| 1 |
//! Composable structures to handle reading an image. |
| 2 |
|
| 3 |
|
| 4 |
use std::convert::TryFrom; |
| 5 |
use std::fmt::Debug; |
| 6 |
use std::io::{Read, Seek}; |
| 7 |
use crate::block::{BlockIndex, UncompressedBlock}; |
| 8 |
use crate::block::chunk::{Chunk, TileCoordinates}; |
| 9 |
use crate::error::{Error, Result, u64_to_usize, UnitResult}; |
| 10 |
use crate::io::{PeekRead, Tracking}; |
| 11 |
use crate::meta::{MetaData, OffsetTables}; |
| 12 |
use crate::meta::header::Header; |
| 13 |
|
| 14 |
/// Decode the meta data from a byte source, keeping the source ready for further reading. |
| 15 |
/// Continue decoding the remaining bytes by calling `filtered_chunks` or `all_chunks`. |
| 16 |
#[derive(Debug)] |
| 17 |
pub struct Reader<R> { |
| 18 |
meta_data: MetaData, |
| 19 |
remaining_reader: PeekRead<Tracking<R>>, // TODO does R need to be Seek or is Tracking enough? |
| 20 |
} |
| 21 |
|
| 22 |
impl<R: Read + Seek> Reader<R> { |
| 23 |
|
| 24 |
/// Start the reading process. |
| 25 |
/// Immediately decodes the meta data into an internal field. |
| 26 |
/// Access it via`meta_data()`. |
| 27 |
pub fn read_from_buffered(read: R, pedantic: bool) -> Result<Self> { |
| 28 |
let mut remaining_reader = PeekRead::new(Tracking::new(read)); |
| 29 |
let meta_data = MetaData::read_validated_from_buffered_peekable(&mut remaining_reader, pedantic)?; |
| 30 |
Ok(Self { meta_data, remaining_reader }) |
| 31 |
} |
| 32 |
|
| 33 |
// must not be mutable, as reading the file later on relies on the meta data |
| 34 |
/// The decoded exr meta data from the file. |
| 35 |
pub fn meta_data(&self) -> &MetaData { &self.meta_data } |
| 36 |
|
| 37 |
/// The decoded exr meta data from the file. |
| 38 |
pub fn headers(&self) -> &[Header] { &self.meta_data.headers } |
| 39 |
|
| 40 |
/// Obtain the meta data ownership. |
| 41 |
pub fn into_meta_data(self) -> MetaData { self.meta_data } |
| 42 |
|
| 43 |
/// Prepare to read all the chunks from the file. |
| 44 |
/// Does not decode the chunks now, but returns a decoder. |
| 45 |
/// Reading all chunks reduces seeking the file, but some chunks might be read without being used. |
| 46 |
pub fn all_chunks(mut self, pedantic: bool) -> Result<AllChunksReader<R>> { |
| 47 |
let total_chunk_count = { |
| 48 |
if pedantic { |
| 49 |
let offset_tables = MetaData::read_offset_tables(&mut self.remaining_reader, &self.meta_data.headers)?; |
| 50 |
validate_offset_tables(self.meta_data.headers.as_slice(), &offset_tables, self.remaining_reader.byte_position())?; |
| 51 |
offset_tables.iter().map(|table| table.len()).sum() |
| 52 |
} |
| 53 |
else { |
| 54 |
MetaData::skip_offset_tables(&mut self.remaining_reader, &self.meta_data.headers)? |
| 55 |
} |
| 56 |
}; |
| 57 |
|
| 58 |
Ok(AllChunksReader { |
| 59 |
meta_data: self.meta_data, |
| 60 |
remaining_chunks: 0 .. total_chunk_count, |
| 61 |
remaining_bytes: self.remaining_reader, |
| 62 |
pedantic |
| 63 |
}) |
| 64 |
} |
| 65 |
|
| 66 |
/// Prepare to read some the chunks from the file. |
| 67 |
/// Does not decode the chunks now, but returns a decoder. |
| 68 |
/// Reading only some chunks may seeking the file, potentially skipping many bytes. |
| 69 |
// TODO tile indices add no new information to block index?? |
| 70 |
pub fn filter_chunks(mut self, pedantic: bool, mut filter: impl FnMut(&MetaData, TileCoordinates, BlockIndex) -> bool) -> Result<FilteredChunksReader<R>> { |
| 71 |
