String Compression (zlib/deflate)

--[[
    -------------------------------
    ---       INFORMATION       ---
    -------------------------------

    Written by Haoqian He
    Edited/Forked by Scarious

    This is the forked version of the LibDeflate library by Haoqian He intended for luau. Credit for the original source code and such
    goes to their respective creators (basic credits can be viewed under CREDITS, and more expansive credits/licensing info can be viewed
    under LICENSING AND COPYRIGHT.

    Original documentation can be viewed here: https://safeteewow.github.io/LibDeflate/source/LibDeflate.lua.html

    You can access the LibDeflate library (and most LibDeflate methods seen in the original documentation) by using Compression.Library


    -------------------------------
    ---      DOCUMENTATION      ---
    -------------------------------

    Compression Methods:

    Compression.Deflate.Compress(data, configs?)
    Compression.Zlib.Compress(data, configs?)


    Decompression Methods:

    Compression.Deflate.Decompress(compressedData)
    Compression.Zlib.Decompress(compressedData)


    USAGE:

        configs table:

        {
            level = 0; -- integer 0 -> 9 where 0 is no compression and 9 is most compression
            strategy = "" -- "huffman_only", "fixed", "dynamic"
        }

        note :: the higher the level, the slower the compression will be
             :: configs table is optional, if not supplied (aka nil) default level+strategy will be used


        methods:

        Method: Compression.Deflate.Compress(data, configs?):

            Description: Compresses a string using the raw deflate format

            Input:
                - String: data = The data to be compressed
                - table?: configs = The configuration table to control the compression

            Output:
                - String: compressedData = The compressed data
                - int: paddedBits = The number of bits padded at the end of the output


        Method: Compression.Deflate.Decompress(compressedData):

            Description: Decompresses a raw deflate compressed data.

            Input:
                - String: compressedData = The data to be decompressed

            Output:
                - String: data = The decompressed data



        Method: Compression.Zlib.Compress(data, configs?):

            Description: Compresses a string using the zlib format

            Input:
                - String: data = The data to be compressed
                - table?: configs = The configuration table to control the compression

            Output:
                - String: compressedData = The compressed data
                - int: paddedBits = The number of bits padded at the end of the output


        Method: Compression.Deflate.Decompress(compressedData):

            Description: Decompresses a zlib compressed data.

            Input:
                - String: compressedData = The data to be decompressed

            Output:
                - String: data = The decompressed data


    -------------------------------
    ---         CREDITS         ---
    -------------------------------

    - LibDeflate Library: Haoqian He
    - zlib: Jean-loup Gailly and Mark Adler
    - puff: Mark Adler
    - LibCompress: jjsheets and Galmok (WoW)
    - 6bit encoding/decoding: WeakAuras2 (WoW)


    -------------------------------
    --- LICENSING AND COPYRIGHT ---
    -------------------------------

    LibDeflate 1.0.2-release <br>
    Pure Lua compressor and decompressor with high compression ratio using
    DEFLATE/zlib format.
    @file LibDeflate.lua
    @author Haoqian He (Github: SafeteeWoW; World of Warcraft: Safetyy-Illidan(US))
    @copyright LibDeflate <2018-2020> Haoqian He
    @license zlib License
    This library is implemented according to the following specifications.
    Report a bug if LibDeflate is not fully compliant with those specs.
    Both compressors and decompressors have been implemented in the library.
    1. RFC1950: DEFLATE Compressed Data Format Specification version 1.3
    https://tools.ietf.org/html/rfc1951
    2. RFC1951: ZLIB Compressed Data Format Specification version 3.3
    https://tools.ietf.org/html/rfc1950


    zlib License
    (C) 2018-2020 Haoqian He
    This software is provided 'as-is', without any express or implied
    warranty.  In no event will the authors be held liable for any damages
    arising from the use of this software.
    Permission is granted to anyone to use this software for any purpose,
    including commercial applications, and to alter it and redistribute it
    freely, subject to the following restrictions:
    1. The origin of this software must not be misrepresented; you must not
       claim that you wrote the original software. If you use this software
       in a product, an acknowledgment in the product documentation would be
       appreciated but is not required.
    2. Altered source versions must be plainly marked as such, and must not be
       misrepresented as being the original software.
    3. This notice may not be removed or altered from any source distribution.
    License History:
        1. GNU General Public License Version 3 in v1.0.0 and earlier versions.
        2. GNU Lesser General Public License Version 3 in v1.0.1
        3. the zlib License since v1.0.2

    Credits and Disclaimer:
    This library rewrites the code from the algorithm
    and the ideas of the following projects,
    and uses their code to help to test the correctness of this library,
    but their code is not included directly in the library itself.
    Their original licenses shall be comply when used:
        1. zlib, by Jean-loup Gailly (compression) and Mark Adler (decompression).
            http://www.zlib.net/
            Licensed under zlib License. http://www.zlib.net/zlib_license.html
            For the compression algorithm.
        2. puff, by Mark Adler. https://github.com/madler/zlib/tree/master/contrib/puff
            Licensed under zlib License. http://www.zlib.net/zlib_license.html
            For the decompression algorithm.
        3. LibCompress, by jjsheets and Galmok of European Stormrage (Horde)
            https://www.wowace.com/projects/libcompress
            Licensed under GPLv2.
            https://www.gnu.org/licenses/old-licenses/gpl-2.0.html
            For the code to create customized codec.
        4. WeakAuras2,
            https://github.com/WeakAuras/WeakAuras2
            Licensed under GPLv2.
            For the 6bit encoding and decoding.
]]

local Compression = {}
local LibDeflate = {}

Compression.Deflate = {}
Compression.Zlib = {}
Compression.Library = LibDeflate

--[[
    Method: Compression.Deflate.Compress

    Description: Compresses a string using the raw deflate format

    Input:
        - String: data = The data to be compressed
        - table?: configs = The configuration table to control the compression

    Output:
        - String: compressedData = The compressed data
        - int: paddedBits = The number of bits padded at the end of the output

    For more information see:
        - LibDeflate:CompressDeflate
        - compression_configs
]]

function Compression.Deflate.Compress(data, configs)
    return LibDeflate:CompressDeflate(data, configs)
end


--[[
    Method: Compression.Deflate.Decompress

    Description: Decompresses a raw deflate compressed data.

    Input:
        - String: compressedData = The data to be decompressed

    Output:
        - String: data = The decompressed data

    For more information see:
        - LibDeflate:DecompressDeflate
        - compression_configs
]]

function Compression.Deflate.Decompress(compressedData)
    return LibDeflate:DecompressDeflate(compressedData)
end



--[[
    Method: Compression.Zlib.Compress

    Description: Compresses a string using the zlib format

    Input:
        - String: data = The data to be compressed
        - table?: configs = The configuration table to control the compression

    Output:
        - String: compressedData = The compressed data
        - int: paddedBits = The number of bits padded at the end of the output

    For more information see:
        - LibDeflate:CompressZlib
        - compression_configs
]]

function Compression.Zlib.Compress(data, configs)
    return LibDeflate:CompressZlib(data, configs)
end


--[[
    Method: Compression.Deflate.Decompress

    Description: Decompresses a zlib compressed data.

    Input:
        - String: compressedData = The data to be decompressed

    Output:
        - String: data = The decompressed data

    For more information see:
        - LibDeflate:DecompressZlib
        - compression_configs
]]

function Compression.Zlib.Decompress(compressedData)
    return LibDeflate:DecompressZlib(compressedData)
end





--[[

    LIBDEFLATE LIBRARY:

]]


do
    -- Semantic version. all lowercase.
    -- Suffix can be alpha1, alpha2, beta1, beta2, rc1, rc2, etc.
    -- NOTE: Two version numbers needs to modify.
    -- 1. On the top of LibDeflate.lua
    -- 2. _VERSION
    -- 3. _MINOR

    -- version to store the official version of LibDeflate
    local _VERSION = "1.0.2-release"

    -- When MAJOR is changed, I should name it as LibDeflate2
    local _MAJOR = "LibDeflate"

    -- Update this whenever a new version, for LibStub version registration.
    -- 0 : v0.x
    -- 1 : v1.0.0
    -- 2 : v1.0.1
    -- 3 : v1.0.2
    local _MINOR = 3

    local _COPYRIGHT =
        "LibDeflate ".._VERSION
        .." Copyright (C) 2018-2020 Haoqian He."
        .." Licensed under the zlib License"

    -- Register in the World of Warcraft library "LibStub" if detected.
    LibDeflate = {}

    LibDeflate._VERSION = _VERSION
    LibDeflate._MAJOR = _MAJOR
    LibDeflate._MINOR = _MINOR
    LibDeflate._COPYRIGHT = _COPYRIGHT
end

-- localize Lua api for faster access.
local assert = assert
local error = error
local pairs = pairs
local string_byte = string.byte
local string_char = string.char
local string_find = string.find
local string_gsub = string.gsub
local string_sub = string.sub
local table_concat = table.concat
local table_sort = table.sort
local tostring = tostring
local type = type

-- Converts i to 2^i, (0<=i<=32)
-- This is used to implement bit left shift and bit right shift.
-- "x >> y" in C:   "(x-x%_pow2[y])/_pow2[y]" in Lua
-- "x << y" in C:   "x*_pow2[y]" in Lua
local _pow2 = {}

-- Converts any byte to a character, (0<=byte<=255)
local _byte_to_char = {}

-- _reverseBitsTbl[len][val] stores the bit reverse of
-- the number with bit length "len" and value "val"
-- For example, decimal number 6 with bits length 5 is binary 00110
-- It's reverse is binary 01100,
-- which is decimal 12 and 12 == _reverseBitsTbl[5][6]
-- 1<=len<=9, 0<=val<=2^len-1
-- The reason for 1<=len<=9 is that the max of min bitlen of huffman code
-- of a huffman alphabet is 9?
local _reverse_bits_tbl = {}

-- Convert a LZ77 length (3<=len<=258) to
-- a deflate literal/LZ77_length code (257<=code<=285)
local _length_to_deflate_code = {}

-- convert a LZ77 length (3<=len<=258) to
-- a deflate literal/LZ77_length code extra bits.
local _length_to_deflate_extra_bits = {}

-- Convert a LZ77 length (3<=len<=258) to
-- a deflate literal/LZ77_length code extra bit length.
local _length_to_deflate_extra_bitlen = {}

-- Convert a small LZ77 distance (1<=dist<=256) to a deflate code.
local _dist256_to_deflate_code = {}

-- Convert a small LZ77 distance (1<=dist<=256) to
-- a deflate distance code extra bits.
local _dist256_to_deflate_extra_bits = {}

-- Convert a small LZ77 distance (1<=dist<=256) to
-- a deflate distance code extra bit length.
local _dist256_to_deflate_extra_bitlen = {}

-- Convert a literal/LZ77_length deflate code to LZ77 base length
-- The key of the table is (code - 256), 257<=code<=285
local _literal_deflate_code_to_base_len = {
    3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31,
    35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258,
}

-- Convert a literal/LZ77_length deflate code to base LZ77 length extra bits
-- The key of the table is (code - 256), 257<=code<=285
local _literal_deflate_code_to_extra_bitlen = {
    0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2,
    3, 3, 3, 3, 4, 4, 4, 4, 5, 5, 5, 5, 0,
}

-- Convert a distance deflate code to base LZ77 distance. (0<=code<=29)
local _dist_deflate_code_to_base_dist = {
    [0] = 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193,
    257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145,
    8193, 12289, 16385, 24577,
}

-- Convert a distance deflate code to LZ77 bits length. (0<=code<=29)
local _dist_deflate_code_to_extra_bitlen = {
    [0] = 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6,
    7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13,
}

-- The code order of the first huffman header in the dynamic deflate block.
-- See the page 12 of RFC1951
local _rle_codes_huffman_bitlen_order = {16, 17, 18,
    0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15,
}

-- The following tables are used by fixed deflate block.
-- The value of these tables are assigned at the bottom of the source.

-- The huffman code of the literal/LZ77_length deflate codes,
-- in fixed deflate block.
local _fix_block_literal_huffman_code

-- Convert huffman code of the literal/LZ77_length to deflate codes,
-- in fixed deflate block.
local _fix_block_literal_huffman_to_deflate_code

-- The bit length of the huffman code of literal/LZ77_length deflate codes,
-- in fixed deflate block.
local _fix_block_literal_huffman_bitlen

-- The count of each bit length of the literal/LZ77_length deflate codes,
-- in fixed deflate block.
local _fix_block_literal_huffman_bitlen_count

-- The huffman code of the distance deflate codes,
-- in fixed deflate block.
local _fix_block_dist_huffman_code

-- Convert huffman code of the distance to deflate codes,
-- in fixed deflate block.
local _fix_block_dist_huffman_to_deflate_code

-- The bit length of the huffman code of the distance deflate codes,
-- in fixed deflate block.
local _fix_block_dist_huffman_bitlen

-- The count of each bit length of the huffman code of
-- the distance deflate codes,
-- in fixed deflate block.
local _fix_block_dist_huffman_bitlen_count

for i = 0, 255 do
    _byte_to_char[i] = string_char(i)
end

do
    local pow = 1
    for i = 0, 32 do
        _pow2[i] = pow
        pow = pow * 2
    end
end

for i = 1, 9 do
    _reverse_bits_tbl[i] = {}
    for j=0, _pow2[i+1]-1 do
        local reverse = 0
        local value = j
        for _ = 1, i do
            -- The following line is equivalent to "res | (code %2)" in C.
            reverse = reverse - reverse%2
            + (((reverse%2==1) or (value % 2) == 1) and 1 or 0)
            value = (value-value%2)/2
            reverse = reverse * 2
        end
        _reverse_bits_tbl[i][j] = (reverse-reverse%2)/2
    end
end

-- The source code is written according to the pattern in the numbers
-- in RFC1951 Page10.
do
    local a = 18
    local b = 16
    local c = 265
    local bitlen = 1
    for len = 3, 258 do
        if len <= 10 then
            _length_to_deflate_code[len] = len + 254
            _length_to_deflate_extra_bitlen[len] = 0
        elseif len == 258 then
            _length_to_deflate_code[len] = 285
            _length_to_deflate_extra_bitlen[len] = 0
        else
            if len > a then
                a = a + b
                b = b * 2
                c = c + 4
                bitlen = bitlen + 1
            end
            local t = len-a-1+b/2
            _length_to_deflate_code[len] = (t-(t%(b/8)))/(b/8) + c
            _length_to_deflate_extra_bitlen[len] = bitlen
            _length_to_deflate_extra_bits[len] = t % (b/8)
        end
    end
end

-- The source code is written according to the pattern in the numbers
-- in RFC1951 Page11.
do
    _dist256_to_deflate_code[1] = 0
    _dist256_to_deflate_code[2] = 1
    _dist256_to_deflate_extra_bitlen[1] = 0
    _dist256_to_deflate_extra_bitlen[2] = 0

    local a = 3
    local b = 4
    local code = 2
    local bitlen = 0
    for dist = 3, 256 do
        if dist > b then
            a = a * 2
            b = b * 2
            code = code + 2
            bitlen = bitlen + 1
        end
        _dist256_to_deflate_code[dist] = (dist <= a) and code or (code+1)
        _dist256_to_deflate_extra_bitlen[dist] = (bitlen < 0) and 0 or bitlen
        if b >= 8 then
            _dist256_to_deflate_extra_bits[dist] = (dist-b/2-1) % (b/4)
        end
    end
end

