData; |
252 | Ok(()) |
253 | } |
254 | 0b10 => { |
255 | if self.nbits < 17 { |
256 | return Ok(()); |
257 | } |
258 | |
259 | self.header.hlit = (self.peak_bits(8) >> 3) as usize + 257; |
260 | self.header.hdist = (self.peak_bits(13) >> 8) as usize + 1; |
261 | self.header.hclen = (self.peak_bits(17) >> 13) as usize + 4; |
262 | if self.header.hlit > 286 { |
263 | return Err(DecompressionError::InvalidHlit); |
264 | } |
265 | if self.header.hdist > 30 { |
266 | return Err(DecompressionError::InvalidHdist); |
267 | } |
268 | |
269 | self.consume_bits(17); |
270 | self.state = State::CodeLengthCodes; |
271 | Ok(()) |
272 | } |
273 | 0b11 => Err(DecompressionError::InvalidBlockType), |
274 | _ => unreachable!(), |
275 | } |
276 | } |
277 | |
278 | fn read_code_length_codes( |
279 | &mut self, |
280 | remaining_input: &mut &[u8], |
281 | ) -> Result<(), DecompressionError> { |
282 | self.fill_buffer(remaining_input); |
283 | if self.nbits as usize + remaining_input.len() * 8 < 3 * self.header.hclen { |
284 | return Ok(()); |
285 | } |
286 | |
287 | let mut code_length_lengths = [0; 19]; |
288 | for i in 0..self.header.hclen { |
289 | code_length_lengths[CLCL_ORDER[i]] = self.peak_bits(3) as u8; |
290 | self.consume_bits(3); |
291 | |
292 | // We need to refill the buffer after reading 3 * 18 = 54 bits since the buffer holds |
293 | // between 56 and 63 bits total. |
294 | if i == 17 { |
295 | self.fill_buffer(remaining_input); |
296 | } |
297 | } |
298 | let code_length_codes: [u16; 19] = crate::compute_codes(&code_length_lengths) |
299 | .ok_or(DecompressionError::BadCodeLengthHuffmanTree)?; |
300 | |
301 | self.header.table = [255; 128]; |
302 | for i in 0..19 { |
303 | let length = code_length_lengths[i]; |
304 | if length > 0 { |
305 | let mut j = code_length_codes[i]; |
306 | while j < 128 { |
307 | self.header.table[j as usize] = ((i as u8) << 3) | length; |
308 | j += 1 << length; |
309 | } |
310 | } |
311 | } |
312 | |
313 | self.state = State::CodeLengths; |
314 | self.header.num_lengths_read = 0; |
315 | Ok(()) |
316 | } |
317 | |
318 | fn read_code_lengths(&mut self, remaining_input: &mut &[u8]) -> Result<(), DecompressionError> { |
319 | let total_lengths = self.header.hlit + self.header.hdist; |
320 | while self.header.num_lengths_read < total_lengths { |
321 | self.fill_buffer(remaining_input); |
322 | if self.nbits < 7 { |
323 | return Ok(()); |
324 | } |
325 | |
326 | let code = self.peak_bits(7); |
327 | let entry = self.header.table[code as usize]; |
328 | let length = entry & 0x7; |
329 | let symbol = entry >> 3; |
330 | |
331 | debug_assert!(length != 0); |
332 | match symbol { |
333 | 0..=15 => { |
334 | self.header.code_lengths[self.header.num_lengths_read] = symbol; |
335 | self.header.num_lengths_read += 1; |
336 | self.consume_bits(length); |
337 | } |
338 | 16..=18 => { |
339 | let (base_repeat, extra_bits) = match symbol { |
340 | 16 => (3, 2), |
341 | 17 => (3, 3), |
342 | 18 => (11, 7), |
343 | _ => unreachable!(), |
344 | }; |
345 | |
346 | if self.nbits < length + extra_bits { |
347 | return Ok(()); |
348 | } |
349 | |
350 | let value = match symbol { |
351 | 16 => { |
352 | self.header.code_lengths[self |
353 | .header |
354 | .num_lengths_read |
355 | .checked_sub(1) |
356 | .ok_or(DecompressionError::InvalidCodeLengthRepeat)?] |
357 | // TODO: is this right? |
358 | } |
359 | 17 => 0, |
360 | 18 => 0, |
361 | _ => unreachable!(), |
362 | }; |
363 | |
364 | let repeat = |
365 | (self.peak_bits(length + extra_bits) >> length) as usize + base_repeat; |
366 | if self.header.num_lengths_read + repeat > total_lengths { |
367 | return Err(DecompressionError::InvalidCodeLengthRepeat); |
368 | } |
369 | |
370 | for i in 0..repeat { |
371 | self.header.code_lengths[self.header.num_lengths_read + i] = value; |
372 | } |
373 | self.header.num_lengths_read += repeat; |
374 | self.consume_bits(length + extra_bits); |
375 | } |
376 | _ => unreachable!(), |
377 | } |
378 | } |
379 | |
380 | self.header |
381 | .code_lengths |
382 | .copy_within(self.header.hlit..total_lengths, 288); |
383 | for i in self.header.hlit..288 { |
384 | self.header.code_lengths[i] = 0; |
385 | } |
386 | for i in 288 + self.header.hdist..320 { |
387 | self.header.code_lengths[i] = 0; |
388 | } |
389 | |
390 | if self.header.hdist == 1 |
391 | && self.header.code_lengths[..286] == tables::HUFFMAN_LENGTHS |