let offset_tables = MetaData::read_offset_tables(&mut self.remaining_reader, &self.meta_data.headers)?; |
| 72 |
|
| 73 |
// TODO regardless of pedantic, if invalid, read all chunks instead, and filter after reading each chunk? |
| 74 |
if pedantic { |
| 75 |
validate_offset_tables( |
| 76 |
self.meta_data.headers.as_slice(), &offset_tables, |
| 77 |
self.remaining_reader.byte_position() |
| 78 |
)?; |
| 79 |
} |
| 80 |
|
| 81 |
let mut filtered_offsets = Vec::with_capacity( |
| 82 |
(self.meta_data.headers.len() * 32).min(2*2048) |
| 83 |
); |
| 84 |
|
| 85 |
// TODO detect whether the filter actually would skip chunks, and aviod sorting etc when not filtering is applied |
| 86 |
|
| 87 |
for (header_index, header) in self.meta_data.headers.iter().enumerate() { // offset tables are stored same order as headers |
| 88 |
for (block_index, tile) in header.blocks_increasing_y_order().enumerate() { // in increasing_y order |
| 89 |
let data_indices = header.get_absolute_block_pixel_coordinates(tile.location)?; |
| 90 |
|
| 91 |
let block = BlockIndex { |
| 92 |
layer: header_index, |
| 93 |
level: tile.location.level_index, |
| 94 |
pixel_position: data_indices.position.to_usize("data indices start")?, |
| 95 |
pixel_size: data_indices.size, |
| 96 |
}; |
| 97 |
|
| 98 |
if filter(&self.meta_data, tile.location, block) { |
| 99 |
filtered_offsets.push(offset_tables[header_index][block_index]) // safe indexing from `enumerate()` |
| 100 |
} |
| 101 |
}; |
| 102 |
} |
| 103 |
|
| 104 |
filtered_offsets.sort_unstable(); // enables reading continuously if possible (already sorted where line order increasing) |
| 105 |
|
| 106 |
if pedantic { |
| 107 |
// table is sorted. if any two neighbours are equal, we have duplicates. this is invalid. |
| 108 |
if filtered_offsets.windows(2).any(|pair| pair[0] == pair[1]) { |
| 109 |
return Err(Error::invalid("chunk offset table")) |
| 110 |
} |
| 111 |
} |
| 112 |
|
| 113 |
Ok(FilteredChunksReader { |
| 114 |
meta_data: self.meta_data, |
| 115 |
expected_filtered_chunk_count: filtered_offsets.len(), |
| 116 |
remaining_filtered_chunk_indices: filtered_offsets.into_iter(), |
| 117 |
remaining_bytes: self.remaining_reader |
| 118 |
}) |
| 119 |
} |
| 120 |
} |
| 121 |
|
| 122 |
|
| 123 |
fn validate_offset_tables(headers: &[Header], offset_tables: &OffsetTables, chunks_start_byte: usize) -> UnitResult { |
| 124 |
let max_pixel_bytes: usize = headers.iter() // when compressed, chunks are smaller, but never larger than max |
| 125 |
.map(|header| header.max_pixel_file_bytes()) |
| 126 |
.sum(); |
| 127 |
|
| 128 |
// check that each offset is within the bounds |
| 129 |
let end_byte = chunks_start_byte + max_pixel_bytes; |
| 130 |
let is_invalid = offset_tables.iter().flatten().map(|&u64| u64_to_usize(u64, "chunk start")) |
| 131 |
.any(|maybe_chunk_start| match maybe_chunk_start { |
| 132 |
Ok(chunk_start) => chunk_start < chunks_start_byte || chunk_start > end_byte, |
| 133 |
Err(_) => true |
| 134 |
}); |
| 135 |
|
| 136 |
if is_invalid { Err(Error::invalid("offset table")) } |
| 137 |
else { Ok(()) } |
| 138 |
} |
| 139 |
|
| 140 |
|
| 141 |
|
| 142 |
|
| 143 |
/// Decode the desired chunks and skip the unimportant chunks in the file. |
| 144 |
/// The decoded chunks can be decompressed by calling |
| 145 |
/// `decompress_parallel`, `decompress_sequential`, or `sequential_decompressor` or `parallel_decompressor`. |
| 146 |
/// Call `on_progress` to have a callback with each block. |
| 147 |