--- Calculate the Adler-32 checksum of the string. <br>
-- See RFC1950 Page 9 https://tools.ietf.org/html/rfc1950 for the
-- definition of Adler-32 checksum.
-- @param str [string] the input string to calcuate its Adler-32 checksum.
-- @return [integer] The Adler-32 checksum, which is greater or equal to 0,
-- and less than 2^32 (4294967296).
function LibDeflate:Adler32(str)
    -- This function is loop unrolled by better performance.
    --
    -- Here is the minimum code:
    --
    -- local a = 1
    -- local b = 0
    -- for i=1, #str do
    --      local s = string.byte(str, i, i)
    --      a = (a+s)%65521
    --      b = (b+a)%65521
    --      end
    -- return b*65536+a
    if type(str) ~= "string" then
        error(("Usage: LibDeflate:Adler32(str):"
            .." 'str' - string expected got '%s'."):format(type(str)), 2)
    end
    local strlen = #str

    local i = 1
    local a = 1
    local b = 0
    while i <= strlen - 15 do
        local x1, x2, x3, x4, x5, x6, x7, x8,
        x9, x10, x11, x12, x13, x14, x15, x16 = string_byte(str, i, i+15)
        b = (b+16*a+16*x1+15*x2+14*x3+13*x4+12*x5+11*x6+10*x7+9*x8+8*x9
            +7*x10+6*x11+5*x12+4*x13+3*x14+2*x15+x16)%65521
        a = (a+x1+x2+x3+x4+x5+x6+x7+x8+x9+x10+x11+x12+x13+x14+x15+x16)%65521
        i =  i + 16
    end
    while (i <= strlen) do
        local x = string_byte(str, i, i)
        a = (a + x) % 65521
        b = (b + a) % 65521
        i = i + 1
    end
    return (b*65536+a) % 4294967296
end

-- Compare adler32 checksum.
-- adler32 should be compared with a mod to avoid sign problem
-- 4072834167 (unsigned) is the same adler32 as -222133129
local function IsEqualAdler32(actual, expected)
    return (actual % 4294967296) == (expected % 4294967296)
end

--- Create a preset dictionary.
--
-- This function is not fast, and the memory consumption of the produced
-- dictionary is about 50 times of the input string. Therefore, it is suggestted
-- to run this function only once in your program.
--
-- It is very important to know that if you do use a preset dictionary,
-- compressors and decompressors MUST USE THE SAME dictionary. That is,
-- dictionary must be created using the same string. If you update your program
-- with a new dictionary, people with the old version won't be able to transmit
-- data with people with the new version. Therefore, changing the dictionary
-- must be very careful.
--
-- The parameters "strlen" and "adler32" add a layer of verification to ensure
-- the parameter "str" is not modified unintentionally during the program
-- development.
--
-- @usage local dict_str = "1234567890"
--
-- -- print(dict_str:len(), LibDeflate:Adler32(dict_str))
-- -- Hardcode the print result below to verify it to avoid acciently
-- -- modification of 'str' during the program development.
-- -- string length: 10, Adler-32: 187433486,
-- -- Don't calculate string length and its Adler-32 at run-time.
--
-- local dict = LibDeflate:CreateDictionary(dict_str, 10, 187433486)
--
-- @param str [string] The string used as the preset dictionary. <br>
-- You should put stuffs that frequently appears in the dictionary
-- string and preferablely put more frequently appeared stuffs toward the end
-- of the string. <br>
-- Empty string and string longer than 32768 bytes are not allowed.
-- @param strlen [integer] The length of 'str'. Please pass in this parameter
-- as a hardcoded constant, in order to verify the content of 'str'. The value
-- of this parameter should be known before your program runs.
-- @param adler32 [integer] The Adler-32 checksum of 'str'. Please pass in this
-- parameter as a hardcoded constant, in order to verify the content of 'str'.
-- The value of this parameter should be known before your program runs.
-- @return  [table] The dictionary used for preset dictionary compression and
-- decompression.
-- @raise error if 'strlen' does not match the length of 'str',
-- or if 'adler32' does not match the Adler-32 checksum of 'str'.
function LibDeflate:CreateDictionary(str, strlen, adler32)
    if type(str) ~= "string" then
        error(("Usage: LibDeflate:CreateDictionary(str, strlen, adler32):"
            .." 'str' - string expected got '%s'."):format(type(str)), 2)
    end
    if type(strlen) ~= "number" then
        error(("Usage: LibDeflate:CreateDictionary(str, strlen, adler32):"
            .." 'strlen' - number expected got '%s'."):format(
                type(strlen)), 2)
    end
    if type(adler32) ~= "number" then
        error(("Usage: LibDeflate:CreateDictionary(str, strlen, adler32):"
            .." 'adler32' - number expected got '%s'."):format(
                type(adler32)), 2)
    end
    if strlen ~= #str then
        error(("Usage: LibDeflate:CreateDictionary(str, strlen, adler32):"
            .." 'strlen' does not match the actual length of 'str'."
            .." 'strlen': %u, '#str': %u ."
            .." Please check if 'str' is modified unintentionally.")
            :format(strlen, #str))
    end
    if strlen == 0 then
        error(("Usage: LibDeflate:CreateDictionary(str, strlen, adler32):"
            .." 'str' - Empty string is not allowed."), 2)
    end
    if strlen > 32768 then
        error(("Usage: LibDeflate:CreateDictionary(str, strlen, adler32):"
            .." 'str' - string longer than 32768 bytes is not allowed."
            .." Got %d bytes."):format(strlen), 2)
    end
    local actual_adler32 = self:Adler32(str)
    if not IsEqualAdler32(adler32, actual_adler32) then
        error(("Usage: LibDeflate:CreateDictionary(str, strlen, adler32):"
            .." 'adler32' does not match the actual adler32 of 'str'."
            .." 'adler32': %u, 'Adler32(str)': %u ."
            .." Please check if 'str' is modified unintentionally.")
            :format(adler32, actual_adler32))
    end

    local dictionary = {}
    dictionary.adler32 = adler32
    dictionary.hash_tables = {}
    dictionary.string_table = {}
    dictionary.strlen = strlen
    local string_table = dictionary.string_table
    local hash_tables = dictionary.hash_tables
    string_table[1] = string_byte(str, 1, 1)
    string_table[2] = string_byte(str, 2, 2)
    if strlen >= 3 then
        local i = 1
        local hash = string_table[1]*256+string_table[2]
        while i <= strlen - 2 - 3 do
            local x1, x2, x3, x4 = string_byte(str, i+2, i+5)
            string_table[i+2] = x1
            string_table[i+3] = x2
            string_table[i+4] = x3
            string_table[i+5] = x4
            hash = (hash*256+x1)%16777216
            local t = hash_tables[hash]
            if not t then t = {}; hash_tables[hash] = t end
            t[#t+1] = i-strlen
            i = i + 1
            hash = (hash*256+x2)%16777216
            t = hash_tables[hash]
            if not t then t = {}; hash_tables[hash] = t end
            t[#t+1] = i-strlen
            i = i + 1
            hash = (hash*256+x3)%16777216
            t = hash_tables[hash]
            if not t then t = {}; hash_tables[hash] = t end
            t[#t+1] = i-strlen
            i = i + 1
            hash = (hash*256+x4)%16777216
            t = hash_tables[hash]
            if not t then t = {}; hash_tables[hash] = t end
            t[#t+1] = i-strlen
            i = i + 1
        end
        while i <= strlen - 2 do
            local x = string_byte(str, i+2)
            string_table[i+2] = x
            hash = (hash*256+x)%16777216
            local t = hash_tables[hash]
            if not t then t = {}; hash_tables[hash] = t end
            t[#t+1] = i-strlen
            i = i + 1
        end
    end
    return dictionary
end

-- Check if the dictionary is valid.
-- @param dictionary The preset dictionary for compression and decompression.
-- @return true if valid, false if not valid.
-- @return if not valid, the error message.
local function IsValidDictionary(dictionary)
    if type(dictionary) ~= "table" then
        return false, ("'dictionary' - table expected got '%s'.")
        :format(type(dictionary))
    end
    if type(dictionary.adler32) ~= "number"
        or type(dictionary.string_table) ~= "table"
        or type(dictionary.strlen) ~= "number"
        or dictionary.strlen <= 0
        or dictionary.strlen > 32768
        or dictionary.strlen ~= #dictionary.string_table
        or type(dictionary.hash_tables) ~= "table"
    then
        return false, ("'dictionary' - corrupted dictionary.")
        :format(type(dictionary))
    end
    return true, ""
end

--[[
    key of the configuration table is the compression level,
    and its value stores the compression setting.
    These numbers come from zlib source code.
    Higher compression level usually means better compression.
    (Because LibDeflate uses a simplified version of zlib algorithm,
    there is no guarantee that higher compression level does not create
    bigger file than lower level, but I can say it's 99% likely)
    Be careful with the high compression level. This is a pure lua
    implementation compressor/decompressor, which is significant slower than
    a C/C++ equivalant compressor/decompressor. Very high compression level
    costs significant more CPU time, and usually compression size won't be
    significant smaller when you increase compression level by 1, when the
    level is already very high. Benchmark yourself if you can afford it.
    See also https://github.com/madler/zlib/blob/master/doc/algorithm.txt,
    https://github.com/madler/zlib/blob/master/deflate.c for more information.
    The meaning of each field:
    @field 1 use_lazy_evaluation:
        true/false. Whether the program uses lazy evaluation.
        See what is "lazy evaluation" in the link above.
        lazy_evaluation improves ratio, but relatively slow.
    @field 2 good_prev_length:
        Only effective if lazy is set, Only use 1/4 of max_chain,
        if prev length of lazy match is above this.
    @field 3 max_insert_length/max_lazy_match:
        If not using lazy evaluation,
        insert new strings in the hash table only if the match length is not
        greater than this length.
        If using lazy evaluation, only continue lazy evaluation,
        if previous match length is strictly smaller than this value.
    @field 4 nice_length:
        Number. Don't continue to go down the hash chain,
        if match length is above this.
    @field 5 max_chain:
        Number. The maximum number of hash chains we look.
--]]
local _compression_level_configs = {
    [0] = {false, nil, 0, 0, 0}, -- level 0, no compression
    [1] = {false, nil, 4, 8, 4}, -- level 1, similar to zlib level 1
    [2] = {false, nil, 5, 18, 8}, -- level 2, similar to zlib level 2
    [3] = {false, nil, 6, 32, 32},  -- level 3, similar to zlib level 3
    [4] = {true, 4, 4, 16, 16}, -- level 4, similar to zlib level 4
    [5] = {true, 8, 16, 32, 32}, -- level 5, similar to zlib level 5
    [6] = {true, 8, 16, 128, 128}, -- level 6, similar to zlib level 6
    [7] = {true, 8, 32, 128, 256}, -- (SLOW) level 7, similar to zlib level 7
    [8] = {true, 32, 128, 258, 1024} , --(SLOW) level 8,similar to zlib level 8
    [9] = {true, 32, 258, 258, 4096},
    -- (VERY SLOW) level 9, similar to zlib level 9
}

-- Check if the compression/decompression arguments is valid
-- @param str The input string.
-- @param check_dictionary if true, check if dictionary is valid.
-- @param dictionary The preset dictionary for compression and decompression.
-- @param check_configs if true, check if config is valid.
-- @param configs The compression configuration table
-- @return true if valid, false if not valid.
-- @return if not valid, the error message.
local function IsValidArguments(str,
    check_dictionary, dictionary,
    check_configs, configs)

    if type(str) ~= "string" then
        return false,
            ("'str' - string expected got '%s'."):format(type(str))
    end
    if check_dictionary then
        local dict_valid, dict_err = IsValidDictionary(dictionary)
        if not dict_valid then
            return false, dict_err
        end
    end
    if check_configs then
        local type_configs = type(configs)
        if type_configs ~= "nil" and type_configs ~= "table" then
            return false,
                ("'configs' - nil or table expected got '%s'.")
            :format(type(configs))
        end
        if type_configs == "table" then
            for k, v in pairs(configs) do
                if k ~= "level" and k ~= "strategy" then
                    return false,
                        ("'configs' - unsupported table key in the configs: '%s'.")
                    :format(k)
                elseif k == "level" and not _compression_level_configs[v] then
                    return false,
                        ("'configs' - unsupported 'level': %s."):format(tostring(v))
                elseif k == "strategy" and v ~= "fixed" and v ~= "huffman_only"
                    and v ~= "dynamic" then
                    -- random_block_type is for testing purpose
                    return false, ("'configs' - unsupported 'strategy': '%s'.")
                    :format(tostring(v))
                end
            end
        end
    end
    return true, ""
end



--[[ --------------------------------------------------------------------------
    Compress code
--]] --------------------------------------------------------------------------

-- partial flush to save memory
local _FLUSH_MODE_MEMORY_CLEANUP = 0
-- full flush with partial bytes
local _FLUSH_MODE_OUTPUT = 1
-- write bytes to get to byte boundary
local _FLUSH_MODE_BYTE_BOUNDARY = 2
-- no flush, just get num of bits written so far
local _FLUSH_MODE_NO_FLUSH = 3

--[[
    Create an empty writer to easily write stuffs as the unit of bits.
    Return values:
    1. WriteBits(code, bitlen):
    2. WriteString(str):
    3. Flush(mode):
--]]
local function CreateWriter()
    local buffer_size = 0
    local cache = 0
    local cache_bitlen = 0
    local total_bitlen = 0
    local buffer = {}
    -- When buffer is big enough, flush into result_buffer to save memory.
    local result_buffer = {}

    -- Write bits with value "value" and bit length of "bitlen" into writer.
    -- @param value: The value being written
    -- @param bitlen: The bit length of "value"
    -- @return nil
    local function WriteBits(value, bitlen)
        cache = cache + value * _pow2[cache_bitlen]
        cache_bitlen = cache_bitlen + bitlen
        total_bitlen = total_bitlen + bitlen
        -- Only bulk to buffer every 4 bytes. This is quicker.
        if cache_bitlen >= 32 then
            buffer_size = buffer_size + 1
            buffer[buffer_size] =
                _byte_to_char[cache % 256]
                .._byte_to_char[((cache-cache%256)/256 % 256)]
                .._byte_to_char[((cache-cache%65536)/65536 % 256)]
                .._byte_to_char[((cache-cache%16777216)/16777216 % 256)]
            local rshift_mask = _pow2[32 - cache_bitlen + bitlen]
            cache = (value - value%rshift_mask)/rshift_mask
            cache_bitlen = cache_bitlen - 32
        end
    end

    -- Write the entire string into the writer.
    -- @param str The string being written
    -- @return nil
    local function WriteString(str)
        for _ = 1, cache_bitlen, 8 do
            buffer_size = buffer_size + 1
            buffer[buffer_size] = string_char(cache % 256)
            cache = (cache-cache%256)/256
        end
        cache_bitlen = 0
        buffer_size = buffer_size + 1
        buffer[buffer_size] = str
        total_bitlen = total_bitlen + #str*8
    end

