fix: patch Huffman OOB read and optimize encoders

algorithms/arithmetic/cpp/main.cpp

@@ -147,6 +147,7 @@ std::vector<uint8_t> arithmetic_encode_buffer(const std::vector<uint8_t>& input)
     }
 
     std::vector<uint8_t> out;
+    out.reserve(input.size() + compresskit::INITIAL_ENCODE_OVERHEAD);
     compresskit::write_frequency_header(out, compresskit::ARITHMETIC_MAGIC, freq);
 
     compresskit::BitWriter writer;
@@ -176,11 +177,10 @@ std::vector<uint8_t> arithmetic_decode_buffer(const std::vector<uint8_t>& input)
         throw std::runtime_error("arithmetic: invalid frequency table");
     }
 
-    std::vector<uint8_t> payload(input.begin() + pos, input.end());
-    compresskit::BitReader reader(payload);
+    std::vector<uint8_t> out;
+    compresskit::BitReader reader(input.data() + pos, input.size() - pos);
     ArithmeticDecoder decoder(reader);
 
-    std::vector<uint8_t> out;
     for (;;) {
         uint32_t sym = decoder.decode_symbol(cumulative);
         if (sym == compresskit::EOF_SYMBOL) {

algorithms/huffman/cpp/main.cpp

@@ -124,21 +124,22 @@ void build_decode_table(const std::vector<Node>& nodes, int32_t root,
         for (uint32_t b = 0; b < compresskit::BYTE_VALUES; ++b) {
             DecodeEntry& e = table[node][b];
             int32_t cur = node;
+            bool corrupt = false;
             for (int bit = compresskit::BITS_PER_BYTE - 1; bit >= 0; --bit) {
                 int v = (b >> bit) & 1;
                 cur = (v == 0) ? nodes[cur].left : nodes[cur].right;
                 if (cur < 0) {
                     // Corrupt stream during table build: shouldn't happen for valid trees.
                     e.count = 0;
-                    e.next = root;
+                    corrupt = true;
                     break;
                 }
                 if (is_leaf(nodes, cur)) {
                     e.symbols[e.count++] = nodes[cur].symbol;
                     cur = root;
                 }
             }
-            e.next = cur;
+            e.next = corrupt ? root : cur;
         }
     }
 }

algorithms/rle/cpp/main.cpp

@@ -15,6 +15,7 @@
 
 std::vector<uint8_t> rle_encode_buffer(const std::vector<uint8_t>& input) {
     std::vector<uint8_t> out;
+    out.reserve(input.size() / 8 + compresskit::MAGIC_SIZE + compresskit::RLE_PAIR_SIZE);
     compresskit::write_magic(out, compresskit::RLE_MAGIC);
 
     if (input.empty()) {

algorithms/shared/cpp/include/compresskit/bit_io.hpp

@@ -44,9 +44,10 @@ class BitWriter {
 // MSB-first bit reader. Returns 0 for bits read past the end of the stream.
 class BitReader {
 public:
-    explicit BitReader(const std::vector<uint8_t>& data) : data_(data) {}
+    explicit BitReader(const std::vector<uint8_t>& data) : data_(data.data()), size_(data.size()) {}
+    BitReader(const uint8_t* data, std::size_t size) : data_(data), size_(size) {}
     int read_bit() {
-        if (byte_pos_ >= data_.size()) {
+        if (byte_pos_ >= size_) {
             return 0;
         }
         int bit = (data_[byte_pos_] >> ((BITS_PER_BYTE - 1) - bit_pos_)) & 1;
@@ -56,10 +57,11 @@ class BitReader {
         }
         return bit;
     }
-    bool eof() const { return byte_pos_ >= data_.size(); }
+    bool eof() const { return byte_pos_ >= size_; }
 
 private:
-    const std::vector<uint8_t>& data_;
+    const uint8_t* data_;
+    std::size_t size_;
     std::size_t byte_pos_ = 0;
     int bit_pos_ = 0;
 };

tests/conformance/run_cli_smoke.py

@@ -25,6 +25,7 @@
     ROOT / "tests/data/empty.bin",
     ROOT / "tests/data/single_byte.bin",
     ROOT / "tests/data/alternating.bin",
+    ROOT / "tests/data/all_same_byte.bin",
     ROOT / "tests/data/small_dictionary_like.bin",
 )
 

tests/gen_testdata.py

@@ -12,6 +12,7 @@
 # - empty.bin              空文件
 # - single_byte.bin        单字节边界样本
 # - alternating.bin        交替字节模式
+# - all_same_byte.bin      全相同字节样本（RLE/Huffman 边界情况）
 # - small_dictionary_like.bin  小型重复词典风格样本
 
 ROOT = Path(__file__).resolve().parent.parent
@@ -87,6 +88,7 @@ def main():
     generate_literal_file(DATA_DIR / "empty.bin", b"")
     generate_literal_file(DATA_DIR / "single_byte.bin", b"\x00")
     generate_literal_file(DATA_DIR / "alternating.bin", (b"\xAA\x55" * 512))
+    generate_literal_file(DATA_DIR / "all_same_byte.bin", (b"\x00" * 4096))
     generate_literal_file(
         DATA_DIR / "small_dictionary_like.bin",
         (b"compresskit-dict-alpha\n" * 128)