commit 8f23ec711fd63e7cc9f22ee53a913258bafdb9bc
Author: Dennis Ranke <dennis.ranke@gmail.com>
Date:   Fri Nov 19 21:40:37 2021 +0100

    first version, only very simple greedy packer

diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..ea8c4bf
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1 @@
+/target
diff --git a/Cargo.lock b/Cargo.lock
new file mode 100644
index 0000000..c241a73
--- /dev/null
+++ b/Cargo.lock
@@ -0,0 +1,50 @@
+# This file is automatically @generated by Cargo.
+# It is not intended for manual editing.
+version = 3
+
+[[package]]
+name = "autocfg"
+version = "1.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "cdb031dd78e28731d87d56cc8ffef4a8f36ca26c38fe2de700543e627f8a464a"
+
+[[package]]
+name = "cc"
+version = "1.0.72"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "22a9137b95ea06864e018375b72adfb7db6e6f68cfc8df5a04d00288050485ee"
+
+[[package]]
+name = "cdivsufsort"
+version = "2.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "edefce019197609da416762da75bb000bbd2224b2d89a7e722c2296cbff79b8c"
+dependencies = [
+ "cc",
+ "sacabase",
+]
+
+[[package]]
+name = "num-traits"
+version = "0.2.14"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9a64b1ec5cda2586e284722486d802acf1f7dbdc623e2bfc57e65ca1cd099290"
+dependencies = [
+ "autocfg",
+]
+
+[[package]]
+name = "sacabase"
+version = "2.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9883fc3d6ce3d78bb54d908602f8bc1f7b5f983afe601dabe083009d86267a84"
+dependencies = [
+ "num-traits",
+]
+
+[[package]]
+name = "upkr"
+version = "0.1.0"
+dependencies = [
+ "cdivsufsort",
+]
diff --git a/Cargo.toml b/Cargo.toml
new file mode 100644
index 0000000..1bb3bfa
--- /dev/null
+++ b/Cargo.toml
@@ -0,0 +1,9 @@
+[package]
+name = "upkr"
+version = "0.1.0"
+edition = "2021"
+
+# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
+
+[dependencies]
+cdivsufsort = "2"
diff --git a/src/context_state.rs b/src/context_state.rs
new file mode 100644
index 0000000..81cb3e0
--- /dev/null
+++ b/src/context_state.rs
@@ -0,0 +1,39 @@
+const INIT_PROB: u16 = 0x8000;
+const UPDATE_RATE: u32 = 4;
+
+#[derive(Clone)]
+pub struct ContextState {
+    contexts: Vec<u16>,
+}
+
+pub struct Context<'a> {
+    state: &'a mut ContextState,
+    index: usize,
+}
+
+impl ContextState {
+    pub fn new(size: usize) -> ContextState {
+        ContextState {
+            contexts: vec![INIT_PROB; size],
+        }
+    }
+
+    pub fn context_mut(&mut self, index: usize) -> Context {
+        Context { state: self, index }
+    }
+}
+
+impl<'a> Context<'a> {
+    pub fn prob(&self) -> u16 {
+        self.state.contexts[self.index]
+    }
+
+    pub fn update(&mut self, bit: bool) {
+        let old = self.state.contexts[self.index];
+        self.state.contexts[self.index] = if bit {
+            old + (((1 << 16) - old as u32) >> UPDATE_RATE) as u16
+        } else {
+            old - (old >> UPDATE_RATE)
+        };
+    }
+}
diff --git a/src/greedy_packer.rs b/src/greedy_packer.rs
new file mode 100644
index 0000000..684b719
--- /dev/null
+++ b/src/greedy_packer.rs
@@ -0,0 +1,44 @@
+use crate::lz::LzCoder;
+use crate::match_finder::MatchFinder;
+
+pub fn pack(data: &[u8]) -> Vec<u8> {
+    let match_finder = MatchFinder::new(data);
+    let mut lz = LzCoder::new();
+
+    let mut pos = 0;
+    while pos < data.len() {
+        let mut encoded_match = false;
+        if let Some(m) = match_finder.matches(pos).next() {
+            let max_offset = 1 << (m.length * 3 - 1).min(31);
+            let offset = pos - m.pos;
+            if offset < max_offset {
+                lz.encode_match(offset, m.length);
+                pos += m.length;
+                encoded_match = true;
+            }
+        }
+
+        if !encoded_match {
+            let offset = lz.last_offset();
+            if offset != 0 {
+                let length = data[pos..]