    -- Flush current stuffs in the writer and return it.
    -- This operation will free most of the memory.
    -- @param mode See the descrtion of the constant and the source code.
    -- @return The total number of bits stored in the writer right now.
    -- for byte boundary mode, it includes the padding bits.
    -- for output mode, it does not include padding bits.
    -- @return Return the outputs if mode is output.
    local function FlushWriter(mode)
        if mode == _FLUSH_MODE_NO_FLUSH then
            return total_bitlen
        end

        if mode == _FLUSH_MODE_OUTPUT
            or mode == _FLUSH_MODE_BYTE_BOUNDARY then
            -- Full flush, also output cache.
            -- Need to pad some bits if cache_bitlen is not multiple of 8.
            local padding_bitlen = (8 - cache_bitlen % 8) % 8

            if cache_bitlen > 0 then
                -- padding with all 1 bits, mainly because "\000" is not
                -- good to be tranmitted. I do this so "\000" is a little bit
                -- less frequent.
                cache = cache - _pow2[cache_bitlen]
                + _pow2[cache_bitlen+padding_bitlen]
                for _ = 1, cache_bitlen, 8 do
                    buffer_size = buffer_size + 1
                    buffer[buffer_size] = _byte_to_char[cache % 256]
                    cache = (cache-cache%256)/256
                end

                cache = 0
                cache_bitlen = 0
            end
            if mode == _FLUSH_MODE_BYTE_BOUNDARY then
                total_bitlen = total_bitlen + padding_bitlen
                return total_bitlen
            end
        end

        local flushed = table_concat(buffer)
        buffer = {}
        buffer_size = 0
        result_buffer[#result_buffer+1] = flushed

        if mode == _FLUSH_MODE_MEMORY_CLEANUP then
            return total_bitlen
        else
            return total_bitlen, table_concat(result_buffer)
        end
    end

    return WriteBits, WriteString, FlushWriter
end

-- Push an element into a max heap
-- @param heap A max heap whose max element is at index 1.
-- @param e The element to be pushed. Assume element "e" is a table
--  and comparison is done via its first entry e[1]
-- @param heap_size current number of elements in the heap.
--  NOTE: There may be some garbage stored in
--  heap[heap_size+1], heap[heap_size+2], etc..
-- @return nil
local function MinHeapPush(heap, e, heap_size)
    heap_size = heap_size + 1
    heap[heap_size] = e
    local value = e[1]
    local pos = heap_size
    local parent_pos = (pos-pos%2)/2

    while (parent_pos >= 1 and heap[parent_pos][1] > value) do
        local t = heap[parent_pos]
        heap[parent_pos] = e
        heap[pos] = t
        pos = parent_pos
        parent_pos = (parent_pos-parent_pos%2)/2
    end
end

-- Pop an element from a max heap
-- @param heap A max heap whose max element is at index 1.
-- @param heap_size current number of elements in the heap.
-- @return the poped element
-- Note: This function does not change table size of "heap" to save CPU time.
local function MinHeapPop(heap, heap_size)
    local top = heap[1]
    local e = heap[heap_size]
    local value = e[1]
    heap[1] = e
    heap[heap_size] = top
    heap_size = heap_size - 1

    local pos = 1
    local left_child_pos = pos * 2
    local right_child_pos = left_child_pos + 1

    while (left_child_pos <= heap_size) do
        local left_child = heap[left_child_pos]
        if (right_child_pos <= heap_size
            and heap[right_child_pos][1] < left_child[1]) then
            local right_child = heap[right_child_pos]
            if right_child[1] < value then
                heap[right_child_pos] = e
                heap[pos] = right_child
                pos = right_child_pos
                left_child_pos = pos * 2
                right_child_pos = left_child_pos + 1
            else
                break
            end
        else
            if left_child[1] < value then
                heap[left_child_pos] = e
                heap[pos] = left_child
                pos = left_child_pos
                left_child_pos = pos * 2
                right_child_pos = left_child_pos + 1
            else
                break
            end
        end
    end

    return top
end

-- Deflate defines a special huffman tree, which is unique once the bit length
-- of huffman code of all symbols are known.
-- @param bitlen_count Number of symbols with a specific bitlen
-- @param symbol_bitlen The bit length of a symbol
-- @param max_symbol The max symbol among all symbols,
--      which is (number of symbols - 1)
-- @param max_bitlen The max huffman bit length among all symbols.
-- @return The huffman code of all symbols.
local function GetHuffmanCodeFromBitlen(bitlen_counts, symbol_bitlens
    , max_symbol, max_bitlen)
    local huffman_code = 0
    local next_codes = {}
    local symbol_huffman_codes = {}
    for bitlen = 1, max_bitlen do
        huffman_code = (huffman_code+(bitlen_counts[bitlen-1] or 0))*2
        next_codes[bitlen] = huffman_code
    end
    for symbol = 0, max_symbol do
        local bitlen = symbol_bitlens[symbol]
        if bitlen then
            huffman_code = next_codes[bitlen]
            next_codes[bitlen] = huffman_code + 1

            -- Reverse the bits of huffman code,
            -- because most signifant bits of huffman code
            -- is stored first into the compressed data.
            -- @see RFC1951 Page5 Section 3.1.1
            if bitlen <= 9 then -- Have cached reverse for small bitlen.
                symbol_huffman_codes[symbol] =
                    _reverse_bits_tbl[bitlen][huffman_code]
            else
                local reverse = 0
                for _ = 1, bitlen do
                    reverse = reverse - reverse%2
                    + (((reverse%2==1)
                        or (huffman_code % 2) == 1) and 1 or 0)
                    huffman_code = (huffman_code-huffman_code%2)/2
                    reverse = reverse*2
                end
                symbol_huffman_codes[symbol] = (reverse-reverse%2)/2
            end
        end
    end
    return symbol_huffman_codes
end

-- A helper function to sort heap elements
-- a[1], b[1] is the huffman frequency
-- a[2], b[2] is the symbol value.
local function SortByFirstThenSecond(a, b)
    return a[1] < b[1] or
        (a[1] == b[1] and a[2] < b[2])
end

-- Calculate the huffman bit length and huffman code.
-- @param symbol_count: A table whose table key is the symbol, and table value
--      is the symbol frenquency (nil means 0 frequency).
-- @param max_bitlen: See description of return value.
-- @param max_symbol: The maximum symbol
-- @return a table whose key is the symbol, and the value is the huffman bit
--      bit length. We guarantee that all bit length <= max_bitlen.
--      For 0<=symbol<=max_symbol, table value could be nil if the frequency
--      of the symbol is 0 or nil.
-- @return a table whose key is the symbol, and the value is the huffman code.
-- @return a number indicating the maximum symbol whose bitlen is not 0.
local function GetHuffmanBitlenAndCode(symbol_counts, max_bitlen, max_symbol)
    local heap_size
    local max_non_zero_bitlen_symbol = -1
    local leafs = {}
    local heap = {}
    local symbol_bitlens = {}
    local symbol_codes = {}
    local bitlen_counts = {}

    --[[
        tree[1]: weight, temporarily used as parent and bitLengths
        tree[2]: symbol
        tree[3]: left child
        tree[4]: right child
    --]]
    local number_unique_symbols = 0
    for symbol, count in pairs(symbol_counts) do
        number_unique_symbols = number_unique_symbols + 1
        leafs[number_unique_symbols] = {count, symbol}
    end

    if (number_unique_symbols == 0) then
        -- no code.
        return {}, {}, -1
    elseif (number_unique_symbols == 1) then
        -- Only one code. In this case, its huffman code
        -- needs to be assigned as 0, and bit length is 1.
        -- This is the only case that the return result
        -- represents an imcomplete huffman tree.
        local symbol = leafs[1][2]
        symbol_bitlens[symbol] = 1
        symbol_codes[symbol] = 0
        return symbol_bitlens, symbol_codes, symbol
    else
        table_sort(leafs, SortByFirstThenSecond)
        heap_size = number_unique_symbols
        for i = 1, heap_size do
            heap[i] = leafs[i]
        end

        while (heap_size > 1) do
            -- Note: pop does not change table size of heap
            local leftChild = MinHeapPop(heap, heap_size)
            heap_size = heap_size - 1
            local rightChild = MinHeapPop(heap, heap_size)
            heap_size = heap_size - 1
            local newNode =
            {leftChild[1]+rightChild[1], -1, leftChild, rightChild}
            MinHeapPush(heap, newNode, heap_size)
            heap_size = heap_size + 1
        end

        -- Number of leafs whose bit length is greater than max_len.
        local number_bitlen_overflow = 0

        -- Calculate bit length of all nodes
        local fifo = {heap[1], 0, 0, 0} -- preallocate some spaces.
        local fifo_size = 1
        local index = 1
        heap[1][1] = 0
        while (index <= fifo_size) do -- Breath first search
            local e = fifo[index]
            local bitlen = e[1]
            local symbol = e[2]
            local left_child = e[3]
            local right_child = e[4]
            if left_child then
                fifo_size = fifo_size + 1
                fifo[fifo_size] = left_child
                left_child[1] = bitlen + 1
            end
            if right_child then
                fifo_size = fifo_size + 1
                fifo[fifo_size] = right_child
                right_child[1] = bitlen + 1
            end
            index = index + 1

            if (bitlen > max_bitlen) then
                number_bitlen_overflow = number_bitlen_overflow + 1
                bitlen = max_bitlen
            end
            if symbol >= 0 then
                symbol_bitlens[symbol] = bitlen
                max_non_zero_bitlen_symbol =
                    (symbol > max_non_zero_bitlen_symbol)
                    and symbol or max_non_zero_bitlen_symbol
                bitlen_counts[bitlen] = (bitlen_counts[bitlen] or 0) + 1
            end
        end

        -- Resolve bit length overflow
        -- @see ZLib/trees.c:gen_bitlen(s, desc), for reference
        if (number_bitlen_overflow > 0) then
            repeat
                local bitlen = max_bitlen - 1
                while ((bitlen_counts[bitlen] or 0) == 0) do
                    bitlen = bitlen - 1
                end
                -- move one leaf down the tree
                bitlen_counts[bitlen] = bitlen_counts[bitlen] - 1
                -- move one overflow item as its brother
                bitlen_counts[bitlen+1] = (bitlen_counts[bitlen+1] or 0) + 2
                bitlen_counts[max_bitlen] = bitlen_counts[max_bitlen] - 1
                number_bitlen_overflow = number_bitlen_overflow - 2
            until (number_bitlen_overflow <= 0)

            index = 1
            for bitlen = max_bitlen, 1, -1 do
                local n = bitlen_counts[bitlen] or 0
                while (n > 0) do
                    local symbol = leafs[index][2]
                    symbol_bitlens[symbol] = bitlen
                    n = n - 1
                    index = index + 1
                end
            end
        end

        symbol_codes = GetHuffmanCodeFromBitlen(bitlen_counts, symbol_bitlens,
            max_symbol, max_bitlen)
        return symbol_bitlens, symbol_codes, max_non_zero_bitlen_symbol
    end
end

-- Calculate the first huffman header in the dynamic huffman block
-- @see RFC1951 Page 12
-- @param lcode_bitlen: The huffman bit length of literal/LZ77_length.
-- @param max_non_zero_bitlen_lcode: The maximum literal/LZ77_length symbol
--      whose huffman bit length is not zero.
-- @param dcode_bitlen: The huffman bit length of LZ77 distance.
-- @param max_non_zero_bitlen_dcode: The maximum LZ77 distance symbol
--      whose huffman bit length is not zero.
-- @return The run length encoded codes.
-- @return The extra bits. One entry for each rle code that needs extra bits.
--      (code == 16 or 17 or 18).
-- @return The count of appearance of each rle codes.
local function RunLengthEncodeHuffmanBitlen(
    lcode_bitlens,
    max_non_zero_bitlen_lcode,
    dcode_bitlens,
    max_non_zero_bitlen_dcode)
    local rle_code_tblsize = 0
    local rle_codes = {}
    local rle_code_counts = {}
    local rle_extra_bits_tblsize = 0
    local rle_extra_bits = {}
    local prev = nil
    local count = 0

    -- If there is no distance code, assume one distance code of bit length 0.
    -- RFC1951: One distance code of zero bits means that
    -- there are no distance codes used at all (the data is all literals).
    max_non_zero_bitlen_dcode = (max_non_zero_bitlen_dcode < 0)
    and 0 or max_non_zero_bitlen_dcode
    local max_code = max_non_zero_bitlen_lcode+max_non_zero_bitlen_dcode+1

    for code = 0, max_code+1 do
        local len = (code <= max_non_zero_bitlen_lcode)
        and (lcode_bitlens[code] or 0)
        or ((code <= max_code)
            and (dcode_bitlens[code-max_non_zero_bitlen_lcode-1] or 0) or nil)
        if len == prev then
            count = count + 1
            if len ~= 0 and count == 6 then
                rle_code_tblsize = rle_code_tblsize + 1
                rle_codes[rle_code_tblsize] = 16
                rle_extra_bits_tblsize = rle_extra_bits_tblsize + 1
                rle_extra_bits[rle_extra_bits_tblsize] = 3
                rle_code_counts[16] = (rle_code_counts[16] or 0) + 1
                count = 0
            elseif len == 0 and count == 138 then
                rle_code_tblsize = rle_code_tblsize + 1
                rle_codes[rle_code_tblsize] = 18
                rle_extra_bits_tblsize = rle_extra_bits_tblsize + 1
                rle_extra_bits[rle_extra_bits_tblsize] = 127
                rle_code_counts[18] = (rle_code_counts[18] or 0) + 1
                count = 0
            end
        else
            if count == 1 then
                rle_code_tblsize = rle_code_tblsize + 1
                rle_codes[rle_code_tblsize] = prev
                rle_code_counts[prev] = (rle_code_counts[prev] or 0) + 1
            elseif count == 2 then
                rle_code_tblsize = rle_code_tblsize + 1
                rle_codes[rle_code_tblsize] = prev
                rle_code_tblsize = rle_code_tblsize + 1
                rle_codes[rle_code_tblsize] = prev
                rle_code_counts[prev] = (rle_code_counts[prev] or 0) + 2
            elseif count >= 3 then
                rle_code_tblsize = rle_code_tblsize + 1
                local rleCode = (prev ~= 0) and 16 or (count <= 10 and 17 or 18)
                rle_codes[rle_code_tblsize] = rleCode
                rle_code_counts[rleCode] = (rle_code_counts[rleCode] or 0) + 1
                rle_extra_bits_tblsize = rle_extra_bits_tblsize + 1
                rle_extra_bits[rle_extra_bits_tblsize] =
                    (count <= 10) and (count - 3) or (count - 11)
            end

            prev = len
            if len and len ~= 0 then
                rle_code_tblsize = rle_code_tblsize + 1
                rle_codes[rle_code_tblsize] = len
                rle_code_counts[len] = (rle_code_counts[len] or 0) + 1
                count = 0
            else
                count = 1
            end
        end
    end

    return rle_codes, rle_extra_bits, rle_code_counts
end

-- Load the string into a table, in order to speed up LZ77.
-- Loop unrolled 16 times to speed this function up.
-- @param str The string to be loaded.
-- @param t The load destination
-- @param start str[index] will be the first character to be loaded.
-- @param end str[index] will be the last character to be loaded
-- @param offset str[index] will be loaded into t[index-offset]
-- @return t
local function LoadStringToTable(str, t, start, stop, offset)
    local i = start - offset
    while i <= stop - 15 - offset do
        t[i], t[i+1], t[i+2], t[i+3], t[i+4], t[i+5], t[i+6], t[i+7], t[i+8],
            t[i+9], t[i+10], t[i+11], t[i+12], t[i+13], t[i+14], t[i+15] =
            string_byte(str, i + offset, i + 15 + offset)
        i = i + 16
    end
    while (i <= stop - offset) do
        t[i] = string_byte(str, i + offset, i + offset)
        i = i + 1
    end
    return t
end

