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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415
use crate::leb128;
use crate::serialize::{Decodable, Decoder, Encodable, Encoder};
use std::fs::File;
use std::io::{self, Write};
use std::marker::PhantomData;
use std::ops::Range;
use std::path::Path;
// -----------------------------------------------------------------------------
// Encoder
// -----------------------------------------------------------------------------
pub type FileEncodeResult = Result<usize, io::Error>;
/// The size of the buffer in `FileEncoder`.
const BUF_SIZE: usize = 8192;
/// `FileEncoder` encodes data to file via fixed-size buffer.
///
/// There used to be a `MemEncoder` type that encoded all the data into a
/// `Vec`. `FileEncoder` is better because its memory use is determined by the
/// size of the buffer, rather than the full length of the encoded data, and
/// because it doesn't need to reallocate memory along the way.
pub struct FileEncoder {
// The input buffer. For adequate performance, we need to be able to write
// directly to the unwritten region of the buffer, without calling copy_from_slice.
// Note that our buffer is always initialized so that we can do that direct access
// without unsafe code. Users of this type write many more than BUF_SIZE bytes, so the
// initialization is approximately free.
buf: Box<[u8; BUF_SIZE]>,
buffered: usize,
flushed: usize,
file: File,
// This is used to implement delayed error handling, as described in the
// comment on `trait Encoder`.
res: Result<(), io::Error>,
}
impl FileEncoder {
pub fn new<P: AsRef<Path>>(path: P) -> io::Result<Self> {
// File::create opens the file for writing only. When -Zmeta-stats is enabled, the metadata
// encoder rewinds the file to inspect what was written. So we need to always open the file
// for reading and writing.
let file = File::options().read(true).write(true).create(true).truncate(true).open(path)?;
Ok(FileEncoder {
buf: vec![0u8; BUF_SIZE].into_boxed_slice().try_into().unwrap(),
buffered: 0,
flushed: 0,
file,
res: Ok(()),
})
}
#[inline]
pub fn position(&self) -> usize {
// Tracking position this way instead of having a `self.position` field
// means that we only need to update `self.buffered` on a write call,
// as opposed to updating `self.position` and `self.buffered`.
self.flushed + self.buffered
}
#[cold]
#[inline(never)]
pub fn flush(&mut self) {
if self.res.is_ok() {
self.res = self.file.write_all(&self.buf[..self.buffered]);
}
self.flushed += self.buffered;
self.buffered = 0;
}
pub fn file(&self) -> &File {
&self.file
}
#[inline]
fn buffer_empty(&mut self) -> &mut [u8] {
// SAFETY: self.buffered is inbounds as an invariant of the type
unsafe { self.buf.get_unchecked_mut(self.buffered..) }
}
#[cold]
#[inline(never)]
fn write_all_cold_path(&mut self, buf: &[u8]) {
self.flush();
if let Some(dest) = self.buf.get_mut(..buf.len()) {
dest.copy_from_slice(buf);
self.buffered += buf.len();
} else {
if self.res.is_ok() {
self.res = self.file.write_all(buf);
}
self.flushed += buf.len();
}
}
#[inline]
fn write_all(&mut self, buf: &[u8]) {
if let Some(dest) = self.buffer_empty().get_mut(..buf.len()) {
dest.copy_from_slice(buf);
self.buffered += buf.len();
} else {
self.write_all_cold_path(buf);
}
}
/// Write up to `N` bytes to this encoder.
///
/// This function can be used to avoid the overhead of calling memcpy for writes that
/// have runtime-variable length, but are small and have a small fixed upper bound.
///
/// This can be used to do in-place encoding as is done for leb128 (without this function
/// we would need to write to a temporary buffer then memcpy into the encoder), and it can
/// also be used to implement the varint scheme we use for rmeta and dep graph encoding,
/// where we only want to encode the first few bytes of an integer. Copying in the whole
/// integer then only advancing the encoder state for the few bytes we care about is more
/// efficient than calling [`FileEncoder::write_all`], because variable-size copies are
/// always lowered to `memcpy`, which has overhead and contains a lot of logic we can bypass
/// with this function. Note that common architectures support fixed-size writes up to 8 bytes
/// with one instruction, so while this does in some sense do wasted work, we come out ahead.
#[inline]
pub fn write_with<const N: usize>(&mut self, visitor: impl FnOnce(&mut [u8; N]) -> usize) {
let flush_threshold = const { BUF_SIZE.checked_sub(N).unwrap() };
if std::intrinsics::unlikely(self.buffered > flush_threshold) {
self.flush();
}
// SAFETY: We checked above that that N < self.buffer_empty().len(),
// and if isn't, flush ensures that our empty buffer is now BUF_SIZE.
