structured-zstd 0.0.15

//! Structures and utilities used for compressing/encoding data into the Zstd format.

pub(crate) mod block_header;
pub(crate) mod blocks;
pub(crate) mod frame_header;
pub(crate) mod match_generator;
pub(crate) mod util;

mod frame_compressor;
mod levels;
mod streaming_encoder;
pub use frame_compressor::FrameCompressor;
pub use match_generator::MatchGeneratorDriver;
pub use streaming_encoder::StreamingEncoder;

use crate::io::{Read, Write};
use alloc::vec::Vec;

/// Convenience function to compress some source into a target without reusing any resources of the compressor
/// ```rust
/// use structured_zstd::encoding::{compress, CompressionLevel};
/// let data: &[u8] = &[0,0,0,0,0,0,0,0,0,0,0,0];
/// let mut target = Vec::new();
/// compress(data, &mut target, CompressionLevel::Fastest);
/// ```
pub fn compress<R: Read, W: Write>(source: R, target: W, level: CompressionLevel) {
    let mut frame_enc = FrameCompressor::new(level);
    frame_enc.set_source(source);
    frame_enc.set_drain(target);
    frame_enc.compress();
}

/// Convenience function to compress some source into a Vec without reusing any resources of the compressor.
///
/// This helper eagerly buffers the full input (`Read`) before compression so it
/// can provide a source-size hint to the one-shot encoder path. Peak memory can
/// therefore be roughly `input_size + output_size`. For very large payloads or
/// tighter memory budgets, prefer streaming APIs such as [`StreamingEncoder`].
/// ```rust
/// use structured_zstd::encoding::{compress_to_vec, CompressionLevel};
/// let data: &[u8] = &[0,0,0,0,0,0,0,0,0,0,0,0];
/// let compressed = compress_to_vec(data, CompressionLevel::Fastest);
/// ```
pub fn compress_to_vec<R: Read>(source: R, level: CompressionLevel) -> Vec<u8> {
    let mut source = source;
    let mut input = Vec::new();
    source.read_to_end(&mut input).unwrap();

    let mut vec = Vec::new();
    let mut frame_enc = FrameCompressor::new(level);
    frame_enc.set_source_size_hint(input.len() as u64);
    frame_enc.set_source(input.as_slice());
    frame_enc.set_drain(&mut vec);
    frame_enc.compress();
    vec
}

/// The compression mode used impacts the speed of compression,
/// and resulting compression ratios. Faster compression will result
/// in worse compression ratios, and vice versa.
#[derive(Copy, Clone, Debug)]
pub enum CompressionLevel {
    /// This level does not compress the data at all, and simply wraps
    /// it in a Zstandard frame.
    Uncompressed,
    /// This level is roughly equivalent to Zstd compression level 1
    Fastest,
    /// This level uses the crate's dedicated `dfast`-style matcher to
    /// target a better speed/ratio tradeoff than [`CompressionLevel::Fastest`].
    ///
    /// It represents this crate's "default" compression setting and may
    /// evolve in future versions as the implementation moves closer to
    /// reference zstd level 3 behavior.
    Default,
    /// This level is roughly equivalent to Zstd level 7.
    ///
    /// Uses the hash-chain matcher with a lazy2 matching strategy: the encoder
    /// evaluates up to two positions ahead before committing to a match,
    /// trading speed for a better compression ratio than [`CompressionLevel::Default`].
    Better,
    /// This level is roughly equivalent to Zstd level 11.
    ///
    /// Uses the hash-chain matcher with a deep lazy2 matching strategy and
    /// a 16 MiB window. Compared to [`CompressionLevel::Better`], this level
    /// uses larger hash and chain tables (2 M / 1 M entries vs 1 M / 512 K),
    /// a deeper search (32 candidates vs 16), and a higher target match
    /// length (128 vs 48), trading speed for the best compression ratio
    /// available in this crate.
    Best,
    /// Numeric compression level.
    ///
    /// Levels 1–22 correspond to the C zstd level numbering.  Higher values
    /// produce smaller output at the cost of more CPU time.  Negative values
    /// select ultra-fast modes that trade ratio for speed.  Level 0 is
    /// treated as [`DEFAULT_LEVEL`](Self::DEFAULT_LEVEL), matching C zstd
    /// semantics.
    ///
    /// Named variants map to specific numeric levels:
    /// [`Fastest`](Self::Fastest) = 1, [`Default`](Self::Default) = 3,
    /// [`Better`](Self::Better) = 7, [`Best`](Self::Best) = 11.
    /// [`Best`](Self::Best) remains the highest-ratio named preset, but
    /// [`Level`](Self::Level) values above 11 can target stronger (slower)
    /// tuning than the named hierarchy.
    ///
    /// Levels above 11 use progressively larger windows and deeper search
    /// with the lazy2 hash-chain backend.  Levels that require strategies
    /// this crate has not yet implemented (btopt, btultra) are approximated
    /// with the closest available matcher.
    ///
    /// Semver note: this variant was added after the initial enum shape and
    /// is a breaking API change for downstream crates that exhaustively
    /// `match` on [`CompressionLevel`] without a wildcard arm.
    Level(i32),
}

impl CompressionLevel {
    /// The minimum supported numeric compression level (ultra-fast mode).
    pub const MIN_LEVEL: i32 = -131072;
    /// The maximum supported numeric compression level.
    pub const MAX_LEVEL: i32 = 22;
    /// The default numeric compression level (equivalent to [`Default`](Self::Default)).
    pub const DEFAULT_LEVEL: i32 = 3;