-- Do LZ77 process. This function uses the majority of the CPU time.
-- @see zlib/deflate.c:deflate_fast(), zlib/deflate.c:deflate_slow()
-- @see https://github.com/madler/zlib/blob/master/doc/algorithm.txt
-- This function uses the algorithms used above. You should read the
-- algorithm.txt above to understand what is the hash function and the
-- lazy evaluation.
--
-- The special optimization used here is hash functions used here.
-- The hash function is just the multiplication of the three consective
-- characters. So if the hash matches, it guarantees 3 characters are matched.
-- This optimization can be implemented because Lua table is a hash table.
--
-- @param level integer that describes compression level.
-- @param string_table table that stores the value of string to be compressed.
--          The index of this table starts from 1.
--          The caller needs to make sure all values needed by this function
--          are loaded.
--          Assume "str" is the origin input string into the compressor
--          str[block_start]..str[block_end+3] needs to be loaded into
--          string_table[block_start-offset]..string_table[block_end-offset]
--          If dictionary is presented, the last 258 bytes of the dictionary
--          needs to be loaded into sing_table[-257..0]
--          (See more in the description of offset.)
-- @param hash_tables. The table key is the hash value (0<=hash<=16777216=256^3)
--          The table value is an array0 that stores the indexes of the
--          input data string to be compressed, such that
--          hash == str[index]*str[index+1]*str[index+2]
--          Indexes are ordered in this array.
-- @param block_start The indexes of the input data string to be compressed.
--              that starts the LZ77 block.
-- @param block_end The indexes of the input data string to be compressed.
--              that stores the LZ77 block.
-- @param offset str[index] is stored in string_table[index-offset],
--          This offset is mainly an optimization to limit the index
--          of string_table, so lua can access this table quicker.
-- @param dictionary See LibDeflate:CreateDictionary
-- @return literal/LZ77_length deflate codes.
-- @return the extra bits of literal/LZ77_length deflate codes.
-- @return the count of each literal/LZ77 deflate code.
-- @return LZ77 distance deflate codes.
-- @return the extra bits of LZ77 distance deflate codes.
-- @return the count of each LZ77 distance deflate code.
local function GetBlockLZ77Result(level, string_table, hash_tables, block_start,
    block_end, offset, dictionary)
    local config = _compression_level_configs[level]
    local config_use_lazy
    , config_good_prev_length
    , config_max_lazy_match
    , config_nice_length
    , config_max_hash_chain =
        config[1], config[2], config[3], config[4], config[5]

    local config_max_insert_length = (not config_use_lazy)
    and config_max_lazy_match or 2147483646
    local config_good_hash_chain =
        (config_max_hash_chain-config_max_hash_chain%4/4)

    local hash

    local dict_hash_tables
    local dict_string_table
    local dict_string_len = 0

    if dictionary then
        dict_hash_tables = dictionary.hash_tables
        dict_string_table = dictionary.string_table
        dict_string_len = dictionary.strlen
        assert(block_start == 1)
        if block_end >= block_start and dict_string_len >= 2 then
            hash = dict_string_table[dict_string_len-1]*65536
            + dict_string_table[dict_string_len]*256 + string_table[1]
            local t = hash_tables[hash]
            if not t then t = {}; hash_tables[hash] = t end
            t[#t+1] = -1
        end
        if block_end >= block_start+1 and dict_string_len >= 1 then
            hash = dict_string_table[dict_string_len]*65536
            + string_table[1]*256 + string_table[2]
            local t = hash_tables[hash]
            if not t then t = {}; hash_tables[hash] = t end
            t[#t+1] = 0
        end
    end

    local dict_string_len_plus3 = dict_string_len + 3

    hash = (string_table[block_start-offset] or 0)*256
    + (string_table[block_start+1-offset] or 0)

    local lcodes = {}
    local lcode_tblsize = 0
    local lcodes_counts = {}
    local dcodes = {}
    local dcodes_tblsize = 0
    local dcodes_counts = {}

    local lextra_bits = {}
    local lextra_bits_tblsize = 0
    local dextra_bits = {}
    local dextra_bits_tblsize = 0

    local match_available = false
    local prev_len
    local prev_dist
    local cur_len = 0
    local cur_dist = 0

    local index = block_start
    local index_end = block_end + (config_use_lazy and 1 or 0)

    -- the zlib source code writes separate code for lazy evaluation and
    -- not lazy evaluation, which is easier to understand.
    -- I put them together, so it is a bit harder to understand.
    -- because I think this is easier for me to maintain it.
    while (index <= index_end) do
        local string_table_index = index - offset
        local offset_minus_three = offset - 3
        prev_len = cur_len
        prev_dist = cur_dist
        cur_len = 0

        hash = (hash*256+(string_table[string_table_index+2] or 0))%16777216

        local chain_index
        local cur_chain
        local hash_chain = hash_tables[hash]
        local chain_old_size
        if not hash_chain then
            chain_old_size = 0
            hash_chain = {}
            hash_tables[hash] = hash_chain
            if dict_hash_tables then
                cur_chain = dict_hash_tables[hash]
                chain_index = cur_chain and #cur_chain or 0
            else
                chain_index = 0
            end
        else
            chain_old_size = #hash_chain
            cur_chain = hash_chain
            chain_index = chain_old_size
        end

        if index <= block_end then
            hash_chain[chain_old_size+1] = index
        end

        if (chain_index > 0 and index + 2 <= block_end
            and (not config_use_lazy or prev_len < config_max_lazy_match)) then

            local depth =
                (config_use_lazy and prev_len >= config_good_prev_length)
                and config_good_hash_chain or config_max_hash_chain

            local max_len_minus_one = block_end - index
            max_len_minus_one = (max_len_minus_one >= 257) and 257 or max_len_minus_one
            max_len_minus_one = max_len_minus_one + string_table_index
            local string_table_index_plus_three = string_table_index + 3

            while chain_index >= 1 and depth > 0 do
                local prev = cur_chain[chain_index]

                if index - prev > 32768 then
                    break
                end
                if prev < index then
                    local sj = string_table_index_plus_three

                    if prev >= -257 then
                        local pj = prev - offset_minus_three
                        while (sj <= max_len_minus_one
                            and string_table[pj]
                            == string_table[sj]) do
                            sj = sj + 1
                            pj = pj + 1
                        end
                    else
                        local pj = dict_string_len_plus3 + prev
                        while (sj <= max_len_minus_one
                            and dict_string_table[pj]
                            == string_table[sj]) do
                            sj = sj + 1
                            pj = pj + 1
                        end
                    end
                    local j = sj - string_table_index
                    if j > cur_len then
                        cur_len = j
                        cur_dist = index - prev
                    end
                    if cur_len >= config_nice_length then
                        break
                    end
                end

                chain_index = chain_index - 1
                depth = depth - 1
                if chain_index == 0 and prev > 0 and dict_hash_tables then
                    cur_chain = dict_hash_tables[hash]
                    chain_index = cur_chain and #cur_chain or 0
                end
            end
        end

        if not config_use_lazy then
            prev_len, prev_dist = cur_len, cur_dist
        end
        if ((not config_use_lazy or match_available)
            and (prev_len > 3 or (prev_len == 3 and prev_dist < 4096))
            and cur_len <= prev_len )then
            local code = _length_to_deflate_code[prev_len]
            local length_extra_bits_bitlen =
                _length_to_deflate_extra_bitlen[prev_len]
            local dist_code, dist_extra_bits_bitlen, dist_extra_bits
            if prev_dist <= 256 then -- have cached code for small distance.
                dist_code = _dist256_to_deflate_code[prev_dist]
                dist_extra_bits = _dist256_to_deflate_extra_bits[prev_dist]
                dist_extra_bits_bitlen =
                    _dist256_to_deflate_extra_bitlen[prev_dist]
            else
                dist_code = 16
                dist_extra_bits_bitlen = 7
                local a = 384
                local b = 512

                while true do
                    if prev_dist <= a then
                        dist_extra_bits = (prev_dist-(b/2)-1) % (b/4)
                        break
                    elseif prev_dist <= b then
                        dist_extra_bits = (prev_dist-(b/2)-1) % (b/4)
                        dist_code = dist_code + 1
                        break
                    else
                        dist_code = dist_code + 2
                        dist_extra_bits_bitlen = dist_extra_bits_bitlen + 1
                        a = a*2
                        b = b*2
                    end
                end
            end
            lcode_tblsize = lcode_tblsize + 1
            lcodes[lcode_tblsize] = code
            lcodes_counts[code] = (lcodes_counts[code] or 0) + 1

            dcodes_tblsize = dcodes_tblsize + 1
            dcodes[dcodes_tblsize] = dist_code
            dcodes_counts[dist_code] = (dcodes_counts[dist_code] or 0) + 1

            if length_extra_bits_bitlen > 0 then
                local lenExtraBits = _length_to_deflate_extra_bits[prev_len]
                lextra_bits_tblsize = lextra_bits_tblsize + 1
                lextra_bits[lextra_bits_tblsize] = lenExtraBits
            end
            if dist_extra_bits_bitlen > 0 then
                dextra_bits_tblsize = dextra_bits_tblsize + 1
                dextra_bits[dextra_bits_tblsize] = dist_extra_bits
            end

            for i=index+1, index+prev_len-(config_use_lazy and 2 or 1) do
                hash = (hash*256+(string_table[i-offset+2] or 0))%16777216
                if prev_len <= config_max_insert_length then
                    hash_chain = hash_tables[hash]
                    if not hash_chain then
                        hash_chain = {}
                        hash_tables[hash] = hash_chain
                    end
                    hash_chain[#hash_chain+1] = i
                end
            end
            index = index + prev_len - (config_use_lazy and 1 or 0)
            match_available = false
        elseif (not config_use_lazy) or match_available then
            local code = string_table[config_use_lazy
            and (string_table_index-1) or string_table_index]
            lcode_tblsize = lcode_tblsize + 1
            lcodes[lcode_tblsize] = code
            lcodes_counts[code] = (lcodes_counts[code] or 0) + 1
            index = index + 1
        else
            match_available = true
            index = index + 1
        end
    end

    -- Write "end of block" symbol
    lcode_tblsize = lcode_tblsize + 1
    lcodes[lcode_tblsize] = 256
    lcodes_counts[256] = (lcodes_counts[256] or 0) + 1

    return lcodes, lextra_bits, lcodes_counts, dcodes, dextra_bits
    , dcodes_counts
end

-- Get the header data of dynamic block.
-- @param lcodes_count The count of each literal/LZ77_length codes.
-- @param dcodes_count The count of each Lz77 distance codes.
-- @return a lots of stuffs.
-- @see RFC1951 Page 12
local function GetBlockDynamicHuffmanHeader(lcodes_counts, dcodes_counts)
    local lcodes_huffman_bitlens, lcodes_huffman_codes
    , max_non_zero_bitlen_lcode =
        GetHuffmanBitlenAndCode(lcodes_counts, 15, 285)
    local dcodes_huffman_bitlens, dcodes_huffman_codes
    , max_non_zero_bitlen_dcode =
        GetHuffmanBitlenAndCode(dcodes_counts, 15, 29)

    local rle_deflate_codes, rle_extra_bits, rle_codes_counts =
        RunLengthEncodeHuffmanBitlen(lcodes_huffman_bitlens
            ,max_non_zero_bitlen_lcode, dcodes_huffman_bitlens
            , max_non_zero_bitlen_dcode)

    local rle_codes_huffman_bitlens, rle_codes_huffman_codes =
        GetHuffmanBitlenAndCode(rle_codes_counts, 7, 18)

    local HCLEN = 0
    for i = 1, 19 do
        local symbol = _rle_codes_huffman_bitlen_order[i]
        local length = rle_codes_huffman_bitlens[symbol] or 0
        if length ~= 0 then
            HCLEN = i
        end
    end

    HCLEN = HCLEN - 4
    local HLIT = max_non_zero_bitlen_lcode + 1 - 257
    local HDIST = max_non_zero_bitlen_dcode + 1 - 1
    if HDIST < 0 then HDIST = 0 end

    return HLIT, HDIST, HCLEN, rle_codes_huffman_bitlens
    , rle_codes_huffman_codes, rle_deflate_codes, rle_extra_bits
    , lcodes_huffman_bitlens, lcodes_huffman_codes
    , dcodes_huffman_bitlens, dcodes_huffman_codes
end

-- Get the size of dynamic block without writing any bits into the writer.
-- @param ... Read the source code of GetBlockDynamicHuffmanHeader()
-- @return the bit length of the dynamic block
local function GetDynamicHuffmanBlockSize(lcodes, dcodes, HCLEN
    , rle_codes_huffman_bitlens, rle_deflate_codes
    , lcodes_huffman_bitlens, dcodes_huffman_bitlens)

    local block_bitlen = 17 -- 1+2+5+5+4
    block_bitlen = block_bitlen + (HCLEN+4)*3

    for i = 1, #rle_deflate_codes do
        local code = rle_deflate_codes[i]
        block_bitlen = block_bitlen + rle_codes_huffman_bitlens[code]
        if code >= 16 then
            block_bitlen = block_bitlen +
                ((code == 16) and 2 or (code == 17 and 3 or 7))
        end
    end

    local length_code_count = 0
    for i = 1, #lcodes do
        local code = lcodes[i]
        local huffman_bitlen = lcodes_huffman_bitlens[code]
        block_bitlen = block_bitlen + huffman_bitlen
        if code > 256 then -- Length code
            length_code_count = length_code_count + 1
            if code > 264 and code < 285 then -- Length code with extra bits
                local extra_bits_bitlen =
                    _literal_deflate_code_to_extra_bitlen[code-256]
                block_bitlen = block_bitlen + extra_bits_bitlen
            end
            local dist_code = dcodes[length_code_count]
            local dist_huffman_bitlen = dcodes_huffman_bitlens[dist_code]
            block_bitlen = block_bitlen + dist_huffman_bitlen

            if dist_code > 3 then -- dist code with extra bits
                local dist_extra_bits_bitlen = (dist_code-dist_code%2)/2 - 1
                block_bitlen = block_bitlen + dist_extra_bits_bitlen
            end
        end
    end
    return block_bitlen
end

-- Write dynamic block.
-- @param ... Read the source code of GetBlockDynamicHuffmanHeader()
local function CompressDynamicHuffmanBlock(WriteBits, is_last_block
    , lcodes, lextra_bits, dcodes, dextra_bits, HLIT, HDIST, HCLEN
    , rle_codes_huffman_bitlens, rle_codes_huffman_codes
    , rle_deflate_codes, rle_extra_bits
    , lcodes_huffman_bitlens, lcodes_huffman_codes
    , dcodes_huffman_bitlens, dcodes_huffman_codes)