/// Also contains the image meta data. |
| 148 |
#[derive(Debug)] |
| 149 |
pub struct FilteredChunksReader<R> { |
| 150 |
meta_data: MetaData, |
| 151 |
expected_filtered_chunk_count: usize, |
| 152 |
remaining_filtered_chunk_indices: std::vec::IntoIter<u64>, |
| 153 |
remaining_bytes: PeekRead<Tracking<R>>, |
| 154 |
} |
| 155 |
|
| 156 |
/// Decode all chunks in the file without seeking. |
| 157 |
/// The decoded chunks can be decompressed by calling |
| 158 |
/// `decompress_parallel`, `decompress_sequential`, or `sequential_decompressor` or `parallel_decompressor`. |
| 159 |
/// Call `on_progress` to have a callback with each block. |
| 160 |
/// Also contains the image meta data. |
| 161 |
#[derive(Debug)] |
| 162 |
pub struct AllChunksReader<R> { |
| 163 |
meta_data: MetaData, |
| 164 |
remaining_chunks: std::ops::Range<usize>, |
| 165 |
remaining_bytes: PeekRead<Tracking<R>>, |
| 166 |
pedantic: bool, |
| 167 |
} |
| 168 |
|
| 169 |
/// Decode chunks in the file without seeking. |
| 170 |
/// Calls the supplied closure for each chunk. |
| 171 |
/// The decoded chunks can be decompressed by calling |
| 172 |
/// `decompress_parallel`, `decompress_sequential`, or `sequential_decompressor`. |
| 173 |
/// Also contains the image meta data. |
| 174 |
#[derive(Debug)] |
| 175 |
pub struct OnProgressChunksReader<R, F> { |
| 176 |
chunks_reader: R, |
| 177 |
decoded_chunks: usize, |
| 178 |
callback: F, |
| 179 |
} |
| 180 |
|
| 181 |
/// Decode chunks in the file. |
| 182 |
/// The decoded chunks can be decompressed by calling |
| 183 |
/// `decompress_parallel`, `decompress_sequential`, or `sequential_decompressor`. |
| 184 |
/// Call `on_progress` to have a callback with each block. |
| 185 |
/// Also contains the image meta data. |
| 186 |
pub trait ChunksReader: Sized + Iterator<Item=Result<Chunk>> + ExactSizeIterator { |
| 187 |
|
| 188 |
/// The decoded exr meta data from the file. |
| 189 |
fn meta_data(&self) -> &MetaData; |
| 190 |
|
| 191 |
/// The decoded exr headers from the file. |
| 192 |
fn headers(&self) -> &[Header] { &self.meta_data().headers } |
| 193 |
|
| 194 |
/// The number of chunks that this reader will return in total. |
| 195 |
/// Can be less than the total number of chunks in the file, if some chunks are skipped. |
| 196 |
fn expected_chunk_count(&self) -> usize; |
| 197 |
|
| 198 |
/// Read the next compressed chunk from the file. |
| 199 |
/// Equivalent to `.next()`, as this also is an iterator. |
| 200 |
/// Returns `None` if all chunks have been read. |
| 201 |
fn read_next_chunk(&mut self) -> Option<Result<Chunk>> { self.next() } |
| 202 |
|
| 203 |
/// Create a new reader that calls the provided progress |
| 204 |
/// callback for each chunk that is read from the file. |
| 205 |
/// If the file can be successfully decoded, |
| 206 |
/// the progress will always at least once include 0.0 at the start and 1.0 at the end. |
| 207 |
fn on_progress<F>(self, on_progress: F) -> OnProgressChunksReader<Self, F> where F: FnMut(f64) { |
| 208 |
OnProgressChunksReader { chunks_reader: self, callback: on_progress, decoded_chunks: 0 } |
| 209 |
} |
| 210 |
|
| 211 |
#[cfg(feature = "rayon")] |
| 212 |
/// Decompress all blocks in the file, using multiple cpu cores, and call the supplied closure for each block. |
| 213 |
/// The order of the blocks is not deterministic. |
| 214 |
/// You can also use `parallel_decompressor` to obtain an iterator instead. |
| 215 |