+                    .iter()
+                    .zip(data[(pos - offset)..].iter())
+                    .take_while(|(a, b)| a == b)
+                    .count();
+                if length > 0 {
+                    lz.encode_match(offset, length);
+                    pos += length;
+                    encoded_match = true;
+                }
+            }
+        }
+
+        if !encoded_match {
+            lz.encode_literal(data[pos]);
+            pos += 1;
+        }
+    }
+
+    lz.finish()
+}
diff --git a/src/lz.rs b/src/lz.rs
new file mode 100644
index 0000000..f220a28
--- /dev/null
+++ b/src/lz.rs
@@ -0,0 +1,116 @@
+use crate::context_state::ContextState;
+use crate::range_coder::{RangeCoder, RangeDecoder};
+
+pub struct LzCoder {
+    contexts: ContextState,
+    range_coder: RangeCoder,
+    last_offset: usize,
+}
+
+impl LzCoder {
+    pub fn new() -> LzCoder {
+        LzCoder {
+            contexts: ContextState::new(1 + 255 + 1 + 64 + 64),
+            range_coder: RangeCoder::new(),
+            last_offset: 0,
+        }
+    }
+
+    pub fn encode_literal(&mut self, byte: u8) {
+        self.bit(false, 0);
+        let mut context_index = 1;
+        for i in (0..8).rev() {
+            let bit = (byte >> i) & 1 != 0;
+            self.bit(bit, context_index);
+            context_index = (context_index << 1) | bit as usize;
+        }
+    }
+
+    pub fn encode_match(&mut self, offset: usize, length: usize) {
+        self.bit(true, 0);
+        if offset != self.last_offset {
+            self.last_offset = offset;
+            self.bit(true, 256);
+            self.length(offset + 1, 257);
+        } else {
+            self.bit(false, 256);
+        }
+        self.length(length, 257 + 64);
+    }
+
+    pub fn finish(mut self) -> Vec<u8> {
+        self.bit(true, 0);
+        self.bit(true, 256);
+        self.length(1, 257);
+        self.range_coder.finish()
+    }
+
+    pub fn last_offset(&self) -> usize {
+        self.last_offset
+    }
+
+    fn length(&mut self, value: usize, context_start: usize) {
+        assert!(value >= 1);
+        let top_bit = usize::BITS - 1 - value.leading_zeros();
+        let mut context_index = context_start;
+        for i in (0..top_bit).rev() {
+            self.bit(true, context_index);
+            self.bit((value >> i) & 1 != 0, context_index + 1);
+            context_index += 2;
+        }
+        self.bit(false, context_index);
+    }
+
+    fn bit(&mut self, b: bool, context_index: usize) {
+        self.range_coder
+            .encode_with_context(b, &mut self.contexts.context_mut(context_index));
+    }
+}
+
+pub fn unpack(packed_data: &[u8]) -> Vec<u8> {
+    let mut decoder = RangeDecoder::new(packed_data);
+    let mut contexts = ContextState::new(1 + 255 + 1 + 64 + 64);
+    let mut result = vec![];
+    let mut offset = 0;
+
+    fn decode_length(
+        decoder: &mut RangeDecoder,
+        contexts: &mut ContextState,
+        mut context_index: usize,
+    ) -> usize {
+        let mut length = 1;
+        while decoder.decode_with_context(&mut contexts.context_mut(context_index)) {
+            length = (length << 1)
+                | decoder.decode_with_context(&mut contexts.context_mut(context_index + 1))
+                    as usize;
+            context_index += 2;
+        }
+        length
+    }
+
+    loop {
+        if decoder.decode_with_context(&mut contexts.context_mut(0)) {
+            if decoder.decode_with_context(&mut contexts.context_mut(256)) {
+                offset = decode_length(&mut decoder, &mut contexts, 257) - 1;
+                if offset == 0 {
+                    break;
+                }
+            }
+            let length = decode_length(&mut decoder, &mut contexts, 257 + 64);
+            for _ in 0..length {
+                result.push(result[result.len() - offset]);
+            }
+        } else {
+            let mut context_index = 1;
+            let mut byte = 0;
+            for i in (0..8).rev() {
+                let bit = decoder.decode_with_context(&mut contexts.context_mut(context_index));
+                context_index = (context_index << 1) | bit as usize;
+                byte |= (bit as u8) << i;
+            }