    WriteBits(is_last_block and 1 or 0, 1) -- Last block identifier
    WriteBits(2, 2) -- Dynamic Huffman block identifier

    WriteBits(HLIT, 5)
    WriteBits(HDIST, 5)
    WriteBits(HCLEN, 4)

    for i = 1, HCLEN+4 do
        local symbol = _rle_codes_huffman_bitlen_order[i]
        local length = rle_codes_huffman_bitlens[symbol] or 0
        WriteBits(length, 3)
    end

    local rleExtraBitsIndex = 1
    for i=1, #rle_deflate_codes do
        local code = rle_deflate_codes[i]
        WriteBits(rle_codes_huffman_codes[code]
            , rle_codes_huffman_bitlens[code])
        if code >= 16 then
            local extraBits = rle_extra_bits[rleExtraBitsIndex]
            WriteBits(extraBits, (code == 16) and 2 or (code == 17 and 3 or 7))
            rleExtraBitsIndex = rleExtraBitsIndex + 1
        end
    end

    local length_code_count = 0
    local length_code_with_extra_count = 0
    local dist_code_with_extra_count = 0

    for i=1, #lcodes do
        local deflate_codee = lcodes[i]
        local huffman_code = lcodes_huffman_codes[deflate_codee]
        local huffman_bitlen = lcodes_huffman_bitlens[deflate_codee]
        WriteBits(huffman_code, huffman_bitlen)
        if deflate_codee > 256 then -- Length code
            length_code_count = length_code_count + 1
            if deflate_codee > 264 and deflate_codee < 285 then
                -- Length code with extra bits
                length_code_with_extra_count = length_code_with_extra_count + 1
                local extra_bits = lextra_bits[length_code_with_extra_count]
                local extra_bits_bitlen =
                    _literal_deflate_code_to_extra_bitlen[deflate_codee-256]
                WriteBits(extra_bits, extra_bits_bitlen)
            end
            -- Write distance code
            local dist_deflate_code = dcodes[length_code_count]
            local dist_huffman_code = dcodes_huffman_codes[dist_deflate_code]
            local dist_huffman_bitlen =
                dcodes_huffman_bitlens[dist_deflate_code]
            WriteBits(dist_huffman_code, dist_huffman_bitlen)

            if dist_deflate_code > 3 then -- dist code with extra bits
                dist_code_with_extra_count = dist_code_with_extra_count + 1
                local dist_extra_bits = dextra_bits[dist_code_with_extra_count]
                local dist_extra_bits_bitlen =
                    (dist_deflate_code-dist_deflate_code%2)/2 - 1
                WriteBits(dist_extra_bits, dist_extra_bits_bitlen)
            end
        end
    end
end

-- Get the size of fixed block without writing any bits into the writer.
-- @param lcodes literal/LZ77_length deflate codes
-- @param decodes LZ77 distance deflate codes
-- @return the bit length of the fixed block
local function GetFixedHuffmanBlockSize(lcodes, dcodes)
    local block_bitlen = 3
    local length_code_count = 0
    for i=1, #lcodes do
        local code = lcodes[i]
        local huffman_bitlen = _fix_block_literal_huffman_bitlen[code]
        block_bitlen = block_bitlen + huffman_bitlen
        if code > 256 then -- Length code
            length_code_count = length_code_count + 1
            if code > 264 and code < 285 then -- Length code with extra bits
                local extra_bits_bitlen =
                    _literal_deflate_code_to_extra_bitlen[code-256]
                block_bitlen = block_bitlen + extra_bits_bitlen
            end
            local dist_code = dcodes[length_code_count]
            block_bitlen = block_bitlen + 5

            if dist_code > 3 then -- dist code with extra bits
                local dist_extra_bits_bitlen =
                    (dist_code-dist_code%2)/2 - 1
                block_bitlen = block_bitlen + dist_extra_bits_bitlen
            end
        end
    end
    return block_bitlen
end

-- Write fixed block.
-- @param lcodes literal/LZ77_length deflate codes
-- @param decodes LZ77 distance deflate codes
local function CompressFixedHuffmanBlock(WriteBits, is_last_block,
    lcodes, lextra_bits, dcodes, dextra_bits)
    WriteBits(is_last_block and 1 or 0, 1) -- Last block identifier
    WriteBits(1, 2) -- Fixed Huffman block identifier
    local length_code_count = 0
    local length_code_with_extra_count = 0
    local dist_code_with_extra_count = 0
    for i=1, #lcodes do
        local deflate_code = lcodes[i]
        local huffman_code = _fix_block_literal_huffman_code[deflate_code]
        local huffman_bitlen = _fix_block_literal_huffman_bitlen[deflate_code]
        WriteBits(huffman_code, huffman_bitlen)
        if deflate_code > 256 then -- Length code
            length_code_count = length_code_count + 1
            if deflate_code > 264 and deflate_code < 285 then
                -- Length code with extra bits
                length_code_with_extra_count = length_code_with_extra_count + 1
                local extra_bits = lextra_bits[length_code_with_extra_count]
                local extra_bits_bitlen =
                    _literal_deflate_code_to_extra_bitlen[deflate_code-256]
                WriteBits(extra_bits, extra_bits_bitlen)
            end
            -- Write distance code
            local dist_code = dcodes[length_code_count]
            local dist_huffman_code = _fix_block_dist_huffman_code[dist_code]
            WriteBits(dist_huffman_code, 5)

            if dist_code > 3 then -- dist code with extra bits
                dist_code_with_extra_count = dist_code_with_extra_count + 1
                local dist_extra_bits = dextra_bits[dist_code_with_extra_count]
                local dist_extra_bits_bitlen = (dist_code-dist_code%2)/2 - 1
                WriteBits(dist_extra_bits, dist_extra_bits_bitlen)
            end
        end
    end
end

-- Get the size of store block without writing any bits into the writer.
-- @param block_start The start index of the origin input string
-- @param block_end The end index of the origin input string
-- @param Total bit lens had been written into the compressed result before,
-- because store block needs to shift to byte boundary.
-- @return the bit length of the fixed block
local function GetStoreBlockSize(block_start, block_end, total_bitlen)
    assert(block_end-block_start+1 <= 65535)
    local block_bitlen = 3
    total_bitlen = total_bitlen + 3
    local padding_bitlen = (8-total_bitlen%8)%8
    block_bitlen = block_bitlen + padding_bitlen
    block_bitlen = block_bitlen + 32
    block_bitlen = block_bitlen + (block_end - block_start + 1) * 8
    return block_bitlen
end

-- Write the store block.
-- @param ... lots of stuffs
-- @return nil
local function CompressStoreBlock(WriteBits, WriteString, is_last_block, str
    , block_start, block_end, total_bitlen)
    assert(block_end-block_start+1 <= 65535)
    WriteBits(is_last_block and 1 or 0, 1) -- Last block identifer.
    WriteBits(0, 2) -- Store block identifier.
    total_bitlen = total_bitlen + 3
    local padding_bitlen = (8-total_bitlen%8)%8
    if padding_bitlen > 0 then
        WriteBits(_pow2[padding_bitlen]-1, padding_bitlen)
    end
    local size = block_end - block_start + 1
    WriteBits(size, 16)

    -- Write size's one's complement
    local comp = (255 - size % 256) + (255 - (size-size%256)/256)*256
    WriteBits(comp, 16)

    WriteString(str:sub(block_start, block_end))
end

-- Do the deflate
-- Currently using a simple way to determine the block size
-- (This is why the compression ratio is little bit worse than zlib when
-- the input size is very large
-- The first block is 64KB, the following block is 32KB.
-- After each block, there is a memory cleanup operation.
-- This is not a fast operation, but it is needed to save memory usage, so
-- the memory usage does not grow unboundly. If the data size is less than
-- 64KB, then memory cleanup won't happen.
-- This function determines whether to use store/fixed/dynamic blocks by
-- calculating the block size of each block type and chooses the smallest one.
local function Deflate(configs, WriteBits, WriteString, FlushWriter, str
    , dictionary)
    local string_table = {}
    local hash_tables = {}
    local is_last_block = nil
    local block_start
    local block_end
    local bitlen_written
    local total_bitlen = FlushWriter(_FLUSH_MODE_NO_FLUSH)
    local strlen = #str
    local offset

    local level
    local strategy
    if configs then
        if configs.level then
            level = configs.level
        end
        if configs.strategy then
            strategy = configs.strategy
        end
    end

    if not level then
        if strlen < 2048 then
            level = 7
        elseif strlen > 65536 then
            level = 3
        else
            level = 5
        end
    end

    while not is_last_block do
        if not block_start then
            block_start = 1
            block_end = 64*1024 - 1
            offset = 0
        else
            block_start = block_end + 1
            block_end = block_end + 32*1024
            offset = block_start - 32*1024 - 1
        end

        if block_end >= strlen then
            block_end = strlen
            is_last_block = true
        else
            is_last_block = false
        end

        local lcodes, lextra_bits, lcodes_counts, dcodes, dextra_bits
        , dcodes_counts

        local HLIT, HDIST, HCLEN, rle_codes_huffman_bitlens
        , rle_codes_huffman_codes, rle_deflate_codes
        , rle_extra_bits, lcodes_huffman_bitlens, lcodes_huffman_codes
        , dcodes_huffman_bitlens, dcodes_huffman_codes

        local dynamic_block_bitlen
        local fixed_block_bitlen
        local store_block_bitlen

        if level ~= 0 then

            -- GetBlockLZ77 needs block_start to block_end+3 to be loaded.
            LoadStringToTable(str, string_table, block_start, block_end + 3
                , offset)
            if block_start == 1 and dictionary then
                local dict_string_table = dictionary.string_table
                local dict_strlen = dictionary.strlen
                for i=0, (-dict_strlen+1)<-257
                    and -257 or (-dict_strlen+1), -1 do
                    string_table[i] = dict_string_table[dict_strlen+i]
                end
            end

            if strategy == "huffman_only" then
                lcodes = {}
                LoadStringToTable(str, lcodes, block_start, block_end
                    , block_start-1)
                lextra_bits = {}
                lcodes_counts = {}
                lcodes[block_end - block_start+2] = 256 -- end of block
                for i=1, block_end - block_start+2 do
                    local code = lcodes[i]
                    lcodes_counts[code] = (lcodes_counts[code] or 0) + 1
                end
                dcodes = {}
                dextra_bits = {}
                dcodes_counts = {}
            else
                lcodes, lextra_bits, lcodes_counts, dcodes, dextra_bits
                , dcodes_counts = GetBlockLZ77Result(level, string_table
                    , hash_tables, block_start, block_end, offset, dictionary
                )
            end

            HLIT, HDIST, HCLEN, rle_codes_huffman_bitlens
            , rle_codes_huffman_codes, rle_deflate_codes
            , rle_extra_bits, lcodes_huffman_bitlens, lcodes_huffman_codes
            , dcodes_huffman_bitlens, dcodes_huffman_codes =
                GetBlockDynamicHuffmanHeader(lcodes_counts, dcodes_counts)
            dynamic_block_bitlen = GetDynamicHuffmanBlockSize(
                lcodes, dcodes, HCLEN, rle_codes_huffman_bitlens
                , rle_deflate_codes, lcodes_huffman_bitlens
                , dcodes_huffman_bitlens)
            fixed_block_bitlen = GetFixedHuffmanBlockSize(lcodes, dcodes)
        end

        store_block_bitlen = GetStoreBlockSize(block_start, block_end
            , total_bitlen)

        local min_bitlen = store_block_bitlen
        min_bitlen = (fixed_block_bitlen and fixed_block_bitlen < min_bitlen)
        and fixed_block_bitlen or min_bitlen
        min_bitlen = (dynamic_block_bitlen
            and dynamic_block_bitlen < min_bitlen)
        and dynamic_block_bitlen or min_bitlen

        if level == 0 or (strategy ~= "fixed" and strategy ~= "dynamic" and
            store_block_bitlen == min_bitlen) then
            CompressStoreBlock(WriteBits, WriteString, is_last_block
                , str, block_start, block_end, total_bitlen)
            total_bitlen = total_bitlen + store_block_bitlen
        elseif strategy ~= "dynamic" and (
            strategy == "fixed" or fixed_block_bitlen == min_bitlen) then
            CompressFixedHuffmanBlock(WriteBits, is_last_block,
                lcodes, lextra_bits, dcodes, dextra_bits)
            total_bitlen = total_bitlen + fixed_block_bitlen
        elseif strategy == "dynamic" or dynamic_block_bitlen == min_bitlen then
            CompressDynamicHuffmanBlock(WriteBits, is_last_block, lcodes
                , lextra_bits, dcodes, dextra_bits, HLIT, HDIST, HCLEN
                , rle_codes_huffman_bitlens, rle_codes_huffman_codes
                , rle_deflate_codes, rle_extra_bits
                , lcodes_huffman_bitlens, lcodes_huffman_codes
                , dcodes_huffman_bitlens, dcodes_huffman_codes)
            total_bitlen = total_bitlen + dynamic_block_bitlen
        end

        if is_last_block then
            bitlen_written = FlushWriter(_FLUSH_MODE_NO_FLUSH)
        else
            bitlen_written = FlushWriter(_FLUSH_MODE_MEMORY_CLEANUP)
        end

        assert(bitlen_written == total_bitlen)

        -- Memory clean up, so memory consumption does not always grow linearly
        -- , even if input string is > 64K.
        -- Not a very efficient operation, but this operation won't happen
        -- when the input data size is less than 64K.
        if not is_last_block then
            local j
            if dictionary and block_start == 1 then
                j = 0
                while (string_table[j]) do
                    string_table[j] = nil
                    j = j - 1
                end
            end
            dictionary = nil
            j = 1
            for i = block_end-32767, block_end do
                string_table[j] = string_table[i-offset]
                j = j + 1
            end

            for k, t in pairs(hash_tables) do
                local tSize = #t
                if tSize > 0 and block_end+1 - t[1] > 32768 then
                    if tSize == 1 then
                        hash_tables[k] = nil
                    else
                        local new = {}
                        local newSize = 0
                        for i = 2, tSize do
                            j = t[i]
                            if block_end+1 - j <= 32768 then
                                newSize = newSize + 1
                                new[newSize] = j
                            end
                        end
                        hash_tables[k] = new
                    end
                end
            end
        end
    end
end

--- The description to compression configuration table. <br>
-- Any field can be nil to use its default. <br>
-- Table with keys other than those below is an invalid table.
-- @class table
-- @name compression_configs
-- @field level The compression level ranged from 0 to 9. 0 is no compression.
-- 9 is the slowest but best compression. Use nil for default level.
-- @field strategy The compression strategy. "fixed" to only use fixed deflate
-- compression block. "dynamic" to only use dynamic block. "huffman_only" to
-- do no LZ77 compression. Only do huffman compression.