/// Will fallback to sequential processing where threads are not available, or where it would not speed up the process. |
| 216 |
// FIXME try async + futures instead of rayon! Maybe even allows for external async decoding? (-> impl Stream<UncompressedBlock>) |
| 217 |
fn decompress_parallel( |
| 218 |
self, pedantic: bool, |
| 219 |
mut insert_block: impl FnMut(&MetaData, UncompressedBlock) -> UnitResult |
| 220 |
) -> UnitResult |
| 221 |
{ |
| 222 |
let mut decompressor = match self.parallel_decompressor(pedantic) { |
| 223 |
Err(old_self) => return old_self.decompress_sequential(pedantic, insert_block), |
| 224 |
Ok(decompressor) => decompressor, |
| 225 |
}; |
| 226 |
|
| 227 |
while let Some(block) = decompressor.next() { |
| 228 |
insert_block(decompressor.meta_data(), block?)?; |
| 229 |
} |
| 230 |
|
| 231 |
debug_assert_eq!(decompressor.len(), 0, "compressed blocks left after decompressing all blocks"); |
| 232 |
Ok(()) |
| 233 |
} |
| 234 |
|
| 235 |
#[cfg(feature = "rayon")] |
| 236 |
/// Return an iterator that decompresses the chunks with multiple threads. |
| 237 |
/// The order of the blocks is not deterministic. |
| 238 |
/// Use `ParallelBlockDecompressor::new` if you want to use your own thread pool. |
| 239 |
/// By default, this uses as many threads as there are CPUs. |
| 240 |
/// Returns the `self` if there is no need for parallel decompression. |
| 241 |
fn parallel_decompressor(self, pedantic: bool) -> std::result::Result<ParallelBlockDecompressor<Self>, Self> { |
| 242 |
ParallelBlockDecompressor::new(self, pedantic) |
| 243 |
} |
| 244 |
|
| 245 |
/// Return an iterator that decompresses the chunks in this thread. |
| 246 |
/// You can alternatively use `sequential_decompressor` if you prefer an external iterator. |
| 247 |
fn decompress_sequential( |
| 248 |
self, pedantic: bool, |
| 249 |
mut insert_block: impl FnMut(&MetaData, UncompressedBlock) -> UnitResult |
| 250 |
) -> UnitResult |
| 251 |
{ |
| 252 |
let mut decompressor = self.sequential_decompressor(pedantic); |
| 253 |
while let Some(block) = decompressor.next() { |
| 254 |
insert_block(decompressor.meta_data(), block?)?; |
| 255 |
} |
| 256 |
|
| 257 |
debug_assert_eq!(decompressor.len(), 0, "compressed blocks left after decompressing all blocks"); |
| 258 |
Ok(()) |
| 259 |
} |
| 260 |
|
| 261 |
/// Prepare reading the chunks sequentially, only a single thread, but with less memory overhead. |
| 262 |
fn sequential_decompressor(self, pedantic: bool) -> SequentialBlockDecompressor<Self> { |
| 263 |
SequentialBlockDecompressor { remaining_chunks_reader: self, pedantic } |
| 264 |
} |
| 265 |
} |
| 266 |
|
| 267 |
impl<R, F> ChunksReader for OnProgressChunksReader<R, F> where R: ChunksReader, F: FnMut(f64) { |
| 268 |
fn meta_data(&self) -> &MetaData { self.chunks_reader.meta_data() } |
| 269 |
fn expected_chunk_count(&self) -> usize { self.chunks_reader.expected_chunk_count() } |
| 270 |
} |
| 271 |
|
| 272 |
impl<R, F> ExactSizeIterator for OnProgressChunksReader<R, F> where R: ChunksReader, F: FnMut(f64) {} |
| 273 |
impl<R, F> Iterator for OnProgressChunksReader<R, F> where R: ChunksReader, F: FnMut(f64) { |
| 274 |
type Item = Result<Chunk>; |
| 275 |
|
| 276 |
fn next(&mut self) -> Option<Self::Item> { |
| 277 |
self.chunks_reader.next().map(|item|{ |
| 278 |
{ |
| 279 |
let total_chunks = self.expected_chunk_count() as f64; |
| 280 |
let callback = &mut self.callback; |
| 281 |
callback(self.decoded_chunks as f64 / total_chunks); |
| 282 |