392 | && self.header.code_lengths[288] == 1 |
393 | { |
394 | self.compression = FDEFLATE_COMPRESSED_BLOCK; |
395 | } else { |
396 | Self::build_tables( |
397 | self.header.hlit, |
398 | &self.header.code_lengths, |
399 | &mut self.compression, |
400 | 6, |
401 | )?; |
402 | } |
403 | self.state = State::CompressedData; |
404 | Ok(()) |
405 | } |
406 | |
407 | fn build_tables( |
408 | hlit: usize, |
409 | code_lengths: &[u8], |
410 | compression: &mut CompressedBlock, |
411 | max_search_bits: u8, |
412 | ) -> Result<(), DecompressionError> { |
413 | // Build the literal/length code table. |
414 | let lengths = &code_lengths[..288]; |
415 | let codes: [u16; 288] = crate::compute_codes(&lengths.try_into().unwrap()) |
416 | .ok_or(DecompressionError::BadLiteralLengthHuffmanTree)?; |
417 | |
418 | let table_bits = lengths.iter().cloned().max().unwrap().min(12).max(6); |
419 | let table_size = 1 << table_bits; |
420 | |
421 | for i in 0..256 { |
422 | let code = codes[i]; |
423 | let length = lengths[i]; |
424 | let mut j = code; |
425 | |
426 | while j < table_size && length != 0 && length <= 12 { |
427 | compression.litlen_table[j as usize] = |
428 | ((i as u32) << 16) | LITERAL_ENTRY | (1 << 8) | length as u32; |
429 | j += 1 << length; |
430 | } |
431 | |
432 | if length > 0 && length <= max_search_bits { |
433 | for ii in 0..256 { |
434 | let code2 = codes[ii]; |
435 | let length2 = lengths[ii]; |
436 | if length2 != 0 && length + length2 <= table_bits { |
437 | let mut j = code | (code2 << length); |
438 | |
439 | while j < table_size { |
440 | compression.litlen_table[j as usize] = (ii as u32) << 24 |
441 | | (i as u32) << 16 |
442 | | LITERAL_ENTRY |
443 | | (2 << 8) |
444 | | ((length + length2) as u32); |
445 | j += 1 << (length + length2); |
446 | } |
447 | } |
448 | } |
449 | } |
450 | } |
451 | |
452 | if lengths[256] != 0 && lengths[256] <= 12 { |
453 | let mut j = codes[256]; |
454 | while j < table_size { |
455 | compression.litlen_table[j as usize] = EXCEPTIONAL_ENTRY | lengths[256] as u32; |
456 | j += 1 << lengths[256]; |
457 | } |
458 | } |
459 | |
460 | let table_size = table_size as usize; |
461 | for i in (table_size..4096).step_by(table_size) { |
462 | compression.litlen_table.copy_within(0..table_size, i); |
463 | } |
464 | |
465 | compression.eof_code = codes[256]; |
466 | compression.eof_mask = (1 << lengths[256]) - 1; |
467 | compression.eof_bits = lengths[256]; |
468 | |
469 | for i in 257..hlit { |
470 | let code = codes[i]; |
471 | let length = lengths[i]; |
472 | if length != 0 && length <= 12 { |
473 | let mut j = code; |
474 | while j < 4096 { |
475 | compression.litlen_table[j as usize] = if i < 286 { |
476 | (LEN_SYM_TO_LEN_BASE[i - 257] as u32) << 16 |
477 | | (LEN_SYM_TO_LEN_EXTRA[i - 257] as u32) << 8 |
478 | | length as u32 |
479 | } else { |
480 | EXCEPTIONAL_ENTRY |
481 | }; |
482 | j += 1 << length; |
483 | } |
484 | } |
485 | } |
486 | |
487 | for i in 0..hlit { |
488 | if lengths[i] > 12 { |
489 | compression.litlen_table[(codes[i] & 0xfff) as usize] = u32::MAX; |
490 | } |
491 | } |
492 | |
493 | let mut secondary_table_len = 0; |
494 | for i in 0..hlit { |
495 | if lengths[i] > 12 { |
496 | let j = (codes[i] & 0xfff) as usize; |
497 | if compression.litlen_table[j] == u32::MAX { |
498 | compression.litlen_table[j] = |
499 | (secondary_table_len << 16) | EXCEPTIONAL_ENTRY | SECONDARY_TABLE_ENTRY; |
500 | secondary_table_len += 8; |
501 | } |
502 | } |
503 | } |
504 | assert!(secondary_table_len <= 0x7ff); |
505 | compression.secondary_table = vec![0; secondary_table_len as usize]; |
506 | for i in 0..hlit { |
507 | let code = codes[i]; |
508 | let length = lengths[i]; |
509 | if length > 12 { |
510 | let j = (codes[i] & 0xfff) as usize; |
511 | let k = (compression.litlen_table[j] >> 16) as usize; |
512 | |
513 | let mut s = code >> 12; |
514 | while s < 8 { |
515 | debug_assert_eq!(compression.secondary_table[k + s as usize], 0); |
516 | compression.secondary_table[k + s as usize] = |
517 | ((i as u16) << 4) | (length as u16); |
518 | s += 1 << (length - 12); |
519 | } |
520 | } |
521 | } |
522 | debug_assert!(compression |
523 | .secondary_table |
524 | .iter() |
525 | .all(|&x| x != 0 && (x & 0xf) > 12)); |
526 | |
527 | // Build the distance code table. |
528 | let lengths = &code_lengths[288..320]; |
