structured_zstd/

lib.rs

//! Pure-Rust Zstandard codec with a production-grade decoder, dictionary
//! handle reuse, and an actively-improved encoder.
//!
//! The crate ships:
//!
//! * [`decoding`] — [RFC 8878] decoder ([`decoding::StreamingDecoder`],
//!   [`decoding::FrameDecoder`], dictionary-backed paths via
//!   [`decoding::DictionaryHandle`]).
//! * [`encoding`] — frame compressor, streaming encoder, named and numeric
//!   compression levels ([`encoding::CompressionLevel`]).
//! * [`dictionary`] (feature `dict-builder`) — COVER / FastCOVER training
//!   plus raw-to-finalized dictionary helpers.
//! * `decoding::workspace` (feature `caller-workspace`): streaming decoder
//!   that runs entirely in caller-owned memory, never allocating, with
//!   caller-set limits on window, output and work.
//! * v0.5, v0.6 and v0.7 frames (feature `legacy`, on by default): the
//!   formats every libzstd 1.5.7 build reads by default decode through the
//!   same entry points, the frame queries and the streaming decoder
//!   included; `decoding::LegacyVersion` names them. Without the feature
//!   they are refused as an unknown format.
//!
//! No FFI, no cmake, no system zstd. `no_std` builds are supported by
//! disabling the default `std` feature.
//!
//! # CPU kernel features
//!
//! Both the decode and the encode hot paths ship per-CPU-tier SIMD kernels.
//! On x86 and aarch64 with `std` the tier is chosen at runtime (CPU-feature
//! detection, cached on first use); on `no_std` it is chosen at compile time
//! from `cfg(target_feature)`. WebAssembly is compile-time only either way:
//! its kernels additionally require `target_feature = "simd128"`, so a wasm
//! build without `-C target-feature=+simd128` stays scalar.
//!
//! Each tier is gated by a cargo feature: `kernel-scalar`, `kernel-sse`,
//! `kernel-bmi2`, `kernel-avx2`, `kernel-vbmi2` (x86) and `kernel-neon`,
//! `kernel-sve` (aarch64). All are on by default except `kernel-vbmi2`, which
//! is opt-in because the AVX-512 decode tier measures slower than AVX2 on the
//! bursty decode path, so the default build is a universal binary that picks
//! the best available tier per the above. `kernel-scalar` gates no code: the
//! scalar path is the mandatory fallback and is always compiled, so the flag
//! exists only to name that tier explicitly in a feature set. `kernel-vbmi2`
//! and `kernel-sve` are decoder-only; the encoder has no AVX-512 or SVE tier.
//! The chain mirrors the ISA
//! dependency (`kernel-avx2` implies `kernel-bmi2` implies `kernel-sse`;
//! `kernel-sve` implies `kernel-neon`). `kernel-sse` covers two x86 tiers:
//! SSE4.2 where the CPU has it, and a plain-SSE2 tier otherwise, so a
//! pre-SSE4.2 CPU still gets vector match compares. Any subset is valid, and a
//! flag is inert on architectures it doesn't apply to. Constrained targets can
//! shrink the binary by trimming
//! tiers: `--no-default-features --features kernel-scalar` compiles out every
//! per-tier dispatch, the BMI2/AVX2/VBMI2/NEON trampolines, and the explicit
//! SSE2/NEON intrinsics in both the copy primitives and the encoder
//! match-finder. The `kernel_*` features control the crate's own explicit
//! SIMD; they do not constrain the compiler's autovectorizer, which may still
//! emit vector instructions from ordinary scalar code regardless of the
//! enabled tiers.
//!
//! The packaged README is included below for the docs.rs landing page; the
//! API anchors above link straight into the per-module documentation.
//!
//! [RFC 8878]: https://www.rfc-editor.org/rfc/rfc8878
// Keep crate docs aligned with the packaged README via the crate-local symlink in `zstd/README.md`.
#![doc = include_str!("../README.md")]
#![no_std]
#![deny(trivial_casts, trivial_numeric_casts, rust_2018_idioms)]
#![cfg_attr(docsrs, feature(doc_cfg))]