-- @see LibDeflate:CompressDeflate(str, configs)
-- @see LibDeflate:CompressDeflateWithDict(str, dictionary, configs)
local function CompressDeflateInternal(str, dictionary, configs)
    local WriteBits, WriteString, FlushWriter = CreateWriter()
    Deflate(configs, WriteBits, WriteString, FlushWriter, str, dictionary)
    local total_bitlen, result = FlushWriter(_FLUSH_MODE_OUTPUT)
    local padding_bitlen = (8-total_bitlen%8)%8
    return result, padding_bitlen
end

-- @see LibDeflate:CompressZlib
-- @see LibDeflate:CompressZlibWithDict
local function CompressZlibInternal(str, dictionary, configs)
    local WriteBits, WriteString, FlushWriter = CreateWriter()

    local CM = 8 -- Compression method
    local CINFO = 7 --Window Size = 32K
    local CMF = CINFO*16+CM
    WriteBits(CMF, 8)

    local FDIST = dictionary and 1 or 0
    local FLEVEL = 2 -- Default compression
    local FLG = FLEVEL*64+FDIST*32
    local FCHECK = (31-(CMF*256+FLG)%31)
    -- The FCHECK value must be such that CMF and FLG,
    -- when viewed as a 16-bit unsigned integer stored
    -- in MSB order (CMF*256 + FLG), is a multiple of 31.
    FLG = FLG + FCHECK
    WriteBits(FLG, 8)

    if FDIST == 1 then
        local adler32 = dictionary.adler32
        local byte0 = adler32 % 256
        adler32 = (adler32 - byte0) / 256
        local byte1 = adler32 % 256
        adler32 = (adler32 - byte1) / 256
        local byte2 = adler32 % 256
        adler32 = (adler32 - byte2) / 256
        local byte3 = adler32 % 256
        WriteBits(byte3, 8)
        WriteBits(byte2, 8)
        WriteBits(byte1, 8)
        WriteBits(byte0, 8)
    end

    Deflate(configs, WriteBits, WriteString, FlushWriter, str, dictionary)
    FlushWriter(_FLUSH_MODE_BYTE_BOUNDARY)

    local adler32 = LibDeflate:Adler32(str)

    -- Most significant byte first
    local byte3 = adler32%256
    adler32 = (adler32 - byte3) / 256
    local byte2 = adler32%256
    adler32 = (adler32 - byte2) / 256
    local byte1 = adler32%256
    adler32 = (adler32 - byte1) / 256
    local byte0 = adler32%256

    WriteBits(byte0, 8)
    WriteBits(byte1, 8)
    WriteBits(byte2, 8)
    WriteBits(byte3, 8)
    local total_bitlen, result = FlushWriter(_FLUSH_MODE_OUTPUT)
    local padding_bitlen = (8-total_bitlen%8)%8
    return result, padding_bitlen
end

--- Compress using the raw deflate format.
-- @param str [string] The data to be compressed.
-- @param configs [table/nil] The configuration table to control the compression
-- . If nil, use the default configuration.
-- @return [string] The compressed data.
-- @return [integer] The number of bits padded at the end of output.
-- 0 <= bits < 8  <br>
-- This means the most significant "bits" of the last byte of the returned
-- compressed data are padding bits and they don't affect decompression.
-- You don't need to use this value unless you want to do some postprocessing
-- to the compressed data.
-- @see compression_configs
-- @see LibDeflate:DecompressDeflate
function LibDeflate:CompressDeflate(str, configs)
    local arg_valid, arg_err = IsValidArguments(str, false, nil, true, configs)
    if not arg_valid then
        error(("Usage: LibDeflate:CompressDeflate(str, configs): "
            ..arg_err), 2)
    end
    return CompressDeflateInternal(str, nil, configs)
end

--- Compress using the raw deflate format with a preset dictionary.
-- @param str [string] The data to be compressed.
-- @param dictionary [table] The preset dictionary produced by
-- LibDeflate:CreateDictionary
-- @param configs [table/nil] The configuration table to control the compression
-- . If nil, use the default configuration.
-- @return [string] The compressed data.
-- @return [integer] The number of bits padded at the end of output.
-- 0 <= bits < 8  <br>
-- This means the most significant "bits" of the last byte of the returned
-- compressed data are padding bits and they don't affect decompression.
-- You don't need to use this value unless you want to do some postprocessing
-- to the compressed data.
-- @see compression_configs
-- @see LibDeflate:CreateDictionary
-- @see LibDeflate:DecompressDeflateWithDict
function LibDeflate:CompressDeflateWithDict(str, dictionary, configs)
    local arg_valid, arg_err = IsValidArguments(str, true, dictionary
        , true, configs)
    if not arg_valid then
        error(("Usage: LibDeflate:CompressDeflateWithDict"
            .."(str, dictionary, configs): "
            ..arg_err), 2)
    end
    return CompressDeflateInternal(str, dictionary, configs)
end

--- Compress using the zlib format.
-- @param str [string] the data to be compressed.
-- @param configs [table/nil] The configuration table to control the compression
-- . If nil, use the default configuration.
-- @return [string] The compressed data.
-- @return [integer] The number of bits padded at the end of output.
-- Should always be 0.
-- Zlib formatted compressed data never has padding bits at the end.
-- @see compression_configs
-- @see LibDeflate:DecompressZlib
function LibDeflate:CompressZlib(str, configs)
    local arg_valid, arg_err = IsValidArguments(str, false, nil, true, configs)
    if not arg_valid then
        error(("Usage: LibDeflate:CompressZlib(str, configs): "
            ..arg_err), 2)
    end
    return CompressZlibInternal(str, nil, configs)
end

--- Compress using the zlib format with a preset dictionary.
-- @param str [string] the data to be compressed.
-- @param dictionary [table] A preset dictionary produced
-- by LibDeflate:CreateDictionary()
-- @param configs [table/nil] The configuration table to control the compression
-- . If nil, use the default configuration.
-- @return [string] The compressed data.
-- @return [integer] The number of bits padded at the end of output.
-- Should always be 0.
-- Zlib formatted compressed data never has padding bits at the end.
-- @see compression_configs
-- @see LibDeflate:CreateDictionary
-- @see LibDeflate:DecompressZlibWithDict
function LibDeflate:CompressZlibWithDict(str, dictionary, configs)
    local arg_valid, arg_err = IsValidArguments(str, true, dictionary
        , true, configs)
    if not arg_valid then
        error(("Usage: LibDeflate:CompressZlibWithDict"
            .."(str, dictionary, configs): "
            ..arg_err), 2)
    end
    return CompressZlibInternal(str, dictionary, configs)
end

--[[ --------------------------------------------------------------------------
    Decompress code
--]] --------------------------------------------------------------------------

--[[
    Create a reader to easily reader stuffs as the unit of bits.
    Return values:
    1. ReadBits(bitlen)
    2. ReadBytes(bytelen, buffer, buffer_size)
    3. Decode(huffman_bitlen_count, huffman_symbol, min_bitlen)
    4. ReaderBitlenLeft()
    5. SkipToByteBoundary()
--]]
local function CreateReader(input_string)
    local input = input_string
    local input_strlen = #input_string
    local input_next_byte_pos = 1
    local cache_bitlen = 0
    local cache = 0

    -- Read some bits.
    -- To improve speed, this function does not
    -- check if the input has been exhausted.
    -- Use ReaderBitlenLeft() < 0 to check it.
    -- @param bitlen the number of bits to read
    -- @return the data is read.
    local function ReadBits(bitlen)
        local rshift_mask = _pow2[bitlen]
        local code
        if bitlen <= cache_bitlen then
            code = cache % rshift_mask
            cache = (cache - code) / rshift_mask
            cache_bitlen = cache_bitlen - bitlen
        else -- Whether input has been exhausted is not checked.
            local lshift_mask = _pow2[cache_bitlen]
            local byte1, byte2, byte3, byte4 = string_byte(input
                , input_next_byte_pos, input_next_byte_pos+3)
            -- This requires lua number to be at least double ()
            cache = cache + ((byte1 or 0)+(byte2 or 0)*256
                + (byte3 or 0)*65536+(byte4 or 0)*16777216)*lshift_mask
            input_next_byte_pos = input_next_byte_pos + 4
            cache_bitlen = cache_bitlen + 32 - bitlen
            code = cache % rshift_mask
            cache = (cache - code) / rshift_mask
        end
        return code
    end

    -- Read some bytes from the reader.
    -- Assume reader is on the byte boundary.
    -- @param bytelen The number of bytes to be read.
    -- @param buffer The byte read will be stored into this buffer.
    -- @param buffer_size The buffer will be modified starting from
    --  buffer[buffer_size+1], ending at buffer[buffer_size+bytelen-1]
    -- @return the new buffer_size
    local function ReadBytes(bytelen, buffer, buffer_size)
        assert(cache_bitlen % 8 == 0)

        local byte_from_cache = (cache_bitlen/8 < bytelen)
        and (cache_bitlen/8) or bytelen
        for _=1, byte_from_cache do
            local byte = cache % 256
            buffer_size = buffer_size + 1
            buffer[buffer_size] = string_char(byte)
            cache = (cache - byte) / 256
        end
        cache_bitlen = cache_bitlen - byte_from_cache*8
        bytelen = bytelen - byte_from_cache
        if (input_strlen - input_next_byte_pos - bytelen + 1) * 8
            + cache_bitlen < 0 then
            return -1 -- out of input
        end
        for i=input_next_byte_pos, input_next_byte_pos+bytelen-1 do
            buffer_size = buffer_size + 1
            buffer[buffer_size] = string_sub(input, i, i)
        end

        input_next_byte_pos = input_next_byte_pos + bytelen
        return buffer_size
    end

    -- Decode huffman code
    -- To improve speed, this function does not check
    -- if the input has been exhausted.
    -- Use ReaderBitlenLeft() < 0 to check it.
    -- Credits for Mark Adler. This code is from puff:Decode()
    -- @see puff:Decode(...)
    -- @param huffman_bitlen_count
    -- @param huffman_symbol
    -- @param min_bitlen The minimum huffman bit length of all symbols
    -- @return The decoded deflate code.
    --  Negative value is returned if decoding fails.
    local function Decode(huffman_bitlen_counts, huffman_symbols, min_bitlen)
        local code = 0
        local first = 0
        local index = 0
        local count
        if min_bitlen > 0 then
            if cache_bitlen < 15 and input then
                local lshift_mask = _pow2[cache_bitlen]
                local byte1, byte2, byte3, byte4 =
                    string_byte(input, input_next_byte_pos
                        , input_next_byte_pos+3)
                -- This requires lua number to be at least double ()
                cache = cache + ((byte1 or 0)+(byte2 or 0)*256
                    +(byte3 or 0)*65536+(byte4 or 0)*16777216)*lshift_mask
                input_next_byte_pos = input_next_byte_pos + 4
                cache_bitlen = cache_bitlen + 32
            end

            local rshift_mask = _pow2[min_bitlen]
            cache_bitlen = cache_bitlen - min_bitlen
            code = cache % rshift_mask
            cache = (cache - code) / rshift_mask
            -- Reverse the bits
            code = _reverse_bits_tbl[min_bitlen][code]

            count = huffman_bitlen_counts[min_bitlen]
            if code < count then
                return huffman_symbols[code]
            end
            index = count
            first = count * 2
            code = code * 2
        end

        for bitlen = min_bitlen+1, 15 do
            local bit
            bit = cache % 2
            cache = (cache - bit) / 2
            cache_bitlen = cache_bitlen - 1

            code = (bit==1) and (code + 1 - code % 2) or code
            count = huffman_bitlen_counts[bitlen] or 0
            local diff = code - first
            if diff < count then
                return huffman_symbols[index + diff]
            end
            index = index + count
            first = first + count
            first = first * 2
            code = code * 2
        end
        -- invalid literal/length or distance code
        -- in fixed or dynamic block (run out of code)
        return -10
    end

    local function ReaderBitlenLeft()
        return (input_strlen - input_next_byte_pos + 1) * 8 + cache_bitlen
    end

    local function SkipToByteBoundary()
        local skipped_bitlen = cache_bitlen%8
        local rshift_mask = _pow2[skipped_bitlen]
        cache_bitlen = cache_bitlen - skipped_bitlen
        cache = (cache - cache % rshift_mask) / rshift_mask
    end

    return ReadBits, ReadBytes, Decode, ReaderBitlenLeft, SkipToByteBoundary
end

-- Create a deflate state, so I can pass in less arguments to functions.
-- @param str the whole string to be decompressed.
-- @param dictionary The preset dictionary. nil if not provided.
--      This dictionary should be produced by LibDeflate:CreateDictionary(str)
-- @return The decomrpess state.
local function CreateDecompressState(str, dictionary)
    local ReadBits, ReadBytes, Decode, ReaderBitlenLeft
    , SkipToByteBoundary = CreateReader(str)
    local state =
    {
        ReadBits = ReadBits,
        ReadBytes = ReadBytes,
        Decode = Decode,
        ReaderBitlenLeft = ReaderBitlenLeft,
        SkipToByteBoundary = SkipToByteBoundary,
        buffer_size = 0,
        buffer = {},
        result_buffer = {},
        dictionary = dictionary,
    }
    return state
end

-- Get the stuffs needed to decode huffman codes
-- @see puff.c:construct(...)
-- @param huffman_bitlen The huffman bit length of the huffman codes.
-- @param max_symbol The maximum symbol
-- @param max_bitlen The min huffman bit length of all codes
-- @return zero or positive for success, negative for failure.
-- @return The count of each huffman bit length.
-- @return A table to convert huffman codes to deflate codes.
-- @return The minimum huffman bit length.
local function GetHuffmanForDecode(huffman_bitlens, max_symbol, max_bitlen)
    local huffman_bitlen_counts = {}
    local min_bitlen = max_bitlen
    for symbol = 0, max_symbol do
        local bitlen = huffman_bitlens[symbol] or 0
        min_bitlen = (bitlen > 0 and bitlen < min_bitlen)
        and bitlen or min_bitlen
        huffman_bitlen_counts[bitlen] = (huffman_bitlen_counts[bitlen] or 0)+1
    end

    if huffman_bitlen_counts[0] == max_symbol+1 then -- No Codes
        return 0, huffman_bitlen_counts, {}, 0 -- Complete, but decode will fail
    end

    local left = 1
    for len = 1, max_bitlen do
        left = left * 2
        left = left - (huffman_bitlen_counts[len] or 0)
        if left < 0 then
            return left -- Over-subscribed, return negative
        end
    end

    -- Generate offsets info symbol table for each length for sorting
    local offsets = {}
    offsets[1] = 0
    for len = 1, max_bitlen-1 do
        offsets[len + 1] = offsets[len] + (huffman_bitlen_counts[len] or 0)
    end

    local huffman_symbols = {}
    for symbol = 0, max_symbol do
        local bitlen = huffman_bitlens[symbol] or 0
        if bitlen ~= 0 then
            local offset = offsets[bitlen]
            huffman_symbols[offset] = symbol
            offsets[bitlen] = offsets[bitlen] + 1
        end
    end

    -- Return zero for complete set, positive for incomplete set.
    return left, huffman_bitlen_counts, huffman_symbols, min_bitlen
end