529 | if lengths == [0; 32] { |
530 | compression.dist_symbol_masks = [0; 30]; |
531 | compression.dist_symbol_codes = [0xffff; 30]; |
532 | compression.dist_table.fill(0); |
533 | } else { |
534 | let codes: [u16; 32] = match crate::compute_codes(&lengths.try_into().unwrap()) { |
535 | Some(codes) => codes, |
536 | None => { |
537 | if lengths.iter().filter(|&&l| l != 0).count() != 1 { |
538 | return Err(DecompressionError::BadDistanceHuffmanTree); |
539 | } |
540 | [0; 32] |
541 | } |
542 | }; |
543 | |
544 | compression.dist_symbol_codes.copy_from_slice(&codes[..30]); |
545 | compression |
546 | .dist_symbol_lengths |
547 | .copy_from_slice(&lengths[..30]); |
548 | compression.dist_table.fill(0); |
549 | for i in 0..30 { |
550 | let length = lengths[i]; |
551 | let code = codes[i]; |
552 | if length == 0 { |
553 | compression.dist_symbol_masks[i] = 0; |
554 | compression.dist_symbol_codes[i] = 0xffff; |
555 | } else { |
556 | compression.dist_symbol_masks[i] = (1 << lengths[i]) - 1; |
557 | if lengths[i] <= 9 { |
558 | let mut j = code; |
559 | while j < 512 { |
560 | compression.dist_table[j as usize] = (DIST_SYM_TO_DIST_BASE[i] as u32) |
561 | << 16 |
562 | | (DIST_SYM_TO_DIST_EXTRA[i] as u32) << 8 |
563 | | length as u32; |
564 | j += 1 << lengths[i]; |
565 | } |
566 | } |
567 | } |
568 | } |
569 | } |
570 | |
571 | Ok(()) |
572 | } |
573 | |
574 | fn read_compressed( |
575 | &mut self, |
576 | remaining_input: &mut &[u8], |
577 | output: &mut [u8], |
578 | mut output_index: usize, |
579 | ) -> Result<usize, DecompressionError> { |
580 | while let State::CompressedData = self.state { |
581 | self.fill_buffer(remaining_input); |
582 | if output_index == output.len() { |
583 | break; |
584 | } |
585 | |
586 | let mut bits = self.buffer; |
587 | let litlen_entry = self.compression.litlen_table[(bits & 0xfff) as usize]; |
588 | let litlen_code_bits = litlen_entry as u8; |
589 | |
590 | if litlen_entry & LITERAL_ENTRY != 0 { |
591 | // Ultra-fast path: do 3 more consecutive table lookups and bail if any of them need the slow path. |
592 | if self.nbits >= 48 { |
593 | let litlen_entry2 = |
594 | self.compression.litlen_table[(bits >> litlen_code_bits & 0xfff) as usize]; |
595 | let litlen_code_bits2 = litlen_entry2 as u8; |
596 | let litlen_entry3 = self.compression.litlen_table |
597 | [(bits >> (litlen_code_bits + litlen_code_bits2) & 0xfff) as usize]; |
598 | let litlen_code_bits3 = litlen_entry3 as u8; |
599 | let litlen_entry4 = self.compression.litlen_table[(bits |
600 | >> (litlen_code_bits + litlen_code_bits2 + litlen_code_bits3) |
601 | & 0xfff) |
602 | as usize]; |
603 | let litlen_code_bits4 = litlen_entry4 as u8; |
604 | if litlen_entry2 & litlen_entry3 & litlen_entry4 & LITERAL_ENTRY != 0 { |
605 | let advance_output_bytes = ((litlen_entry & 0xf00) >> 8) as usize; |
606 | let advance_output_bytes2 = ((litlen_entry2 & 0xf00) >> 8) as usize; |
607 | let advance_output_bytes3 = ((litlen_entry3 & 0xf00) >> 8) as usize; |
608 | let advance_output_bytes4 = ((litlen_entry4 & 0xf00) >> 8) as usize; |
609 | if output_index |
610 | + advance_output_bytes |
611 | + advance_output_bytes2 |
612 | + advance_output_bytes3 |
613 | + advance_output_bytes4 |
614 | < output.len() |
615 | { |
616 | self.consume_bits( |
617 | litlen_code_bits |
618 | + litlen_code_bits2 |
619 | + litlen_code_bits3 |
620 | + litlen_code_bits4, |
621 | ); |
622 | |
623 | output[output_index] = (litlen_entry >> 16) as u8; |
624 | output[output_index + 1] = (litlen_entry >> 24) as u8; |
625 | output_index += advance_output_bytes; |
626 | output[output_index] = (litlen_entry2 >> 16) as u8; |
627 | output[output_index + 1] = (litlen_entry2 >> 24) as u8; |
628 | output_index += advance_output_bytes2; |
629 | output[output_index] = (litlen_entry3 >> 16) as u8; |
630 | output[output_index + 1] = (litlen_entry3 >> 24) as u8; |
631 | output_index += advance_output_bytes3; |
632 | output[output_index] = (litlen_entry4 >> 16) as u8; |
633 | output[output_index + 1] = (litlen_entry4 >> 24) as u8; |
634 | output_index += advance_output_bytes4; |
635 | continue; |
636 | } |
637 | } |
638 | } |
639 | |
640 | // Fast path: the next symbol is <= 12 bits and a literal, the table specifies the |
641 | // output bytes and we can directly write them to the output buffer. |