#[cfg(feature = "std")]
extern crate std;

#[cfg(not(feature = "rustc-dep-of-std"))]
extern crate alloc;

#[cfg(feature = "std")]
pub(crate) const VERBOSE: bool = false;

macro_rules! vprintln {
    ($($x:expr),*) => {
        #[cfg(feature = "std")]
        if crate::VERBOSE {
            std::println!($($x),*);
        }
    }
}

mod bit_io;
mod common;
/// Largest block the format allows (RFC 8878 3.1.1.2.3, 128 KiB): the single
/// source of truth for every surface that has to name it, the C ABI included.
/// A frame's own maximum is the smaller of this and its window.
pub use common::MAX_BLOCK_SIZE;
/// Smallest accepted maximum block size (the `ZSTD_BLOCKSIZE_MAX_MIN` bound),
/// shared by `set_max_block_size` and the C ABI parameter surface.
pub use common::MIN_MAX_BLOCK_SIZE;
/// Smallest accepted block-size target (the `ZSTD_TARGETCBLOCKSIZE_MIN`
/// bound): the single source of truth shared by the Rust setters
/// (`set_target_block_size`) and the C ABI parameter surface.
pub use common::MIN_TARGET_BLOCK_SIZE;
mod cpu_kernel;
pub mod decoding;
#[cfg(feature = "dict-builder")]
#[cfg_attr(docsrs, doc(cfg(feature = "dict-builder")))]
pub mod dictionary;
pub mod encoding;
mod histogram;

pub mod skippable;

pub(crate) mod blocks;

#[cfg(feature = "fuzz-exports")]
pub mod fse;
#[cfg(feature = "fuzz-exports")]
pub mod huff0;

// `pub fn init_state<K: CpuKernel>` and friends inside the
// fuzz-exports-public `huff0` module name `crate::cpu_kernel::CpuKernel`
// in their signatures. Without a publicly-reachable path to `CpuKernel`
// the bound triggers `private_bounds` / `private_interfaces`. Re-export
// under the same feature gate so the fuzz harness build is clean.
#[cfg(feature = "fuzz-exports")]
pub use crate::cpu_kernel::{CpuKernel, ScalarKernel};

/// Name of the active CPU kernel tier (entropy / sequence hot paths) for this
/// process — for diagnostics and benchmark/dashboard reporting. See
/// [`cpu_kernel::active_cpu_kernel_name`].
pub use crate::cpu_kernel::active_cpu_kernel_name;
/// The ceiling on the instruction sets the kernels may use: see
/// [`set_cpu_ceiling`].
pub use crate::cpu_kernel::{
    CpuCeilingError, CpuLevel, UnknownCpuLevel, cpu_ceiling, set_cpu_ceiling,
};
/// Name of the match-finder kernel tier for this process. See
/// [`encoding::active_match_kernel_name`].
pub use crate::encoding::active_match_kernel_name;

#[cfg(not(feature = "fuzz-exports"))]
pub(crate) mod fse;
#[cfg(not(feature = "fuzz-exports"))]
pub(crate) mod huff0;

#[cfg(feature = "std")]
pub mod io_std;

#[cfg(feature = "std")]
pub use io_std as io;

#[cfg(not(feature = "std"))]
pub mod io_nostd;

#[cfg(not(feature = "std"))]
pub use io_nostd as io;

#[cfg(test)]
mod tests;