// We produce a post-mono error if N > BUF_SIZE.
let buf = unsafe { self.buffer_empty().first_chunk_mut::<N>().unwrap_unchecked() };
let written = visitor(buf);
// We have to ensure that an errant visitor cannot cause self.buffered to exeed BUF_SIZE.
if written > N {
Self::panic_invalid_write::<N>(written);
}
self.buffered += written;
}
#[cold]
#[inline(never)]
fn panic_invalid_write<const N: usize>(written: usize) {
panic!("FileEncoder::write_with::<{N}> cannot be used to write {written} bytes");
}
/// Helper for calls where [`FileEncoder::write_with`] always writes the whole array.
#[inline]
pub fn write_array<const N: usize>(&mut self, buf: [u8; N]) {
self.write_with(|dest| {
*dest = buf;
N
})
}
pub fn finish(mut self) -> Result<usize, io::Error> {
self.flush();
match std::mem::replace(&mut self.res, Ok(())) {
Ok(()) => Ok(self.position()),
Err(e) => Err(e),
}
}
}
impl Drop for FileEncoder {
fn drop(&mut self) {
// Likely to be a no-op, because `finish` should have been called and
// it also flushes. But do it just in case.
self.flush();
}
}
macro_rules! write_leb128 {
($this_fn:ident, $int_ty:ty, $write_leb_fn:ident) => {
#[inline]
fn $this_fn(&mut self, v: $int_ty) {
self.write_with(|buf| leb128::$write_leb_fn(buf, v))
}
};
}
impl Encoder for FileEncoder {
write_leb128!(emit_usize, usize, write_usize_leb128);
write_leb128!(emit_u128, u128, write_u128_leb128);
write_leb128!(emit_u64, u64, write_u64_leb128);
write_leb128!(emit_u32, u32, write_u32_leb128);
#[inline]
fn emit_u16(&mut self, v: u16) {
self.write_array(v.to_le_bytes());
}
#[inline]
fn emit_u8(&mut self, v: u8) {
self.write_array([v]);
}
write_leb128!(emit_isize, isize, write_isize_leb128);
write_leb128!(emit_i128, i128, write_i128_leb128);
write_leb128!(emit_i64, i64, write_i64_leb128);
write_leb128!(emit_i32, i32, write_i32_leb128);
#[inline]
fn emit_i16(&mut self, v: i16) {
self.write_array(v.to_le_bytes());
}
#[inline]
fn emit_raw_bytes(&mut self, s: &[u8]) {
self.write_all(s);
}
}
// -----------------------------------------------------------------------------
// Decoder
// -----------------------------------------------------------------------------
// Conceptually, `MemDecoder` wraps a `&[u8]` with a cursor into it that is always valid.
// This is implemented with three pointers, two which represent the original slice and a
// third that is our cursor.
// It is an invariant of this type that start <= current <= end.
// Additionally, the implementation of this type never modifies start and end.
pub struct MemDecoder<'a> {
start: *const u8,
current: *const u8,
end: *const u8,
_marker: PhantomData<&'a u8>,
}
impl<'a> MemDecoder<'a> {
#[inline]
pub fn new(data: &'a [u8], position: usize) -> MemDecoder<'a> {
let Range { start, end } = data.as_ptr_range();
MemDecoder { start, current: data[position..].as_ptr(), end, _marker: PhantomData }
}
#[inline]
pub fn data(&self) -> &'a [u8] {
// SAFETY: This recovers the original slice, only using members we never modify.
unsafe { std::slice::from_raw_parts(self.start, self.len()) }
}
#[inline]
pub fn len(&self) -> usize {
// SAFETY: This recovers the length of the original slice, only using members we never modify.
unsafe { self.end.sub_ptr(self.start) }
}
#[inline]
pub fn remaining(&self) -> usize {
// SAFETY: This type guarantees current <= end.
unsafe { self.end.sub_ptr(self.current) }
}
#[cold]
#[inline(never)]
fn decoder_exhausted() -> ! {
panic!("MemDecoder exhausted")
}
#[inline]
pub fn read_array<const N: usize>(&mut self) -> [u8; N] {
self.read_raw_bytes(N).try_into().unwrap()
}
/// While we could manually expose manipulation of the decoder position,
/// all current users of that method would need to reset the position later,
/// incurring the bounds check of set_position twice.