642 | let advance_output_bytes = ((litlen_entry & 0xf00) >> 8) as usize; |
643 | |
644 | // match advance_output_bytes { |
645 | // 1 => println!("[{output_index}] LIT1 {}", litlen_entry >> 16), |
646 | // 2 => println!( |
647 | // "[{output_index}] LIT2 {} {} {}", |
648 | // (litlen_entry >> 16) as u8, |
649 | // litlen_entry >> 24, |
650 | // bits & 0xfff |
651 | // ), |
652 | // n => println!( |
653 | // "[{output_index}] LIT{n} {} {}", |
654 | // (litlen_entry >> 16) as u8, |
655 | // litlen_entry >> 24, |
656 | // ), |
657 | // } |
658 | |
659 | if self.nbits < litlen_code_bits { |
660 | break; |
661 | } else if output_index + 1 < output.len() { |
662 | output[output_index] = (litlen_entry >> 16) as u8; |
663 | output[output_index + 1] = (litlen_entry >> 24) as u8; |
664 | output_index += advance_output_bytes; |
665 | self.consume_bits(litlen_code_bits); |
666 | continue; |
667 | } else if output_index + advance_output_bytes == output.len() { |
668 | debug_assert_eq!(advance_output_bytes, 1); |
669 | output[output_index] = (litlen_entry >> 16) as u8; |
670 | output_index += 1; |
671 | self.consume_bits(litlen_code_bits); |
672 | break; |
673 | } else { |
674 | debug_assert_eq!(advance_output_bytes, 2); |
675 | output[output_index] = (litlen_entry >> 16) as u8; |
676 | self.queued_rle = Some(((litlen_entry >> 24) as u8, 1)); |
677 | output_index += 1; |
678 | self.consume_bits(litlen_code_bits); |
679 | break; |
680 | } |
681 | } |
682 | |
683 | let (length_base, length_extra_bits, litlen_code_bits) = |
684 | if litlen_entry & EXCEPTIONAL_ENTRY == 0 { |
685 | ( |
686 | litlen_entry >> 16, |
687 | (litlen_entry >> 8) as u8, |
688 | litlen_code_bits, |
689 | ) |
690 | } else if litlen_entry & SECONDARY_TABLE_ENTRY != 0 { |
691 | let secondary_index = litlen_entry >> 16; |
692 | let secondary_entry = self.compression.secondary_table |
693 | [secondary_index as usize + ((bits >> 12) & 0x7) as usize]; |
694 | let litlen_symbol = secondary_entry >> 4; |
695 | let litlen_code_bits = (secondary_entry & 0xf) as u8; |
696 | |
697 | if self.nbits < litlen_code_bits { |
698 | break; |
699 | } else if litlen_symbol < 256 { |
700 | // println!("[{output_index}] LIT1b {} (val={:04x})", litlen_symbol, self.peak_bits(15)); |
701 | |
702 | self.consume_bits(litlen_code_bits); |
703 | output[output_index] = litlen_symbol as u8; |
704 | output_index += 1; |
705 | continue; |
706 | } else if litlen_symbol == 256 { |
707 | // println!("[{output_index}] EOF"); |
708 | self.consume_bits(litlen_code_bits); |
709 | self.state = match self.last_block { |
710 | true => State::Checksum, |
711 | false => State::BlockHeader, |
712 | }; |
713 | break; |
714 | } |
715 | |
716 | ( |
717 | LEN_SYM_TO_LEN_BASE[litlen_symbol as usize - 257] as u32, |
718 | LEN_SYM_TO_LEN_EXTRA[litlen_symbol as usize - 257], |
719 | litlen_code_bits, |
720 | ) |
721 | } else if litlen_code_bits == 0 { |
722 | return Err(DecompressionError::InvalidLiteralLengthCode); |
723 | } else { |
724 | if self.nbits < litlen_code_bits { |
725 | break; |
726 | } |
727 | // println!("[{output_index}] EOF"); |
728 | self.consume_bits(litlen_code_bits); |
729 | self.state = match self.last_block { |
730 | true => State::Checksum, |
731 | false => State::BlockHeader, |
732 | }; |
733 | break; |
734 | }; |
735 | bits >>= litlen_code_bits; |
736 | |
737 | let length_extra_mask = (1 << length_extra_bits) - 1; |
738 | let length = length_base as usize + (bits & length_extra_mask) as usize; |
739 | bits >>= length_extra_bits; |
740 | |
741 | let dist_entry = self.compression.dist_table[(bits & 0x1ff) as usize]; |
742 | let (dist_base, dist_extra_bits, dist_code_bits) = if dist_entry != 0 { |
743 | ( |
744 | (dist_entry >> 16) as u16, |
745 | (dist_entry >> 8) as u8, |
746 | dist_entry as u8, |
747 | ) |
748 | } else { |
749 | let mut dist_extra_bits = 0; |
750 | let mut dist_base = 0; |
751 | let mut dist_advance_bits = 0; |
752 | for i in 0..self.compression.dist_symbol_lengths.len() { |
753 | if bits as u16 & self.compression.dist_symbol_masks[i] |
754 | == self.compression.dist_symbol_codes[i] |
755 | { |
756 | dist_extra_bits = DIST_SYM_TO_DIST_EXTRA[i]; |
757 | dist_base = DIST_SYM_TO_DIST_BASE[i]; |
758 | dist_advance_bits = self.compression.dist_symbol_lengths[i]; |
759 | break; |
760 | } |
761 | } |
762 | if dist_advance_bits == 0 { |