} |
| 283 |
|
| 284 |
self.decoded_chunks += 1; |
| 285 |
item |
| 286 |
}) |
| 287 |
.or_else(||{ |
| 288 |
debug_assert_eq!( |
| 289 |
self.decoded_chunks, self.expected_chunk_count(), |
| 290 |
"chunks reader finished but not all chunks are decompressed" |
| 291 |
); |
| 292 |
|
| 293 |
let callback = &mut self.callback; |
| 294 |
callback(1.0); |
| 295 |
None |
| 296 |
}) |
| 297 |
} |
| 298 |
|
| 299 |
fn size_hint(&self) -> (usize, Option<usize>) { |
| 300 |
self.chunks_reader.size_hint() |
| 301 |
} |
| 302 |
} |
| 303 |
|
| 304 |
impl<R: Read + Seek> ChunksReader for AllChunksReader<R> { |
| 305 |
fn meta_data(&self) -> &MetaData { &self.meta_data } |
| 306 |
fn expected_chunk_count(&self) -> usize { self.remaining_chunks.end } |
| 307 |
} |
| 308 |
|
| 309 |
impl<R: Read + Seek> ExactSizeIterator for AllChunksReader<R> {} |
| 310 |
impl<R: Read + Seek> Iterator for AllChunksReader<R> { |
| 311 |
type Item = Result<Chunk>; |
| 312 |
|
| 313 |
fn next(&mut self) -> Option<Self::Item> { |
| 314 |
// read as many chunks as the file should contain (inferred from meta data) |
| 315 |
let next_chunk = self.remaining_chunks.next() |
| 316 |
.map(|_| Chunk::read(&mut self.remaining_bytes, &self.meta_data)); |
| 317 |
|
| 318 |
// if no chunks are left, but some bytes remain, return error |
| 319 |
if self.pedantic && next_chunk.is_none() && self.remaining_bytes.peek_u8().is_ok() { |
| 320 |
return Some(Err(Error::invalid("end of file expected"))); |
| 321 |
} |
| 322 |
|
| 323 |
next_chunk |
| 324 |
} |
| 325 |
|
| 326 |
fn size_hint(&self) -> (usize, Option<usize>) { |
| 327 |
(self.remaining_chunks.len(), Some(self.remaining_chunks.len())) |
| 328 |
} |
| 329 |
} |
| 330 |
|
| 331 |
impl<R: Read + Seek> ChunksReader for FilteredChunksReader<R> { |
| 332 |
fn meta_data(&self) -> &MetaData { &self.meta_data } |
| 333 |
fn expected_chunk_count(&self) -> usize { self.expected_filtered_chunk_count } |
| 334 |
} |
| 335 |
|
| 336 |
impl<R: Read + Seek> ExactSizeIterator for FilteredChunksReader<R> {} |
| 337 |
impl<R: Read + Seek> Iterator for FilteredChunksReader<R> { |
| 338 |
type Item = Result<Chunk>; |
| 339 |
|
| 340 |
fn next(&mut self) -> Option<Self::Item> { |
| 341 |
// read as many chunks as we have desired chunk offsets |
| 342 |
self.remaining_filtered_chunk_indices.next().map(|next_chunk_location|{ |
| 343 |
self.remaining_bytes.skip_to( // no-op for seek at current position, uses skip_bytes for small amounts |
| 344 |
usize::try_from(next_chunk_location)? |
| 345 |
)?; |
| 346 |
|
| 347 |
let meta_data = &self.meta_data; |
| 348 |
Chunk::read(&mut self.remaining_bytes, meta_data) |
| 349 |
}) |
| 350 |
|
| 351 |
// TODO remember last chunk index and then seek to index+size and check whether bytes are left? |
| 352 |
} |
| 353 |
|
| 354 |
fn size_hint(&self) -> (usize, Option<usize>) { |
| 355 |
(self.remaining_filtered_chunk_indices.len(), Some(self.remaining_filtered_chunk_indices.len())) |
| 356 |
} |
| 357 |
} |
| 358 |
|
| 359 |
/// Read all chunks from the file, decompressing each chunk immediately. |
| 360 |
/// Implements iterator. |
| 361 |
#[derive(Debug)] |
| 362 |
pub struct SequentialBlockDecompressor<R: ChunksReader> { |
| 363 |
remaining_chunks_reader: R, |
| 364 |
pedantic: bool, |
| 365 |
} |
| 366 |
|
| 367 |
impl<R: ChunksReader> SequentialBlockDecompressor<R> { |
| 368 |
|
| 369 |
/// The extracted meta data from the image file. |
| 370 |