#[inline]
pub fn with_position<F, T>(&mut self, pos: usize, func: F) -> T
where
F: Fn(&mut MemDecoder<'a>) -> T,
{
struct SetOnDrop<'a, 'guarded> {
decoder: &'guarded mut MemDecoder<'a>,
current: *const u8,
}
impl Drop for SetOnDrop<'_, '_> {
fn drop(&mut self) {
self.decoder.current = self.current;
}
}
if pos >= self.len() {
Self::decoder_exhausted();
}
let previous = self.current;
// SAFETY: We just checked if this add is in-bounds above.
unsafe {
self.current = self.start.add(pos);
}
let guard = SetOnDrop { current: previous, decoder: self };
func(guard.decoder)
}
}
macro_rules! read_leb128 {
($this_fn:ident, $int_ty:ty, $read_leb_fn:ident) => {
#[inline]
fn $this_fn(&mut self) -> $int_ty {
leb128::$read_leb_fn(self)
}
};
}
impl<'a> Decoder for MemDecoder<'a> {
read_leb128!(read_usize, usize, read_usize_leb128);
read_leb128!(read_u128, u128, read_u128_leb128);
read_leb128!(read_u64, u64, read_u64_leb128);
read_leb128!(read_u32, u32, read_u32_leb128);
#[inline]
fn read_u16(&mut self) -> u16 {
u16::from_le_bytes(self.read_array())
}
#[inline]
fn read_u8(&mut self) -> u8 {
if self.current == self.end {
Self::decoder_exhausted();
}
// SAFETY: This type guarantees current <= end, and we just checked current == end.
unsafe {
let byte = *self.current;
self.current = self.current.add(1);
byte
}
}
read_leb128!(read_isize, isize, read_isize_leb128);
read_leb128!(read_i128, i128, read_i128_leb128);
read_leb128!(read_i64, i64, read_i64_leb128);
read_leb128!(read_i32, i32, read_i32_leb128);
#[inline]
fn read_i16(&mut self) -> i16 {
i16::from_le_bytes(self.read_array())
}
#[inline]
fn read_raw_bytes(&mut self, bytes: usize) -> &'a [u8] {
if bytes > self.remaining() {
Self::decoder_exhausted();
}
// SAFETY: We just checked if this range is in-bounds above.
unsafe {
let slice = std::slice::from_raw_parts(self.current, bytes);
self.current = self.current.add(bytes);
slice
}
}
#[inline]
fn peek_byte(&self) -> u8 {
if self.current == self.end {
Self::decoder_exhausted();
}
// SAFETY: This type guarantees current is inbounds or one-past-the-end, which is end.
// Since we just checked current == end, the current pointer must be inbounds.
unsafe { *self.current }
}
#[inline]
fn position(&self) -> usize {
// SAFETY: This type guarantees start <= current
unsafe { self.current.sub_ptr(self.start) }
}
}
// Specializations for contiguous byte sequences follow. The default implementations for slices
// encode and decode each element individually. This isn't necessary for `u8` slices when using
// opaque encoders and decoders, because each `u8` is unchanged by encoding and decoding.
// Therefore, we can use more efficient implementations that process the entire sequence at once.
// Specialize encoding byte slices. This specialization also applies to encoding `Vec<u8>`s, etc.,
// since the default implementations call `encode` on their slices internally.
impl Encodable<FileEncoder> for [u8] {
fn encode(&self, e: &mut FileEncoder) {
Encoder::emit_usize(e, self.len());
e.emit_raw_bytes(self);
}
}
// Specialize decoding `Vec<u8>`. This specialization also applies to decoding `Box<[u8]>`s, etc.,
// since the default implementations call `decode` to produce a `Vec<u8>` internally.
impl<'a> Decodable<MemDecoder<'a>> for Vec<u8> {
fn decode(d: &mut MemDecoder<'a>) -> Self {
let len = Decoder::read_usize(d);
d.read_raw_bytes(len).to_owned()
}
}
/// An integer that will always encode to 8 bytes.
pub struct IntEncodedWithFixedSize(pub u64);
impl IntEncodedWithFixedSize {
pub const ENCODED_SIZE: usize = 8;
}
impl Encodable<FileEncoder> for IntEncodedWithFixedSize {
#[inline]
fn encode(&self, e: &mut FileEncoder) {
let _start_pos = e.position();
e.write_array(self.0.to_le_bytes());
let _end_pos = e.position();
debug_assert_eq!((_end_pos - _start_pos), IntEncodedWithFixedSize::ENCODED_SIZE);
}
}
impl<'a> Decodable<MemDecoder<'a>> for IntEncodedWithFixedSize {
#[inline]
fn decode(decoder: &mut MemDecoder<'a>) -> IntEncodedWithFixedSize {
let bytes = decoder.read_array::<{ IntEncodedWithFixedSize::ENCODED_SIZE }>();
IntEncodedWithFixedSize(u64::from_le_bytes(bytes))
}
}