763 | return Err(DecompressionError::InvalidDistanceCode); |
764 | } |
765 | (dist_base, dist_extra_bits, dist_advance_bits) |
766 | }; |
767 | bits >>= dist_code_bits; |
768 | |
769 | let dist = dist_base as usize + (bits & ((1 << dist_extra_bits) - 1)) as usize; |
770 | let total_bits = |
771 | litlen_code_bits + length_extra_bits + dist_code_bits + dist_extra_bits; |
772 | |
773 | if self.nbits < total_bits { |
774 | break; |
775 | } else if dist > output_index { |
776 | return Err(DecompressionError::DistanceTooFarBack); |
777 | } |
778 | |
779 | // println!("[{output_index}] BACKREF len={} dist={} {:x}", length, dist, dist_entry); |
780 | self.consume_bits(total_bits); |
781 | |
782 | let copy_length = length.min(output.len() - output_index); |
783 | if dist == 1 { |
784 | let last = output[output_index - 1]; |
785 | output[output_index..][..copy_length].fill(last); |
786 | |
787 | if copy_length < length { |
788 | self.queued_rle = Some((last, length - copy_length)); |
789 | output_index = output.len(); |
790 | break; |
791 | } |
792 | } else if output_index + length + 15 <= output.len() { |
793 | let start = output_index - dist; |
794 | output.copy_within(start..start + 16, output_index); |
795 | |
796 | if length > 16 || dist < 16 { |
797 | for i in (0..length).step_by(dist.min(16)).skip(1) { |
798 | output.copy_within(start + i..start + i + 16, output_index + i); |
799 | } |
800 | } |
801 | } else { |
802 | if dist < copy_length { |
803 | for i in 0..copy_length { |
804 | output[output_index + i] = output[output_index + i - dist]; |
805 | } |
806 | } else { |
807 | output.copy_within( |
808 | output_index - dist..output_index + copy_length - dist, |
809 | output_index, |
810 | ) |
811 | } |
812 | |
813 | if copy_length < length { |
814 | self.queued_backref = Some((dist, length - copy_length)); |
815 | output_index = output.len(); |
816 | break; |
817 | } |
818 | } |
819 | output_index += copy_length; |
820 | } |
821 | |
822 | if self.state == State::CompressedData |
823 | && self.queued_backref.is_none() |
824 | && self.queued_rle.is_none() |
825 | && self.nbits >= 15 |
826 | && self.peak_bits(15) as u16 & self.compression.eof_mask == self.compression.eof_code |
827 | { |
828 | self.consume_bits(self.compression.eof_bits); |
829 | self.state = match self.last_block { |
830 | true => State::Checksum, |
831 | false => State::BlockHeader, |
832 | }; |
833 | } |
834 | |
835 | Ok(output_index) |
836 | } |
837 | |
838 | /// Decompresses a chunk of data. |
839 | /// |
840 | /// Returns the number of bytes read from `input` and the number of bytes written to `output`, |
841 | /// or an error if the deflate stream is not valid. `input` is the compressed data. `output` is |
842 | /// the buffer to write the decompressed data to, starting at index `output_position`. |
843 | /// `end_of_input` indicates whether more data may be available in the future. |
844 | /// |
845 | /// The contents of `output` after `output_position` are ignored. However, this function may |
846 | /// write additional data to `output` past what is indicated by the return value. |
847 | /// |
848 | /// When this function returns `Ok`, at least one of the following is true: |
849 | /// - The input is fully consumed. |
850 | /// - The output is full but there are more bytes to output. |
851 | /// - The deflate stream is complete (and `is_done` will return true). |
852 | /// |
853 | /// # Panics |
854 | /// |
855 | /// This function will panic if `output_position` is out of bounds. |
856 | pub fn read( |
857 | &mut self, |
858 | input: &[u8], |
859 | output: &mut [u8], |
860 | output_position: usize, |
861 | end_of_input: bool, |
862 | ) -> Result<(usize, usize), DecompressionError> { |
863 | if let State::Done = self.state { |
864 | return Ok((0, 0)); |
865 | } |
866 | |
867 | assert!(output_position <= output.len()); |
868 | |
869 | let mut remaining_input = input; |
870 | let mut output_index = output_position; |
871 | |
872 | if let Some((data, len)) = self.queued_rle.take() { |
873 | let n = len.min(output.len() - output_index); |
874 | output[output_index..][..n].fill(data); |
875 | output_index += n; |
876 | if n < len { |
877 | self.queued_rle = Some((data, len - n)); |
878 | return Ok((0, n)); |
879 | } |
880 | } |
881 | if let Some((dist, len)) = self.queued_backref.take() { |