+            result.push(byte);
+        }
+    }
+
+    result
+}
diff --git a/src/main.rs b/src/main.rs
new file mode 100644
index 0000000..82705bb
--- /dev/null
+++ b/src/main.rs
@@ -0,0 +1,16 @@
+mod context_state;
+mod greedy_packer;
+mod lz;
+mod match_finder;
+mod range_coder;
+
+fn main() {
+    let test_data = include_bytes!("../testcases/skipahead.wasm");
+
+    let packed = greedy_packer::pack(test_data);
+    dbg!((test_data.len(), packed.len()));
+
+    let unpacked = lz::unpack(&packed);
+    dbg!(unpacked.len());
+    assert!(test_data == unpacked.as_slice());
+}
diff --git a/src/match_finder.rs b/src/match_finder.rs
new file mode 100644
index 0000000..e576019
--- /dev/null
+++ b/src/match_finder.rs
@@ -0,0 +1,158 @@
+use std::collections::BinaryHeap;
+use std::ops::Range;
+
+pub struct MatchFinder {
+    suffixes: Vec<i32>,
+    rev_suffixes: Vec<u32>,
+    lcp: Vec<u32>,
+
+    max_matches: usize,
+    patience: usize,
+    max_length_diff: usize,
+}
+
+impl MatchFinder {
+    pub fn new(data: &[u8]) -> MatchFinder {
+        let mut suffixes = vec![0i32; data.len()];
+        cdivsufsort::sort_in_place(data, &mut suffixes);
+
+        let mut rev_suffixes = vec![0u32; data.len()];
+        for (suffix_index, index) in suffixes.iter().enumerate() {
+            rev_suffixes[*index as usize] = suffix_index as u32;
+        }
+
+        let mut lcp = vec![0u32; data.len()];
+        let mut length = 0usize;
+        for suffix_index in &rev_suffixes {
+            if *suffix_index as usize + 1 < suffixes.len() {
+                let i = suffixes[*suffix_index as usize] as usize;
+                let j = suffixes[*suffix_index as usize + 1] as usize;
+                while i + length < data.len()
+                    && j + length < data.len()
+                    && data[i + length] == data[j + length]
+                {
+                    length += 1;
+                }
+                lcp[*suffix_index as usize] = length as u32;
+            }
+            length = length.saturating_sub(1);
+        }
+
+        MatchFinder {
+            suffixes,
+            rev_suffixes,
+            lcp,
+            max_matches: 10,
+            patience: 10,
+            max_length_diff: 2,
+        }
+    }
+
+    pub fn matches(&self, pos: usize) -> Matches {
+        let index = self.rev_suffixes[pos] as usize;
+        let mut matches = Matches {
+            finder: self,
+            pos_range: 0..pos,
+            left_index: index,
+            left_length: usize::MAX,
+            right_index: index,
+            right_length: usize::MAX,
+            current_length: 0,
+            patience_left: 0,
+            matches_left: self.max_matches,
+            max_length: 0,
+            queue: BinaryHeap::new(),
+        };
+
+        matches.move_left();
+        matches.move_right();
+
+        matches
+    }
+}
+
+pub struct Matches<'a> {
+    finder: &'a MatchFinder,
+    pos_range: Range<usize>,
+    left_index: usize,
+    left_length: usize,
+    right_index: usize,
+    right_length: usize,
+    current_length: usize,
+    patience_left: usize,
+    matches_left: usize,
+    max_length: usize,
+    queue: BinaryHeap<usize>,
+}
+
+#[derive(Debug)]
+pub struct Match {
+    pub pos: usize,
+    pub length: usize,
+}
+
+impl<'a> Iterator for Matches<'a> {
+    type Item = Match;
+
+    fn next(&mut self) -> Option<Match> {
+        if self.queue.is_empty() {
+            self.current_length = self.left_length.max(self.right_length);
+            self.max_length = self.max_length.max(self.current_length);
+            if self.current_length < 2
+                || self.current_length + self.finder.max_length_diff < self.max_length
+            {
+                return None;
+            }
+            self.patience_left = self.finder.patience;
+            while self.matches_left > 0
+                && self.patience_left > 0
+                && (self.left_length == self.current_length
+                    || self.right_length == self.current_length)
+            {
+                if self.left_length == self.current_length {
+                    self.add_to_queue(self.finder.suffixes[self.left_index]);
+                    self.move_left();
+                }