/// Re-exports of internal types used by benchmarks.
///
/// Gated behind the `bench-internals` feature so normal builds do not
/// widen the public API surface. Not part of the stable API; items may
/// change or disappear without notice.
#[cfg(feature = "bench-internals")]
#[doc(hidden)]
pub mod testing {
    /// Compression parameters selected for `(level, srcSize, dictSize)` →
    /// `(windowLog, chainLog, hashLog, searchLog, minMatch, targetLength,
    /// strategy)`. Facade for the `ffi-bench` parity test that diffs the
    /// selection against the reference `ZSTD_getCParams`.
    pub fn compression_params(
        level: i32,
        src: u64,
        dict: usize,
    ) -> (u32, u32, u32, u32, u32, u32, u32) {
        // `ZSTD_getCParams` spells an unknown source size as 0.
        let cp =
            crate::encoding::LevelParameters::for_level(level, (src != 0).then_some(src), dict);
        (
            cp.window_log,
            cp.chain_log,
            cp.hash_log,
            cp.search_log,
            cp.min_match,
            cp.target_length,
            cp.strategy.ordinal(),
        )
    }

    /// Force every HUF table build onto the cheap single-build path (skip the
    /// #167 table-log search) so a bench harness can A/B the search across
    /// levels. Measurement-only.
    pub fn set_force_cheap_huf(on: bool) {
        crate::huff0::huff0_encoder::set_force_cheap_huf(on);
    }

    pub use crate::bit_io::BitReaderReversed;
    // `BitReaderReversed` is generic over `K: CpuKernel = ScalarKernel`,
    // so both the trait bound and the default need a `pub` path to
    // match the re-exported type's visibility. Without this the
    // bench-build trips `private_bounds` / `private_interfaces`.
    pub use crate::cpu_kernel::{CpuKernel, ScalarKernel};

    /// Bench-only facade for the decoder wildcopy implementation.
    ///
    /// # Safety
    /// Caller must satisfy the same safety contract as
    /// `decoding::copy_bytes_overshooting_for_bench`.
    #[inline(always)]
    pub unsafe fn copy_bytes_overshooting_for_bench(
        src: (*const u8, usize),
        dst: (*mut u8, usize),
        copy_at_least: usize,
    ) {
        // Keep decoder internals crate-private and expose only this bench shim.
        unsafe { crate::decoding::copy_bytes_overshooting_for_bench(src, dst, copy_at_least) };
    }

    /// Maximum block size per RFC 8878 §3.1.1.2.3 (128 KiB).
    /// Exposed for parity tests that feed exactly-one-block chunks
    /// into the block-splitter comparator.
    pub const MAX_BLOCK_SIZE: u32 = crate::common::MAX_BLOCK_SIZE;

    /// Run our block splitter on a 128 KB chunk.
    ///
    /// `split_level` mirrors upstream zstd `ZSTD_splitBlock(level)`: `0` selects
    /// the borders heuristic (`ZSTD_splitBlock_fromBorders`), `1..=4`
    /// select `ZSTD_splitBlock_byChunks` at the corresponding sampling
    /// level. Returns the split position (or `block.len()` if no split).
    ///
    /// Crate-internal facade for the block-splitter parity comparator test —
    /// the underlying functions stay `fn` so they don't widen the
    /// stable API surface.
    pub fn block_splitter_decision(block: &[u8], split_level: usize) -> usize {
        crate::encoding::frame_compressor::block_splitter_decision_for_bench(block, split_level)
    }

    /// White-box capture of our Huffman weight description for `data`:
    /// `(description, weights)` where `description` is the length-prefixed
    /// FSE payload and `weights` the raw per-symbol weights. Facade for the
    /// `ffi-bench` conformance test that feeds it through the C `HUF_readStats`.
    pub fn huf_weight_description(data: &[u8]) -> (alloc::vec::Vec<u8>, alloc::vec::Vec<u8>) {
        crate::huff0::huff0_encoder::huf_weight_description_for_test(data)
    }

    /// White-box capture of our 4-stream Huffman payload for `data`. Facade for
    /// the `ffi-bench` conformance test that decodes it through the C HUF reader.
    pub fn huf_encode4x(data: &[u8]) -> alloc::vec::Vec<u8> {
        crate::huff0::huff0_encoder::huf_encode4x_for_test(data)
    }