882 | let n = len.min(output.len() - output_index); |
883 | for i in 0..n { |
884 | output[output_index + i] = output[output_index + i - dist]; |
885 | } |
886 | output_index += n; |
887 | if n < len { |
888 | self.queued_backref = Some((dist, len - n)); |
889 | return Ok((0, n)); |
890 | } |
891 | } |
892 | |
893 | // Main decoding state machine. |
894 | let mut last_state = None; |
895 | while last_state != Some(self.state) { |
896 | last_state = Some(self.state); |
897 | match self.state { |
898 | State::ZlibHeader => { |
899 | self.fill_buffer(&mut remaining_input); |
900 | if self.nbits < 16 { |
901 | break; |
902 | } |
903 | |
904 | let input0 = self.peak_bits(8); |
905 | let input1 = self.peak_bits(16) >> 8 & 0xff; |
906 | if input0 & 0x0f != 0x08 |
907 | || (input0 & 0xf0) > 0x70 |
908 | || input1 & 0x20 != 0 |
909 | || (input0 << 8 | input1) % 31 != 0 |
910 | { |
911 | return Err(DecompressionError::BadZlibHeader); |
912 | } |
913 | |
914 | self.consume_bits(16); |
915 | self.state = State::BlockHeader; |
916 | } |
917 | State::BlockHeader => { |
918 | self.read_block_header(&mut remaining_input)?; |
919 | } |
920 | State::CodeLengthCodes => { |
921 | self.read_code_length_codes(&mut remaining_input)?; |
922 | } |
923 | State::CodeLengths => { |
924 | self.read_code_lengths(&mut remaining_input)?; |
925 | } |
926 | State::CompressedData => { |
927 | output_index = |
928 | self.read_compressed(&mut remaining_input, output, output_index)? |
929 | } |
930 | State::UncompressedData => { |
931 | // Drain any bytes from our buffer. |
932 | debug_assert_eq!(self.nbits % 8, 0); |
933 | while self.nbits > 0 |
934 | && self.uncompressed_bytes_left > 0 |
935 | && output_index < output.len() |
936 | { |
937 | output[output_index] = self.peak_bits(8) as u8; |
938 | self.consume_bits(8); |
939 | output_index += 1; |
940 | self.uncompressed_bytes_left -= 1; |
941 | } |
942 | // Buffer may contain one additional byte. Clear it to avoid confusion. |
943 | if self.nbits == 0 { |
944 | self.buffer = 0; |
945 | } |
946 | |
947 | // Copy subsequent bytes directly from the input. |
948 | let copy_bytes = (self.uncompressed_bytes_left as usize) |
949 | .min(remaining_input.len()) |
950 | .min(output.len() - output_index); |
951 | output[output_index..][..copy_bytes] |
952 | .copy_from_slice(&remaining_input[..copy_bytes]); |
953 | remaining_input = &remaining_input[copy_bytes..]; |
954 | output_index += copy_bytes; |
955 | self.uncompressed_bytes_left -= copy_bytes as u16; |
956 | |
957 | if self.uncompressed_bytes_left == 0 { |
958 | self.state = if self.last_block { |
959 | State::Checksum |
960 | } else { |
961 | State::BlockHeader |
962 | }; |
963 | } |
964 | } |
965 | State::Checksum => { |
966 | self.fill_buffer(&mut remaining_input); |
967 | |
968 | let align_bits = self.nbits % 8; |
969 | if self.nbits >= 32 + align_bits { |
970 | self.checksum.write(&output[output_position..output_index]); |
971 | if align_bits != 0 { |
972 | self.consume_bits(align_bits); |
973 | } |
974 | #[cfg (not(fuzzing))] |
975 | if !self.ignore_adler32 |
976 | && (self.peak_bits(32) as u32).swap_bytes() != self.checksum.finish() |
977 | { |
978 | return Err(DecompressionError::WrongChecksum); |
979 | } |
980 | self.state = State::Done; |
981 | self.consume_bits(32); |
982 | break; |
983 | } |
984 | } |
985 | State::Done => unreachable!(), |
986 | } |
987 | } |
988 | |
989 | if !self.ignore_adler32 && self.state != State::Done { |
990 | self.checksum.write(&output[output_position..output_index]); |
991 | } |
992 | |
993 | if self.state == State::Done || !end_of_input || output_index == output.len() { |
994 | let input_left = remaining_input.len(); |
995 | Ok((input.len() - input_left, output_index - output_position)) |
996 | } else { |
997 | Err(DecompressionError::InsufficientInput) |
998 | } |
999 | } |
1000 | |
1001 | /// Returns true if the decompressor has finished decompressing the input. |
1002 | pub fn is_done(&self) -> bool { |
1003 | self.state == State::Done |
1004 | } |
1005 | } |
1006 | |
1007 | /// Decompress the given data. |
1008 | pub fn decompress_to_vec(input: &[u8]) -> Result<Vec<u8>, DecompressionError> { |
1009 | match decompress_to_vec_bounded(input, maxlen:usize::MAX) { |
1010 | Ok(output: Vec) => Ok(output), |
1011 | Err(BoundedDecompressionError::DecompressionError { inner: DecompressionError }) => Err(inner), |