    /// Create a compression level from a numeric value.
    ///
    /// Returns named variants for canonical levels (`0`/`3`, `1`, `7`, `11`)
    /// and [`Level`](Self::Level) for all other values.
    ///
    /// With the default matcher backend (`MatchGeneratorDriver`), values
    /// outside [`MIN_LEVEL`](Self::MIN_LEVEL)..=[`MAX_LEVEL`](Self::MAX_LEVEL)
    /// are silently clamped during built-in level parameter resolution.
    pub const fn from_level(level: i32) -> Self {
        match level {
            0 | Self::DEFAULT_LEVEL => Self::Default,
            1 => Self::Fastest,
            7 => Self::Better,
            11 => Self::Best,
            _ => Self::Level(level),
        }
    }
}

/// Trait used by the encoder that users can use to extend the matching facilities with their own algorithm
/// making their own tradeoffs between runtime, memory usage and compression ratio
///
/// This trait operates on buffers that represent the chunks of data the matching algorithm wants to work on.
/// Each one of these buffers is referred to as a *space*. One or more of these buffers represent the window
/// the decoder will need to decode the data again.
///
/// This library asks the Matcher for a new buffer using `get_next_space` to allow reusing of allocated buffers when they are no longer part of the
/// window of data that is being used for matching.
///
/// The library fills the buffer with data that is to be compressed and commits them back to the matcher using `commit_space`.
///
/// Then it will either call `start_matching` or, if the space is deemed not worth compressing, `skip_matching` is called.
///
/// This is repeated until no more data is left to be compressed.
pub trait Matcher {
    /// Get a space where we can put data to be matched on. Will be encoded as one block. The maximum allowed size is 128 kB.
    fn get_next_space(&mut self) -> alloc::vec::Vec<u8>;
    /// Get a reference to the last commited space
    fn get_last_space(&mut self) -> &[u8];
    /// Commit a space to the matcher so it can be matched against
    fn commit_space(&mut self, space: alloc::vec::Vec<u8>);
    /// Just process the data in the last commited space for future matching
    fn skip_matching(&mut self);
    /// Process the data in the last commited space for future matching AND generate matches for the data
    fn start_matching(&mut self, handle_sequence: impl for<'a> FnMut(Sequence<'a>));
    /// Reset this matcher so it can be used for the next new frame
    fn reset(&mut self, level: CompressionLevel);
    /// Provide a hint about the total uncompressed size for the next frame.
    ///
    /// Implementations may use this to select smaller hash tables and windows
    /// for small inputs, matching the C zstd source-size-class behavior.
    /// Called before [`reset`](Self::reset) when the caller knows the input
    /// size (e.g. from pledged content size or file metadata).
    ///
    /// The default implementation is a no-op for custom matchers and
    /// test stubs. The built-in runtime matcher (`MatchGeneratorDriver`)
    /// overrides this hook and applies the hint during level resolution.
    fn set_source_size_hint(&mut self, _size: u64) {}
    /// Prime matcher state with dictionary history before compressing the next frame.
    /// Default implementation is a no-op for custom matchers that do not support this.
    fn prime_with_dictionary(&mut self, _dict_content: &[u8], _offset_hist: [u32; 3]) {}
    /// Returns whether this matcher can consume dictionary priming state and produce
    /// dictionary-dependent sequences. Defaults to `false` for custom matchers.
    fn supports_dictionary_priming(&self) -> bool {
        false
    }
    /// The size of the window the decoder will need to execute all sequences produced by this matcher.
    ///
    /// Must return a positive (non-zero) value; returning 0 causes
    /// [`StreamingEncoder`] to reject the first write with an invalid-input error
    /// (`InvalidInput` with `std`, `Other` with `no_std`).
    ///
    /// Must remain stable for the lifetime of a frame.
    /// It may change only after `reset()` is called for the next frame
    /// (for example because the compression level changed).
    fn window_size(&self) -> u64;
}

#[derive(PartialEq, Eq, Debug)]
/// Sequences that a [`Matcher`] can produce
pub enum Sequence<'data> {
    /// Is encoded as a sequence for the decoder sequence execution.
    ///
    /// First the literals will be copied to the decoded data,
    /// then `match_len` bytes are copied from `offset` bytes back in the decoded data
    Triple {
        literals: &'data [u8],
        offset: usize,
        match_len: usize,
    },
    /// This is returned as the last sequence in a block
    ///
    /// These literals will just be copied at the end of the sequence execution by the decoder
    Literals { literals: &'data [u8] },
}