pub fn meta_data(&self) -> &MetaData { self.remaining_chunks_reader.meta_data() } |
| 371 |
|
| 372 |
/// Read and then decompress a single block of pixels from the byte source. |
| 373 |
pub fn decompress_next_block(&mut self) -> Option<Result<UncompressedBlock>> { |
| 374 |
self.remaining_chunks_reader.read_next_chunk().map(|compressed_chunk|{ |
| 375 |
UncompressedBlock::decompress_chunk(compressed_chunk?, &self.remaining_chunks_reader.meta_data(), self.pedantic) |
| 376 |
}) |
| 377 |
} |
| 378 |
} |
| 379 |
|
| 380 |
#[cfg(feature = "rayon")] |
| 381 |
/// Decompress the chunks in a file in parallel. |
| 382 |
/// The first call to `next` will fill the thread pool with jobs, |
| 383 |
/// starting to decompress the next few blocks. |
| 384 |
/// These jobs will finish, even if you stop reading more blocks. |
| 385 |
/// Implements iterator. |
| 386 |
#[derive(Debug)] |
| 387 |
pub struct ParallelBlockDecompressor<R: ChunksReader> { |
| 388 |
remaining_chunks: R, |
| 389 |
sender: std::sync::mpsc::Sender<Result<UncompressedBlock>>, |
| 390 |
receiver: std::sync::mpsc::Receiver<Result<UncompressedBlock>>, |
| 391 |
currently_decompressing_count: usize, |
| 392 |
max_threads: usize, |
| 393 |
|
| 394 |
shared_meta_data_ref: std::sync::Arc<MetaData>, |
| 395 |
pedantic: bool, |
| 396 |
|
| 397 |
pool: rayon_core::ThreadPool, |
| 398 |
} |
| 399 |
|
| 400 |
#[cfg(feature = "rayon")] |
| 401 |
impl<R: ChunksReader> ParallelBlockDecompressor<R> { |
| 402 |
|
| 403 |
/// Create a new decompressor. Does not immediately spawn any tasks. |
| 404 |
/// Decompression starts after the first call to `next`. |
| 405 |
/// Returns the chunks if parallel decompression should not be used. |
| 406 |
/// Use `new_with_thread_pool` to customize the threadpool. |
| 407 |
pub fn new(chunks: R, pedantic: bool) -> std::result::Result<Self, R> { |
| 408 |
Self::new_with_thread_pool(chunks, pedantic, ||{ |
| 409 |
rayon_core::ThreadPoolBuilder::new() |
| 410 |
.thread_name(|index| format!("OpenEXR Block Decompressor Thread #{}", index)) |
| 411 |
.build() |
| 412 |
}) |
| 413 |
} |
| 414 |
|
| 415 |
/// Create a new decompressor. Does not immediately spawn any tasks. |
| 416 |
/// Decompression starts after the first call to `next`. |
| 417 |
/// Returns the chunks if parallel decompression should not be used. |
| 418 |
pub fn new_with_thread_pool<CreatePool>(chunks: R, pedantic: bool, try_create_thread_pool: CreatePool) |
| 419 |
-> std::result::Result<Self, R> |
| 420 |
where CreatePool: FnOnce() -> std::result::Result<rayon_core::ThreadPool, rayon_core::ThreadPoolBuildError> |
| 421 |
{ |
| 422 |
use crate::compression::Compression; |
| 423 |
|
| 424 |
let is_entirely_uncompressed = chunks.meta_data().headers.iter() |
| 425 |
.all(|head|head.compression == Compression::Uncompressed); |
| 426 |
|
| 427 |
// if no compression is used in the file, don't use a threadpool |
| 428 |
if is_entirely_uncompressed { |
| 429 |
return Err(chunks); |
| 430 |
} |
| 431 |
|
| 432 |
// in case thread pool creation fails (for example on WASM currently), |
| 433 |
// we revert to sequential decompression |
| 434 |
let pool = match try_create_thread_pool() { |
| 435 |
Ok(pool) => pool, |
| 436 |
|
| 437 |
// TODO print warning? |
| 438 |
Err(_) => return Err(chunks), |
| 439 |
}; |
| 440 |
|
| 441 |
let max_threads = pool.current_num_threads().max(1).min(chunks.len()) + 2; // ca one block for each thread at all times |
| 442 |
|
| 443 |