1012 | Err(BoundedDecompressionError::OutputTooLarge { .. }) => { |
1013 | unreachable!("Impossible to allocate more than isize::MAX bytes" ) |
1014 | } |
1015 | } |
1016 | } |
1017 | |
1018 | /// An error encountered while decompressing a deflate stream given a bounded maximum output. |
1019 | pub enum BoundedDecompressionError { |
1020 | /// The input is not a valid deflate stream. |
1021 | DecompressionError { |
1022 | /// The underlying error. |
1023 | inner: DecompressionError, |
1024 | }, |
1025 | |
1026 | /// The output is too large. |
1027 | OutputTooLarge { |
1028 | /// The output decoded so far. |
1029 | partial_output: Vec<u8>, |
1030 | }, |
1031 | } |
1032 | impl From<DecompressionError> for BoundedDecompressionError { |
1033 | fn from(inner: DecompressionError) -> Self { |
1034 | BoundedDecompressionError::DecompressionError { inner } |
1035 | } |
1036 | } |
1037 | |
1038 | /// Decompress the given data, returning an error if the output is larger than |
1039 | /// `maxlen` bytes. |
1040 | pub fn decompress_to_vec_bounded( |
1041 | input: &[u8], |
1042 | maxlen: usize, |
1043 | ) -> Result<Vec<u8>, BoundedDecompressionError> { |
1044 | let mut decoder = Decompressor::new(); |
1045 | let mut output = vec![0; 1024.min(maxlen)]; |
1046 | let mut input_index = 0; |
1047 | let mut output_index = 0; |
1048 | loop { |
1049 | let (consumed, produced) = |
1050 | decoder.read(&input[input_index..], &mut output, output_index, true)?; |
1051 | input_index += consumed; |
1052 | output_index += produced; |
1053 | if decoder.is_done() || output_index == maxlen { |
1054 | break; |
1055 | } |
1056 | output.resize((output_index + 32 * 1024).min(maxlen), 0); |
1057 | } |
1058 | output.resize(output_index, 0); |
1059 | |
1060 | if decoder.is_done() { |
1061 | Ok(output) |
1062 | } else { |
1063 | Err(BoundedDecompressionError::OutputTooLarge { |
1064 | partial_output: output, |
1065 | }) |
1066 | } |
1067 | } |
1068 | |
1069 | #[cfg (test)] |
1070 | mod tests { |
1071 | use crate::tables::{self, LENGTH_TO_LEN_EXTRA, LENGTH_TO_SYMBOL}; |
1072 | |
1073 | use super::*; |
1074 | use rand::Rng; |
1075 | |
1076 | fn roundtrip(data: &[u8]) { |
1077 | let compressed = crate::compress_to_vec(data); |
1078 | let decompressed = decompress_to_vec(&compressed).unwrap(); |
1079 | assert_eq!(&decompressed, data); |
1080 | } |
1081 | |
1082 | fn roundtrip_miniz_oxide(data: &[u8]) { |
1083 | let compressed = miniz_oxide::deflate::compress_to_vec_zlib(data, 3); |
1084 | let decompressed = decompress_to_vec(&compressed).unwrap(); |
1085 | assert_eq!(decompressed.len(), data.len()); |
1086 | for (i, (a, b)) in decompressed.chunks(1).zip(data.chunks(1)).enumerate() { |
1087 | assert_eq!(a, b, "chunk {}.. {}" , i * 1, i * 1 + 1); |
1088 | } |
1089 | assert_eq!(&decompressed, data); |
1090 | } |
1091 | |
1092 | #[allow (unused)] |
1093 | fn compare_decompression(data: &[u8]) { |
1094 | // let decompressed0 = flate2::read::ZlibDecoder::new(std::io::Cursor::new(&data)) |
1095 | // .bytes() |
1096 | // .collect::<Result<Vec<_>, _>>() |
1097 | // .unwrap(); |
1098 | let decompressed = decompress_to_vec(&data).unwrap(); |
1099 | let decompressed2 = miniz_oxide::inflate::decompress_to_vec_zlib(&data).unwrap(); |
1100 | for i in 0..decompressed.len().min(decompressed2.len()) { |
1101 | if decompressed[i] != decompressed2[i] { |
1102 | panic!( |
1103 | "mismatch at index {} {:?} {:?}" , |
1104 | i, |
1105 | &decompressed[i.saturating_sub(1)..(i + 16).min(decompressed.len())], |
1106 | &decompressed2[i.saturating_sub(1)..(i + 16).min(decompressed2.len())] |
1107 | ); |
1108 | } |
1109 | } |
1110 | if decompressed != decompressed2 { |
1111 | panic!( |
1112 | "length mismatch {} {} {:x?}" , |
1113 | decompressed.len(), |
1114 | decompressed2.len(), |
1115 | &decompressed2[decompressed.len()..][..16] |
1116 | ); |
1117 | } |
1118 | //assert_eq!(decompressed, decompressed2); |
1119 | } |
1120 | |
1121 | #[test ] |
1122 | fn tables() { |
1123 | for (i, &bits) in LEN_SYM_TO_LEN_EXTRA.iter().enumerate() { |
1124 | let len_base = LEN_SYM_TO_LEN_BASE[i]; |
1125 | for j in 0..(1 << bits) { |
1126 | if i == 27 && j == 31 { |
1127 | continue; |
1128 | } |
1129 | assert_eq!(LENGTH_TO_LEN_EXTRA[len_base + j - 3], bits, " {} {}" , i, j); |
1130 | assert_eq!( |
1131 | LENGTH_TO_SYMBOL[len_base + j - 3], |
