Continue the incremental zstd_compress.c migration with its sequence
statistics and seqStore entropy-compression layer. The following now
live in rust/src/zstd_compress_stats.rs:
- ZSTD_seqToCodes(), exported under its original name because the C
dictionary builder (zdict.c) and decodecorpus link against it,
- ZSTD_buildSequencesStatistics() and its dummy variant, whose result
struct no longer crosses the language boundary,
- ZSTD_entropyCompressSeqStore_internal(), _wExtLitBuffer(), and
ZSTD_entropyCompressSeqStore(),
- ZSTD_buildBlockEntropyStats() with its literals/sequences helpers,
- ZSTD_copyBlockSequences() and the ZSTD_updateRep() rules it shares
with the superblock writer.
Boundary: ZSTD_CCtx and ZSTD_CCtx_params stay private to C. These
paths read exactly two parameter fields, so the C shims keep the
original static/exported function names and forward the strategy and
ZSTD_literalsCompressionIsDisabled() as int scalars alongside the
seqStore and entropy-table leaves. The leaf layouts (SeqDef,
SeqStore_t, ZSTD_hufCTables_t, ZSTD_fseCTables_t, entropy metadata,
SeqCollector, ZSTD_Sequence) are pinned by compile-time asserts in
zstd_compress.c and by both-pointer-width layout tests in Rust.
ZSTD_buildSeqStore, block dispatch/splitting, and the block-size
estimation helpers remain C for a later slice.
The superblock module previously round-tripped through the C export of
ZSTD_buildBlockEntropyStats and mirrored the entropy leaf structs
privately. It now calls the crate-internal builder directly, and the
shared struct definitions moved to zstd_compress_stats; consequently
ZSTD_rust_compressSuperBlock() takes the two parameter scalars instead
of an opaque ZSTD_CCtx_params pointer, extracted by its C shim. Its
layout and repcode tests moved with the definitions.
One C helper family gets no shim: the static
ZSTD_entropyCompressSeqStore_wExtLitBuffer() had a single caller and
was folded into the Rust implementation.
Byte-identity was verified against the pre-change compressor: COPYING,
datagen -g5000000 -s7, and datagen -g300000 -s21 -P90, each at levels
1/3/9/19 and --fast=5, plus a superblock-heavy pass at level 19 with
--target-compressed-block-size=1024; all 18 frames are byte-identical,
covering the repeat-mode state machine, longOffsets, RLE/raw fallback,
and dstSize_tooSmall paths through the block splitter and superblock.
Test plan:
- cd rust && cargo fmt --check && cargo clippy --all-targets -- -D
warnings && cargo test --all-targets (131 tests: new reference
vectors for seqToCodes, RLE table headers, empty-seqStore repeat
copies, literal-stats type selection, and repcode resolution in
copyBlockSequences)
- cargo build --release --no-default-features --features compression
and --features decompression
- make -C tests fuzzer && ./tests/fuzzer -i1 --no-big-tests (covers
ZSTD_generateSequences and ZDICT training over the Rust seqToCodes)
- make -C tests test-rust-lib-smoke
- ./tests/zstreamtest -i1 and ./tests/zstreamtest --newapi -t1 -i1
- ./tests/decodecorpus -t -T1s (decodecorpus links Rust seqToCodes)
- /tmp/ref vs /tmp/got frame diff as described above: byte-identical
Move subblock partitioning, literal and sequence section emission, entropy
fallback, and repcode repair into Rust. Keep a narrow C wrapper to read the
opaque compression context and invoke the existing C entropy-stat builder;
the Rust side owns only verified C-shaped leaf layouts.
Test Plan:
- cargo clippy
- cargo clippy --benches
- cargo clippy --tests
- cargo +nightly fmt
- cargo test zstd_compress_superblock::tests -- --nocapture
- cargo test --target i686-unknown-linux-gnu zstd_compress_superblock::tests
- byte-exact C control across 16 target-size and input-shape cases
- fuzzer, zstreamtest, invalidDictionaries, and bounded native test matrix
Refs: Rust sequence entropy port 887af204
and only defines a sub-block boundary when
it believes that it is compressible.
It's effectively an optimization,
avoiding a compression cycle to reach the same conclusion.
note that the size of individual compressed blocks will vary more wildly with this modification.
But it seems good enough for a first test, and fix the speed regression issue.
Further refinements can be attempted later.
and in `decodecorpus`:
the specific case `nbSeq=127` can be represented using the 1-byte format.
Note that both the 1-byte and the 2-bytes formats are valid to represent this case,
so there was no "error", produced data remains valid,
it's just that the 1-byte format is more efficient.
fix#3667
Credit to @ip7z for finding this issue.
```
for f in $(find . \( -path ./.git -o -path ./tests/fuzz/corpora \) -prune -o -type f);
do
sed -i 's/Facebook, Inc\./Meta Platforms, Inc. and affiliates./' $f;
done
```
the new contracts seems to make more sense :
updateRep() updates an array of repeat offsets _in place_,
while newRep() generates a new structure with the updated repeat-offset array.
Most callers are actually expecting the in-place variant,
and a limited sub-section, in `zstd_opt.c` mainly, prefer `newRep()`.
This saves some 1.7Kb in rodata section (x86_64, zstd tool),
while assembler code stays the same except
the type of a few load/extend instructions.
Should not have negative performance implications.
* Switch to yearless copyright per FB policy
* Fix up SPDX-License-Identifier lines in `contrib/linux-kernel` sources
* Add zstd copyright/license header to the `contrib/linux-kernel` sources
* Update the `tests/test-license.py` to check for yearless copyright
* Improvements to `tests/test-license.py`
* Check `contrib/linux-kernel` in `tests/test-license.py`
ddidderr
Yann Collet