+                if self.right_length == self.current_length {
+                    self.add_to_queue(self.finder.suffixes[self.right_index]);
+                    self.move_right();
+                }
+            }
+        }
+
+        self.queue.pop().map(|pos| Match {
+            pos,
+            length: self.current_length,
+        })
+    }
+}
+
+impl<'a> Matches<'a> {
+    fn move_left(&mut self) {
+        if self.left_index > 0 {
+            self.left_index -= 1;
+            self.left_length = self
+                .left_length
+                .min(self.finder.lcp[self.left_index] as usize);
+        } else {
+            self.left_length = 0;
+        }
+    }
+
+    fn move_right(&mut self) {
+        self.right_index += 1;
+        self.right_length = self
+            .right_length
+            .min(self.finder.lcp[self.right_index - 1] as usize);
+    }
+
+    fn add_to_queue(&mut self, pos: i32) {
+        if self.pos_range.contains(&(pos as usize)) {
+            self.queue.push(pos as usize);
+            self.matches_left -= 1;
+            self.patience_left = self.finder.patience;
+        } else {
+            self.patience_left = 0;
+        }
+    }
+}
diff --git a/src/range_coder.rs b/src/range_coder.rs
new file mode 100644
index 0000000..89ea633
--- /dev/null
+++ b/src/range_coder.rs
@@ -0,0 +1,111 @@
+use crate::context_state::Context;
+
+pub struct RangeCoder {
+    buffer: Vec<u8>,
+    low: u64,
+    range: u64,
+}
+
+const TOTAL: u32 = 65536;
+
+impl RangeCoder {
+    pub fn new() -> RangeCoder {
+        RangeCoder {
+            buffer: vec![],
+            low: 0,
+            range: 1 << 40,
+        }
+    }
+
+    pub fn encode_with_context(&mut self, bit: bool, context: &mut Context) {
+        self.encode_bit(bit, context.prob() as u32);
+        context.update(bit);
+    }
+
+    pub fn encode_bit(&mut self, bit: bool, prob: u32) {
+        let (start, size) = if bit { (0, prob) } else { (prob, TOTAL - prob) };
+        self.range /= TOTAL as u64;
+        self.low += start as u64 * self.range;
+        self.range *= size as u64;
+
+        while (self.low >> 32) == (self.low + self.range - 1) >> 32 {
+            self.emit_byte();
+        }
+
+        if self.range < 1 << 24 {
+            self.emit_byte();
+            self.emit_byte();
+            self.range = (1 << 40) - self.low;
+        }
+    }
+
+    pub fn finish(mut self) -> Vec<u8> {
+        while self.range < 1 << 32 {
+            self.emit_byte();
+        }
+        self.low += 1 << 32;
+        self.emit_byte();
+        self.buffer
+    }
+
+    fn emit_byte(&mut self) {
+        self.buffer.push((self.low >> 32).try_into().unwrap());
+        self.low = (self.low & 0xffffffff) << 8;
+        self.range *= 256;
+    }
+}
+
+pub struct RangeDecoder<'a> {
+    data: &'a [u8],
+    code: u64,
+    low: u64,
+    range: u64,
+}
+
+impl<'a> RangeDecoder<'a> {
+    pub fn new(data: &'a [u8]) -> RangeDecoder<'a> {
+        RangeDecoder {
+            data,
+            code: 0,
+            low: 0,
+            range: 1,
+        }
+    }
+
+    pub fn decode_with_context(&mut self, context: &mut Context) -> bool {
+        let bit = self.decode_bit(context.prob() as u32);
+        context.update(bit);
+        bit
+    }
+
+    pub fn decode_bit(&mut self, prob: u32) -> bool {
+        while self.low >> 32 == (self.low + self.range - 1) >> 32 {
+            self.append_byte();
+        }
+
+        if self.range < 1 << 24 {
+            self.append_byte();
+            self.append_byte();
+            self.range = (1 << 40) - self.low;
+        }
+
+        let bit = (self.code - self.low) / (self.range / TOTAL as u64) < prob as u64;
+
+        let (start, size) = if bit { (0, prob) } else { (prob, TOTAL - prob) };
+        self.range /= TOTAL as u64;
+        self.low += start as u64 * self.range;
+        self.range *= size as u64;
+
+        bit
+    }
+
+    fn append_byte(&mut self) {
+        self.code = (self.code & 0xffffffff) << 8;
+        if !self.data.is_empty() {
+            self.code |= self.data[0] as u64;
+            self.data = &self.data[1..];
+        }
+        self.low = (self.low & 0xffffffff) << 8;
+        self.range <<= 8;
+    }
+}