-- Decode a fixed or dynamic huffman blocks, excluding last block identifier
-- and block type identifer.
-- @see puff.c:codes()
-- @param state decompression state that will be modified by this function.
--  @see CreateDecompressState
-- @param ... Read the source code
-- @return 0 on success, other value on failure.
local function DecodeUntilEndOfBlock(state, lcodes_huffman_bitlens
    , lcodes_huffman_symbols, lcodes_huffman_min_bitlen
    , dcodes_huffman_bitlens, dcodes_huffman_symbols
    , dcodes_huffman_min_bitlen)
    local buffer, buffer_size, ReadBits, Decode, ReaderBitlenLeft
    , result_buffer =
        state.buffer, state.buffer_size, state.ReadBits, state.Decode
    , state.ReaderBitlenLeft, state.result_buffer
    local dictionary = state.dictionary
    local dict_string_table
    local dict_strlen

    local buffer_end = 1
    if dictionary and not buffer[0] then
        -- If there is a dictionary, copy the last 258 bytes into
        -- the string_table to make the copy in the main loop quicker.
        -- This is done only once per decompression.
        dict_string_table = dictionary.string_table
        dict_strlen = dictionary.strlen
        buffer_end = -dict_strlen + 1
        for i=0, (-dict_strlen+1)<-257 and -257 or (-dict_strlen+1), -1 do
            buffer[i] = _byte_to_char[dict_string_table[dict_strlen+i]]
        end
    end

    repeat
        local symbol = Decode(lcodes_huffman_bitlens
            , lcodes_huffman_symbols, lcodes_huffman_min_bitlen)
        if symbol < 0 or symbol > 285 then
            -- invalid literal/length or distance code in fixed or dynamic block
            return -10
        elseif symbol < 256 then -- Literal
            buffer_size = buffer_size + 1
            buffer[buffer_size] = _byte_to_char[symbol]
        elseif symbol > 256 then -- Length code
            symbol = symbol - 256
            local bitlen = _literal_deflate_code_to_base_len[symbol]
            bitlen = (symbol >= 8)
            and (bitlen
                + ReadBits(_literal_deflate_code_to_extra_bitlen[symbol]))
            or bitlen
            symbol = Decode(dcodes_huffman_bitlens, dcodes_huffman_symbols
                , dcodes_huffman_min_bitlen)
            if symbol < 0 or symbol > 29 then
                -- invalid literal/length or distance code in fixed or dynamic block
                return -10
            end
            local dist = _dist_deflate_code_to_base_dist[symbol]
            dist = (dist > 4) and (dist
            + ReadBits(_dist_deflate_code_to_extra_bitlen[symbol])) or dist

            local char_buffer_index = buffer_size-dist+1
            if char_buffer_index < buffer_end then
                -- distance is too far back in fixed or dynamic block
                return -11
            end
            if char_buffer_index >= -257 then
                for _=1, bitlen do
                    buffer_size = buffer_size + 1
                    buffer[buffer_size] = buffer[char_buffer_index]
                    char_buffer_index = char_buffer_index + 1
                end
            else
                char_buffer_index = dict_strlen + char_buffer_index
                for _=1, bitlen do
                    buffer_size = buffer_size + 1
                    buffer[buffer_size] =
                        _byte_to_char[dict_string_table[char_buffer_index]]
                    char_buffer_index = char_buffer_index + 1
                end
            end
        end

        if ReaderBitlenLeft() < 0 then
            return 2 -- available inflate data did not terminate
        end

        if buffer_size >= 65536 then
            result_buffer[#result_buffer+1] =
                table_concat(buffer, "", 1, 32768)
            for i=32769, buffer_size do
                buffer[i-32768] = buffer[i]
            end
            buffer_size = buffer_size - 32768
            buffer[buffer_size+1] = nil
            -- NOTE: buffer[32769..end] and buffer[-257..0] are not cleared.
            -- This is why "buffer_size" variable is needed.
        end
    until symbol == 256

    state.buffer_size = buffer_size

    return 0
end

-- Decompress a store block
-- @param state decompression state that will be modified by this function.
-- @return 0 if succeeds, other value if fails.
local function DecompressStoreBlock(state)
    local buffer, buffer_size, ReadBits, ReadBytes, ReaderBitlenLeft
    , SkipToByteBoundary, result_buffer =
        state.buffer, state.buffer_size, state.ReadBits, state.ReadBytes
    , state.ReaderBitlenLeft, state.SkipToByteBoundary, state.result_buffer

    SkipToByteBoundary()
    local bytelen = ReadBits(16)
    if ReaderBitlenLeft() < 0 then
        return 2 -- available inflate data did not terminate
    end
    local bytelenComp = ReadBits(16)
    if ReaderBitlenLeft() < 0 then
        return 2 -- available inflate data did not terminate
    end

    if bytelen % 256 + bytelenComp % 256 ~= 255 then
        return -2 -- Not one's complement
    end
    if (bytelen-bytelen % 256)/256
        + (bytelenComp-bytelenComp % 256)/256 ~= 255 then
        return -2 -- Not one's complement
    end

    -- Note that ReadBytes will skip to the next byte boundary first.
    buffer_size = ReadBytes(bytelen, buffer, buffer_size)
    if buffer_size < 0 then
        return 2 -- available inflate data did not terminate
    end

    -- memory clean up when there are enough bytes in the buffer.
    if buffer_size >= 65536 then
        result_buffer[#result_buffer+1] = table_concat(buffer, "", 1, 32768)
        for i=32769, buffer_size do
            buffer[i-32768] = buffer[i]
        end
        buffer_size = buffer_size - 32768
        buffer[buffer_size+1] = nil
    end
    state.buffer_size = buffer_size
    return 0
end

-- Decompress a fixed block
-- @param state decompression state that will be modified by this function.
-- @return 0 if succeeds other value if fails.
local function DecompressFixBlock(state)
    return DecodeUntilEndOfBlock(state
        , _fix_block_literal_huffman_bitlen_count
        , _fix_block_literal_huffman_to_deflate_code, 7
        , _fix_block_dist_huffman_bitlen_count
        , _fix_block_dist_huffman_to_deflate_code, 5)
end

-- Decompress a dynamic block
-- @param state decompression state that will be modified by this function.
-- @return 0 if success, other value if fails.
local function DecompressDynamicBlock(state)
    local ReadBits, Decode = state.ReadBits, state.Decode
    local nlen = ReadBits(5) + 257
    local ndist = ReadBits(5) + 1
    local ncode = ReadBits(4) + 4
    if nlen > 286 or ndist > 30 then
        -- dynamic block code description: too many length or distance codes
        return -3
    end

    local rle_codes_huffman_bitlens = {}

    for i = 1, ncode do
        rle_codes_huffman_bitlens[_rle_codes_huffman_bitlen_order[i]] =
            ReadBits(3)
    end

    local rle_codes_err, rle_codes_huffman_bitlen_counts,
    rle_codes_huffman_symbols, rle_codes_huffman_min_bitlen =
        GetHuffmanForDecode(rle_codes_huffman_bitlens, 18, 7)
    if rle_codes_err ~= 0 then -- Require complete code set here
        -- dynamic block code description: code lengths codes incomplete
        return -4
    end

    local lcodes_huffman_bitlens = {}
    local dcodes_huffman_bitlens = {}
    -- Read length/literal and distance code length tables
    local index = 0
    while index < nlen + ndist do
        local symbol -- Decoded value
        local bitlen -- Last length to repeat

        symbol = Decode(rle_codes_huffman_bitlen_counts
            , rle_codes_huffman_symbols, rle_codes_huffman_min_bitlen)

        if symbol < 0 then
            return symbol -- Invalid symbol
        elseif symbol < 16 then
            if index < nlen then
                lcodes_huffman_bitlens[index] = symbol
            else
                dcodes_huffman_bitlens[index-nlen] = symbol
            end
            index = index + 1
        else
            bitlen = 0
            if symbol == 16 then
                if index == 0 then
                    -- dynamic block code description: repeat lengths
                    -- with no first length
                    return -5
                end
                if index-1 < nlen then
                    bitlen = lcodes_huffman_bitlens[index-1]
                else
                    bitlen = dcodes_huffman_bitlens[index-nlen-1]
                end
                symbol = 3 + ReadBits(2)
            elseif symbol == 17 then -- Repeat zero 3..10 times
                symbol = 3 + ReadBits(3)
            else -- == 18, repeat zero 11.138 times
                symbol = 11 + ReadBits(7)
            end
            if index + symbol > nlen + ndist then
                -- dynamic block code description:
                -- repeat more than specified lengths
                return -6
            end
            while symbol > 0 do -- Repeat last or zero symbol times
                symbol = symbol - 1
                if index < nlen then
                    lcodes_huffman_bitlens[index] = bitlen
                else
                    dcodes_huffman_bitlens[index-nlen] = bitlen
                end
                index = index + 1
            end
        end
    end

    if (lcodes_huffman_bitlens[256] or 0) == 0 then
        -- dynamic block code description: missing end-of-block code
        return -9
    end

    local lcodes_err, lcodes_huffman_bitlen_counts
    , lcodes_huffman_symbols, lcodes_huffman_min_bitlen =
        GetHuffmanForDecode(lcodes_huffman_bitlens, nlen-1, 15)
    --dynamic block code description: invalid literal/length code lengths,
    -- Incomplete code ok only for single length 1 code
    if (lcodes_err ~=0 and (lcodes_err < 0
        or nlen ~= (lcodes_huffman_bitlen_counts[0] or 0)
        +(lcodes_huffman_bitlen_counts[1] or 0))) then
        return -7
    end

    local dcodes_err, dcodes_huffman_bitlen_counts
    , dcodes_huffman_symbols, dcodes_huffman_min_bitlen =
        GetHuffmanForDecode(dcodes_huffman_bitlens, ndist-1, 15)
    -- dynamic block code description: invalid distance code lengths,
    -- Incomplete code ok only for single length 1 code
    if (dcodes_err ~=0 and (dcodes_err < 0
        or ndist ~= (dcodes_huffman_bitlen_counts[0] or 0)
        + (dcodes_huffman_bitlen_counts[1] or 0))) then
        return -8
    end

    -- Build buffman table for literal/length codes
    return DecodeUntilEndOfBlock(state, lcodes_huffman_bitlen_counts
        , lcodes_huffman_symbols, lcodes_huffman_min_bitlen
        , dcodes_huffman_bitlen_counts, dcodes_huffman_symbols
        , dcodes_huffman_min_bitlen)
end

-- Decompress a deflate stream
-- @param state: a decompression state
-- @return the decompressed string if succeeds. nil if fails.
local function Inflate(state)
    local ReadBits = state.ReadBits

    local is_last_block
    while not is_last_block do
        is_last_block = (ReadBits(1) == 1)
        local block_type = ReadBits(2)
        local status
        if block_type == 0 then
            status = DecompressStoreBlock(state)
        elseif block_type == 1 then
            status = DecompressFixBlock(state)
        elseif block_type == 2 then
            status = DecompressDynamicBlock(state)
        else
            return nil, -1 -- invalid block type (type == 3)
        end
        if status ~= 0 then
            return nil, status
        end
    end

    state.result_buffer[#state.result_buffer+1] =
        table_concat(state.buffer, "", 1, state.buffer_size)
    local result = table_concat(state.result_buffer)
    return result
end

-- @see LibDeflate:DecompressDeflate(str)
-- @see LibDeflate:DecompressDeflateWithDict(str, dictionary)
local function DecompressDeflateInternal(str, dictionary)
    local state = CreateDecompressState(str, dictionary)
    local result, status = Inflate(state)
    if not result then
        return nil, status
    end

    local bitlen_left = state.ReaderBitlenLeft()
    local bytelen_left = (bitlen_left - bitlen_left % 8) / 8
    return result, bytelen_left
end

-- @see LibDeflate:DecompressZlib(str)
-- @see LibDeflate:DecompressZlibWithDict(str)
local function DecompressZlibInternal(str, dictionary)
    local state = CreateDecompressState(str, dictionary)
    local ReadBits = state.ReadBits

    local CMF = ReadBits(8)
    if state.ReaderBitlenLeft() < 0 then
        return nil, 2 -- available inflate data did not terminate
    end
    local CM = CMF % 16
    local CINFO = (CMF - CM) / 16
    if CM ~= 8 then
        return nil, -12 -- invalid compression method
    end
    if CINFO > 7 then
        return nil, -13 -- invalid window size
    end

    local FLG = ReadBits(8)
    if state.ReaderBitlenLeft() < 0 then
        return nil, 2 -- available inflate data did not terminate
    end
    if (CMF*256+FLG)%31 ~= 0 then
        return nil, -14 -- invalid header checksum
    end

    local FDIST = ((FLG-FLG%32)/32 % 2)
    local FLEVEL = ((FLG-FLG%64)/64 % 4) -- luacheck: ignore FLEVEL

    if FDIST == 1 then
        if not dictionary then
            return nil, -16 -- need dictonary, but dictionary is not provided.
        end
        local byte3 = ReadBits(8)
        local byte2 = ReadBits(8)
        local byte1 = ReadBits(8)
        local byte0 = ReadBits(8)
        local actual_adler32 = byte3*16777216+byte2*65536+byte1*256+byte0
        if state.ReaderBitlenLeft() < 0 then
            return nil, 2 -- available inflate data did not terminate
        end
        if not IsEqualAdler32(actual_adler32, dictionary.adler32) then
            return nil, -17 -- dictionary adler32 does not match
        end
    end
    local result, status = Inflate(state)
    if not result then
        return nil, status
    end
    state.SkipToByteBoundary()

    local adler_byte0 = ReadBits(8)
    local adler_byte1 = ReadBits(8)
    local adler_byte2 = ReadBits(8)
    local adler_byte3 = ReadBits(8)
    if state.ReaderBitlenLeft() < 0 then
        return nil, 2 -- available inflate data did not terminate
    end

    local adler32_expected = adler_byte0*16777216
    + adler_byte1*65536 + adler_byte2*256 + adler_byte3
    local adler32_actual = LibDeflate:Adler32(result)
    if not IsEqualAdler32(adler32_expected, adler32_actual) then
        return nil, -15 -- Adler32 checksum does not match
    end

    local bitlen_left = state.ReaderBitlenLeft()
    local bytelen_left = (bitlen_left - bitlen_left % 8) / 8
    return result, bytelen_left
end

--- Decompress a raw deflate compressed data.
-- @param str [string] The data to be decompressed.
-- @return [string/nil] If the decompression succeeds, return the decompressed
-- data. If the decompression fails, return nil. You should check if this return
-- value is non-nil to know if the decompression succeeds.
-- @return [integer] If the decompression succeeds, return the number of
-- unprocessed bytes in the input compressed data. This return value is a
-- positive integer if the input data is a valid compressed data appended by an
-- arbitary non-empty string. This return value is 0 if the input data does not
-- contain any extra bytes.<br>
-- If the decompression fails (The first return value of this function is nil),
-- this return value is undefined.
-- @see LibDeflate:CompressDeflate
function LibDeflate:DecompressDeflate(str)
    local arg_valid, arg_err = IsValidArguments(str)
    if not arg_valid then
        error(("Usage: LibDeflate:DecompressDeflate(str): "
            ..arg_err), 2)
    end
    return DecompressDeflateInternal(str)
end