let (send, recv) = std::sync::mpsc::channel(); // TODO bounded channel simplifies logic? |
| 444 |
|
| 445 |
Ok(Self { |
| 446 |
shared_meta_data_ref: std::sync::Arc::new(chunks.meta_data().clone()), |
| 447 |
currently_decompressing_count: 0, |
| 448 |
remaining_chunks: chunks, |
| 449 |
sender: send, |
| 450 |
receiver: recv, |
| 451 |
pedantic, |
| 452 |
max_threads, |
| 453 |
|
| 454 |
pool, |
| 455 |
}) |
| 456 |
} |
| 457 |
|
| 458 |
/// Fill the pool with decompression jobs. Returns the first job that finishes. |
| 459 |
pub fn decompress_next_block(&mut self) -> Option<Result<UncompressedBlock>> { |
| 460 |
|
| 461 |
while self.currently_decompressing_count < self.max_threads { |
| 462 |
let block = self.remaining_chunks.next(); |
| 463 |
if let Some(block) = block { |
| 464 |
let block = match block { |
| 465 |
Ok(block) => block, |
| 466 |
Err(error) => return Some(Err(error)) |
| 467 |
}; |
| 468 |
|
| 469 |
let sender = self.sender.clone(); |
| 470 |
let meta = self.shared_meta_data_ref.clone(); |
| 471 |
let pedantic = self.pedantic; |
| 472 |
|
| 473 |
self.currently_decompressing_count += 1; |
| 474 |
|
| 475 |
self.pool.spawn(move || { |
| 476 |
let decompressed_or_err = UncompressedBlock::decompress_chunk( |
| 477 |
block, &meta, pedantic |
| 478 |
); |
| 479 |
|
| 480 |
// by now, decompressing could have failed in another thread. |
| 481 |
// the error is then already handled, so we simply |
| 482 |
// don't send the decompressed block and do nothing |
| 483 |
let _ = sender.send(decompressed_or_err); |
| 484 |
}); |
| 485 |
} |
| 486 |
else { |
| 487 |
// there are no chunks left to decompress |
| 488 |
break; |
| 489 |
} |
| 490 |
} |
| 491 |
|
| 492 |
if self.currently_decompressing_count > 0 { |
| 493 |
let next = self.receiver.recv() |
| 494 |
.expect("all decompressing senders hung up but more messages were expected"); |
| 495 |
|
| 496 |
self.currently_decompressing_count -= 1; |
| 497 |
Some(next) |
| 498 |
} |
| 499 |
else { |
| 500 |
debug_assert!(self.receiver.try_recv().is_err(), "uncompressed chunks left in channel after decompressing all chunks"); // TODO not reliable |
| 501 |
debug_assert_eq!(self.len(), 0, "compressed chunks left after decompressing all chunks"); |
| 502 |
None |
| 503 |
} |
| 504 |
} |
| 505 |
|
| 506 |
/// The extracted meta data of the image file. |
| 507 |
pub fn meta_data(&self) -> &MetaData { self.remaining_chunks.meta_data() } |
| 508 |
} |
| 509 |
|
| 510 |
impl<R: ChunksReader> ExactSizeIterator for SequentialBlockDecompressor<R> {} |
| 511 |
impl<R: ChunksReader> Iterator for SequentialBlockDecompressor<R> { |
| 512 |
type Item = Result<UncompressedBlock>; |
| 513 |
fn next(&mut self) -> Option<Self::Item> { self.decompress_next_block() } |
| 514 |
fn size_hint(&self) -> (usize, Option<usize>) { self.remaining_chunks_reader.size_hint() } |
| 515 |
} |
| 516 |
|
| 517 |
#[cfg(feature = "rayon")] |
| 518 |
impl<R: ChunksReader> ExactSizeIterator for ParallelBlockDecompressor<R> {} |
| 519 |
#[cfg(feature = "rayon")] |
| 520 |
impl<R: ChunksReader> Iterator for ParallelBlockDecompressor<R> { |
| 521 |
type Item = Result<UncompressedBlock>; |
| 522 |
fn next(&mut self) -> Option<Self::Item> { self.decompress_next_block() } |
| 523 |
fn size_hint(&self) -> (usize, Option<usize>) { |
| 524 |
let remaining = self.remaining_chunks.len() + self.currently_decompressing_count; |
| 525 |
(remaining, Some(remaining)) |
| 526 |
} |
| 527 |
} |
| 528 |
|
| 529 |
|
| 530 |
|
| 531 |
|
| 532 |
|