    /// White-box capture of the level-22 sequence stream (literal-length,
    /// offset, match-length triples) our match generator emits for `data`.
    /// Facade for the sequence-conformance test in `ffi-bench`, which
    /// compares this stream against the C reference's `ZSTD_generateSequences`
    /// output. Pure Rust; the C side stays out of this crate.
    pub fn collect_level22_sequences(data: &[u8]) -> alloc::vec::Vec<(usize, usize, usize)> {
        crate::encoding::match_generator::collect_level22_sequences(data)
    }

    /// FastCOVER dictionary roundtrip fixture: `(finalized_dictionary,
    /// compressed_frame, original_payload)`. Facade for the `ffi-bench`
    /// conformance test that decodes `compressed_frame` against the dictionary
    /// through the C decoder and compares to `original_payload`.
    #[cfg(feature = "dict-builder")]
    pub fn dict_roundtrip_fixture() -> (
        alloc::vec::Vec<u8>,
        alloc::vec::Vec<u8>,
        alloc::vec::Vec<u8>,
    ) {
        crate::dictionary::dict_roundtrip_fixture()
    }

    /// The content the legacy trainer selects, before any header: facade for
    /// the `ffi-bench` test that compares it with the tail of the reference's
    /// `ZDICT_trainFromBuffer_legacy` dictionary. `None` when the trainer
    /// yields nothing.
    #[cfg(feature = "dict-builder")]
    pub fn legacy_dict_content(
        samples: &[u8],
        sample_sizes: &[usize],
        dict_size: usize,
        selectivity: u32,
    ) -> Option<alloc::vec::Vec<u8>> {
        crate::dictionary::legacy_dict_content(samples, sample_sizes, dict_size, selectivity)
    }

    pub use crate::blocks::block::BlockType;

    /// First block's type (raw / rle / compressed) in a frame. Facade over the
    /// internal block decoder for the FFI parity tests in `ffi-bench`.
    pub fn first_block_type(frame: &[u8]) -> BlockType {
        let (_, header_size) = crate::decoding::frame::read_frame_header_with_format(frame, false)
            .expect("frame header should parse");
        let mut decoder = crate::decoding::block_decoder::new();
        let (header, _) = decoder
            .read_block_header(&frame[header_size as usize..])
            .expect("block header should parse");
        header.block_type
    }

    /// `(single_segment_flag, frame_content_size, fcs_field_size_bytes)` parsed
    /// from a frame header. Facade for the FFI parity tests in `ffi-bench` so
    /// they need not reach into the internal `FrameHeader` type.
    pub fn frame_header_info(frame: &[u8]) -> (bool, u64, u8) {
        let (h, _) = crate::decoding::frame::read_frame_header_with_format(frame, false)
            .expect("frame header should parse");
        (
            h.descriptor.single_segment_flag(),
            h.frame_content_size(),
            h.descriptor.frame_content_size_bytes().unwrap_or(0),
        )
    }
}

/// SIMD wildcopy overshoot slack carried by every decoder backend
/// (currently **32 bytes**). Sized so the AVX2 chunked kernel in
/// `simd_copy::copy_bytes_overshooting` (32-byte stride on x86-64) can
/// fire on tail copies near the end of a fixed-capacity output buffer.
/// Upstream zstd's `WILDCOPY_OVERLENGTH` is also 32 bytes today; this
/// matches that contract.
///
/// Public so callers sizing an output slice for
/// [`crate::decoding::FrameDecoder::decode_all`] can size
/// `frame_content_size + WILDCOPY_OVERLENGTH` symbolically without
/// duplicating the value. Use the const reference rather than a
/// hardcoded literal — `simd_copy::copy_bytes_overshooting` already
/// ships an AVX-512 64-byte chunked kernel, and the slack may grow
/// further to reliably enable that wider kernel at buffer tails
/// (mirroring how the bump from 16 → 32 enabled the AVX2 32-byte
/// kernel at the tail).
pub const WILDCOPY_OVERLENGTH: usize = crate::decoding::buffer_backend::WILDCOPY_OVERLENGTH;