1132 | i as u16 + 257, |
1133 | " {} {}" , |
1134 | i, |
1135 | j |
1136 | ); |
1137 | } |
1138 | } |
1139 | } |
1140 | |
1141 | #[test ] |
1142 | fn fdeflate_table() { |
1143 | let mut compression = CompressedBlock { |
1144 | litlen_table: [0; 4096], |
1145 | dist_table: [0; 512], |
1146 | dist_symbol_lengths: [0; 30], |
1147 | dist_symbol_masks: [0; 30], |
1148 | dist_symbol_codes: [0; 30], |
1149 | secondary_table: Vec::new(), |
1150 | eof_code: 0, |
1151 | eof_mask: 0, |
1152 | eof_bits: 0, |
1153 | }; |
1154 | let mut lengths = tables::HUFFMAN_LENGTHS.to_vec(); |
1155 | lengths.resize(288, 0); |
1156 | lengths.push(1); |
1157 | lengths.resize(320, 0); |
1158 | Decompressor::build_tables(286, &lengths, &mut compression, 11).unwrap(); |
1159 | |
1160 | assert_eq!( |
1161 | compression, FDEFLATE_COMPRESSED_BLOCK, |
1162 | " {:#x?}" , |
1163 | compression |
1164 | ); |
1165 | } |
1166 | |
1167 | #[test ] |
1168 | fn it_works() { |
1169 | roundtrip(b"Hello world!" ); |
1170 | } |
1171 | |
1172 | #[test ] |
1173 | fn constant() { |
1174 | roundtrip_miniz_oxide(&vec![0; 50]); |
1175 | roundtrip_miniz_oxide(&vec![5; 2048]); |
1176 | roundtrip_miniz_oxide(&vec![128; 2048]); |
1177 | roundtrip_miniz_oxide(&vec![254; 2048]); |
1178 | } |
1179 | |
1180 | #[test ] |
1181 | fn random() { |
1182 | let mut rng = rand::thread_rng(); |
1183 | let mut data = vec![0; 50000]; |
1184 | for _ in 0..10 { |
1185 | for byte in &mut data { |
1186 | *byte = rng.gen::<u8>() % 5; |
1187 | } |
1188 | println!("Random data: {:?}" , data); |
1189 | roundtrip_miniz_oxide(&data); |
1190 | } |
1191 | } |
1192 | |
1193 | #[test ] |
1194 | fn ignore_adler32() { |
1195 | let mut compressed = crate::compress_to_vec(b"Hello world!" ); |
1196 | let last_byte = compressed.len() - 1; |
1197 | compressed[last_byte] = compressed[last_byte].wrapping_add(1); |
1198 | |
1199 | match decompress_to_vec(&compressed) { |
1200 | Err(DecompressionError::WrongChecksum) => {} |
1201 | r => panic!("expected WrongChecksum, got {:?}" , r), |
1202 | } |
1203 | |
1204 | let mut decompressor = Decompressor::new(); |
1205 | decompressor.ignore_adler32(); |
1206 | let mut decompressed = vec![0; 1024]; |
1207 | let decompressed_len = decompressor |
1208 | .read(&compressed, &mut decompressed, 0, true) |
1209 | .unwrap() |
1210 | .1; |
1211 | assert_eq!(&decompressed[..decompressed_len], b"Hello world!" ); |
1212 | } |
1213 | |
1214 | #[test ] |
1215 | fn checksum_after_eof() { |
1216 | let input = b"Hello world!" ; |
1217 | let compressed = crate::compress_to_vec(input); |
1218 | |
1219 | let mut decompressor = Decompressor::new(); |
1220 | let mut decompressed = vec![0; 1024]; |
1221 | let (input_consumed, output_written) = decompressor |
1222 | .read( |
1223 | &compressed[..compressed.len() - 1], |
1224 | &mut decompressed, |
1225 | 0, |
1226 | false, |
1227 | ) |
1228 | .unwrap(); |
1229 | assert_eq!(output_written, input.len()); |
1230 | assert_eq!(input_consumed, compressed.len() - 1); |
1231 | |
1232 | let (input_consumed, output_written) = decompressor |
1233 | .read( |
1234 | &compressed[input_consumed..], |
1235 | &mut decompressed[..output_written], |
1236 | output_written, |
1237 | true, |
1238 | ) |
1239 | .unwrap(); |
1240 | assert!(decompressor.is_done()); |
1241 | assert_eq!(input_consumed, 1); |
1242 | assert_eq!(output_written, 0); |
1243 | |
1244 | assert_eq!(&decompressed[..input.len()], input); |
1245 | } |
1246 | |
1247 | #[test ] |
1248 | fn zero_length() { |
1249 | let mut compressed = crate::compress_to_vec(b"" ).to_vec(); |
1250 | |
1251 | // Splice in zero-length non-compressed blocks. |
1252 | for _ in 0..10 { |
1253 | println!("compressed len: {}" , compressed.len()); |
1254 | compressed.splice(2..2, [0u8, 0, 0, 0xff, 0xff].into_iter()); |
1255 | } |
1256 | |
1257 | // Ensure that the full input is decompressed, regardless of whether |
1258 | // `end_of_input` is set. |
1259 | for end_of_input in [true, false] { |
1260 | let mut decompressor = Decompressor::new(); |
1261 | let (input_consumed, output_written) = decompressor |
1262 | .read(&compressed, &mut [], 0, end_of_input) |
1263 | .unwrap(); |
1264 | |
1265 | assert!(decompressor.is_done()); |
1266 | assert_eq!(input_consumed, compressed.len()); |
1267 | assert_eq!(output_written, 0); |
1268 | } |
1269 | } |
1270 | } |
1271 | |