--- Decompress a raw deflate compressed data with a preset dictionary.
-- @param str [string] The data to be decompressed.
-- @param dictionary [table] The preset dictionary used by
-- LibDeflate:CompressDeflateWithDict when the compressed data is produced.
-- Decompression and compression must use the same dictionary.
-- Otherwise wrong decompressed data could be produced without generating any
-- error.
-- @return [string/nil] If the decompression succeeds, return the decompressed
-- data. If the decompression fails, return nil. You should check if this return
-- value is non-nil to know if the decompression succeeds.
-- @return [integer] If the decompression succeeds, return the number of
-- unprocessed bytes in the input compressed data. This return value is a
-- positive integer if the input data is a valid compressed data appended by an
-- arbitary non-empty string. This return value is 0 if the input data does not
-- contain any extra bytes.<br>
-- If the decompression fails (The first return value of this function is nil),
-- this return value is undefined.
-- @see LibDeflate:CompressDeflateWithDict
function LibDeflate:DecompressDeflateWithDict(str, dictionary)
    local arg_valid, arg_err = IsValidArguments(str, true, dictionary)
    if not arg_valid then
        error(("Usage: LibDeflate:DecompressDeflateWithDict(str, dictionary): "
            ..arg_err), 2)
    end
    return DecompressDeflateInternal(str, dictionary)
end

--- Decompress a zlib compressed data.
-- @param str [string] The data to be decompressed
-- @return [string/nil] If the decompression succeeds, return the decompressed
-- data. If the decompression fails, return nil. You should check if this return
-- value is non-nil to know if the decompression succeeds.
-- @return [integer] If the decompression succeeds, return the number of
-- unprocessed bytes in the input compressed data. This return value is a
-- positive integer if the input data is a valid compressed data appended by an
-- arbitary non-empty string. This return value is 0 if the input data does not
-- contain any extra bytes.<br>
-- If the decompression fails (The first return value of this function is nil),
-- this return value is undefined.
-- @see LibDeflate:CompressZlib
function LibDeflate:DecompressZlib(str)
    local arg_valid, arg_err = IsValidArguments(str)
    if not arg_valid then
        error(("Usage: LibDeflate:DecompressZlib(str): "
            ..arg_err), 2)
    end
    return DecompressZlibInternal(str)
end

--- Decompress a zlib compressed data with a preset dictionary.
-- @param str [string] The data to be decompressed
-- @param dictionary [table] The preset dictionary used by
-- LibDeflate:CompressDeflateWithDict when the compressed data is produced.
-- Decompression and compression must use the same dictionary.
-- Otherwise wrong decompressed data could be produced without generating any
-- error.
-- @return [string/nil] If the decompression succeeds, return the decompressed
-- data. If the decompression fails, return nil. You should check if this return
-- value is non-nil to know if the decompression succeeds.
-- @return [integer] If the decompression succeeds, return the number of
-- unprocessed bytes in the input compressed data. This return value is a
-- positive integer if the input data is a valid compressed data appended by an
-- arbitary non-empty string. This return value is 0 if the input data does not
-- contain any extra bytes.<br>
-- If the decompression fails (The first return value of this function is nil),
-- this return value is undefined.
-- @see LibDeflate:CompressZlibWithDict
function LibDeflate:DecompressZlibWithDict(str, dictionary)
    local arg_valid, arg_err = IsValidArguments(str, true, dictionary)
    if not arg_valid then
        error(("Usage: LibDeflate:DecompressZlibWithDict(str, dictionary): "
            ..arg_err), 2)
    end
    return DecompressZlibInternal(str, dictionary)
end

-- Calculate the huffman code of fixed block
do
    _fix_block_literal_huffman_bitlen = {}
    for sym=0, 143 do
        _fix_block_literal_huffman_bitlen[sym] = 8
    end
    for sym=144, 255 do
        _fix_block_literal_huffman_bitlen[sym] = 9
    end
    for sym=256, 279 do
        _fix_block_literal_huffman_bitlen[sym] = 7
    end
    for sym=280, 287 do
        _fix_block_literal_huffman_bitlen[sym] = 8
    end

    _fix_block_dist_huffman_bitlen = {}
    for dist=0, 31 do
        _fix_block_dist_huffman_bitlen[dist] = 5
    end
    local status
    status, _fix_block_literal_huffman_bitlen_count
    , _fix_block_literal_huffman_to_deflate_code =
        GetHuffmanForDecode(_fix_block_literal_huffman_bitlen, 287, 9)
    assert(status == 0)
    status, _fix_block_dist_huffman_bitlen_count,
        _fix_block_dist_huffman_to_deflate_code =
        GetHuffmanForDecode(_fix_block_dist_huffman_bitlen, 31, 5)
    assert(status == 0)

    _fix_block_literal_huffman_code =
        GetHuffmanCodeFromBitlen(_fix_block_literal_huffman_bitlen_count
            , _fix_block_literal_huffman_bitlen, 287, 9)
    _fix_block_dist_huffman_code =
        GetHuffmanCodeFromBitlen(_fix_block_dist_huffman_bitlen_count
            , _fix_block_dist_huffman_bitlen, 31, 5)
end

-- Prefix encoding algorithm
-- Credits to LibCompress.
-- The code has been rewritten by the author of LibDeflate.
------------------------------------------------------------------------------

-- to be able to match any requested byte value, the search
-- string must be preprocessed characters to escape with %:
-- ( ) . % + - * ? [ ] ^ $
-- "illegal" byte values:
-- 0 is replaces %z
local _gsub_escape_table = {
    ["\000"] = "%z", ["("] = "%(", [")"] = "%)", ["."] = "%.",
    ["%"] = "%%", ["+"] = "%+", ["-"] = "%-", ["*"] = "%*",
    ["?"] = "%?", ["["] = "%[", ["]"] = "%]", ["^"] = "%^",
    ["$"] = "%$",
}

local function escape_for_gsub(str)
    return str:gsub("([%z%(%)%.%%%+%-%*%?%[%]%^%$])", _gsub_escape_table)
end

--- Create a custom codec with encoder and decoder. <br>
-- This codec is used to convert an input string to make it not contain
-- some specific bytes.
-- This created codec and the parameters of this function do NOT take
-- localization into account. One byte (0-255) in the string is exactly one
-- character (0-255).
-- Credits to LibCompress.
-- The code has been rewritten by the author of LibDeflate. <br>
-- @param reserved_chars [string] The created encoder will ensure encoded
-- data does not contain any single character in reserved_chars. This parameter
-- should be non-empty.
-- @param escape_chars [string] The escape character(s) used in the created
-- codec. The codec converts any character included in reserved\_chars /
-- escape\_chars / map\_chars to (one escape char + one character not in
-- reserved\_chars / escape\_chars / map\_chars).
-- You usually only need to provide a length-1 string for this parameter.
-- Length-2 string is only needed when
-- reserved\_chars + escape\_chars + map\_chars is longer than 127.
-- This parameter should be non-empty.
-- @param map_chars [string] The created encoder will map every
-- reserved\_chars:sub(i, i) (1 <= i <= #map\_chars) to map\_chars:sub(i, i).
-- This parameter CAN be empty string.
-- @return [table/nil] If the codec cannot be created, return nil.<br>
-- If the codec can be created according to the given
-- parameters, return the codec, which is a encode/decode table.
-- The table contains two functions: <br>
-- t:Encode(str) returns the encoded string. <br>
-- t:Decode(str) returns the decoded string if succeeds. nil if fails.
-- @return [nil/string] If the codec is successfully created, return nil.
-- If not, return a string that describes the reason why the codec cannot be
-- created.
-- @usage
-- -- Create an encoder/decoder that maps all "\000" to "\003",
-- -- and escape "\001" (and "\002" and "\003") properly
-- local codec = LibDeflate:CreateCodec("\000\001", "\002", "\003")
--
-- local encoded = codec:Encode(SOME_STRING)
-- -- "encoded" does not contain "\000" or "\001"
-- local decoded = codec:Decode(encoded)
-- -- assert(decoded == SOME_STRING)
function LibDeflate:CreateCodec(reserved_chars, escape_chars
    , map_chars)
    if type(reserved_chars) ~= "string"
        or type(escape_chars) ~= "string"
        or type(map_chars) ~= "string" then
        error(
            "Usage: LibDeflate:CreateCodec(reserved_chars,"
            .." escape_chars, map_chars):"
            .." All arguments must be string.", 2)
    end

    if escape_chars == "" then
        return nil, "No escape characters supplied."
    end
    if #reserved_chars < #map_chars then
        return nil, "The number of reserved characters must be"
        .." at least as many as the number of mapped chars."
    end
    if reserved_chars == "" then
        return nil, "No characters to encode."
    end

    local encode_bytes = reserved_chars..escape_chars..map_chars
    -- build list of bytes not available as a suffix to a prefix byte
    local taken = {}
    for i = 1, #encode_bytes do
        local byte = string_byte(encode_bytes, i, i)
        if taken[byte] then
            return nil, "There must be no duplicate characters in the"
            .." concatenation of reserved_chars, escape_chars and"
            .." map_chars."
        end
        taken[byte] = true
    end

    local decode_patterns = {}
    local decode_repls = {}

    -- the encoding can be a single gsub
    -- , but the decoding can require multiple gsubs
    local encode_search = {}
    local encode_translate = {}

    -- map single byte to single byte
    if #map_chars > 0 then
        local decode_search = {}
        local decode_translate = {}
        for i = 1, #map_chars do
            local from = string_sub(reserved_chars, i, i)
            local to = string_sub(map_chars, i, i)
            encode_translate[from] = to
            encode_search[#encode_search+1] = from
            decode_translate[to] = from
            decode_search[#decode_search+1] = to
        end
        decode_patterns[#decode_patterns+1] =
            "([".. escape_for_gsub(table_concat(decode_search)).."])"
        decode_repls[#decode_repls+1] = decode_translate
    end

    local escape_char_index = 1
    local escape_char = string_sub(escape_chars
        , escape_char_index, escape_char_index)
    -- map single byte to double-byte
    local r = 0 -- suffix char value to the escapeChar

    local decode_search = {}
    local decode_translate = {}
    for i = 1, #encode_bytes do
        local c = string_sub(encode_bytes, i, i)
        if not encode_translate[c] then
            while r >= 256 or taken[r] do
                r = r + 1
                if r > 255 then -- switch to next escapeChar
                    decode_patterns[#decode_patterns+1] =
                        escape_for_gsub(escape_char)
                        .."(["
                        .. escape_for_gsub(table_concat(decode_search)).."])"
                    decode_repls[#decode_repls+1] = decode_translate

                    escape_char_index = escape_char_index + 1
                    escape_char = string_sub(escape_chars, escape_char_index
                        , escape_char_index)
                    r = 0
                    decode_search = {}
                    decode_translate = {}

                    if not escape_char or escape_char == "" then
                        -- actually I don't need to check
                        -- "not ecape_char", but what if Lua changes
                        -- the behavior of string.sub() in the future?
                        -- we are out of escape chars and we need more!
                        return nil, "Out of escape characters."
                    end
                end
            end

            local char_r = _byte_to_char[r]
            encode_translate[c] = escape_char..char_r
            encode_search[#encode_search+1] = c
            decode_translate[char_r] = c
            decode_search[#decode_search+1] = char_r
            r = r + 1
        end
        if i == #encode_bytes then
            decode_patterns[#decode_patterns+1] =
                escape_for_gsub(escape_char).."(["
                .. escape_for_gsub(table_concat(decode_search)).."])"
            decode_repls[#decode_repls+1] = decode_translate
        end
    end

    local codec = {}

    local encode_pattern = "(["
    .. escape_for_gsub(table_concat(encode_search)).."])"
    local encode_repl = encode_translate

    function codec:Encode(str)
        if type(str) ~= "string" then
            error(("Usage: codec:Encode(str):"
                .." 'str' - string expected got '%s'."):format(type(str)), 2)
        end
        return string_gsub(str, encode_pattern, encode_repl)
    end

    local decode_tblsize = #decode_patterns
    local decode_fail_pattern = "(["
    .. escape_for_gsub(reserved_chars).."])"

    function codec:Decode(str)
        if type(str) ~= "string" then
            error(("Usage: codec:Decode(str):"
                .." 'str' - string expected got '%s'."):format(type(str)), 2)
        end
        if string_find(str, decode_fail_pattern) then
            return nil
        end
        for i = 1, decode_tblsize do
            str = string_gsub(str, decode_patterns[i], decode_repls[i])
        end
        return str
    end

    return codec
end

local _addon_channel_codec


-- Credits to WeakAuras2 and Galmok for the 6 bit encoding algorithm.
-- The code has been rewritten by the author of LibDeflate.
-- The result of encoding will be 25% larger than the
-- origin string, but every single byte of the encoding result will be
-- printable characters as the following.
local _byte_to_6bit_char = {
    [0]="a", "b", "c", "d", "e", "f", "g", "h",
    "i", "j", "k", "l", "m", "n", "o", "p",
    "q", "r", "s", "t", "u", "v", "w", "x",
    "y", "z", "A", "B", "C", "D", "E", "F",
    "G", "H", "I", "J", "K", "L", "M", "N",
    "O", "P", "Q", "R", "S", "T", "U", "V",
    "W", "X", "Y", "Z", "0", "1", "2", "3",
    "4", "5", "6", "7", "8", "9", "(", ")",
}

local _6bit_to_byte = {
    [97]=0,[98]=1,[99]=2,[100]=3,[101]=4,[102]=5,[103]=6,[104]=7,
    [105]=8,[106]=9,[107]=10,[108]=11,[109]=12,[110]=13,[111]=14,[112]=15,
    [113]=16,[114]=17,[115]=18,[116]=19,[117]=20,[118]=21,[119]=22,[120]=23,
    [121]=24,[122]=25,[65]=26,[66]=27,[67]=28,[68]=29,[69]=30,[70]=31,
    [71]=32,[72]=33,[73]=34,[74]=35,[75]=36,[76]=37,[77]=38,[78]=39,
    [79]=40,[80]=41,[81]=42,[82]=43,[83]=44,[84]=45,[85]=46,[86]=47,
    [87]=48,[88]=49,[89]=50,[90]=51,[48]=52,[49]=53,[50]=54,[51]=55,
    [52]=56,[53]=57,[54]=58,[55]=59,[56]=60,[57]=61,[40]=62,[41]=63,
}

--- Encode the string to make it printable. <br>
--
-- Credit to WeakAuras2, this function is equivalant to the implementation
-- it is using right now. <br>
-- The code has been rewritten by the author of LibDeflate. <br>
-- The encoded string will be 25% larger than the origin string. However, every
-- single byte of the encoded string will be one of 64 printable ASCII
-- characters, which are can be easier copied, pasted and displayed.
-- (26 lowercase letters, 26 uppercase letters, 10 numbers digits,
-- left parenthese, or right parenthese)
-- @param str [string] The string to be encoded.
-- @return [string] The encoded string.
function LibDeflate:EncodeForPrint(str)
    if type(str) ~= "string" then
        error(("Usage: LibDeflate:EncodeForPrint(str):"
            .." 'str' - string expected got '%s'."):format(type(str)), 2)
    end
    local strlen = #str
    local strlenMinus2 = strlen - 2
    local i = 1
    local buffer = {}
    local buffer_size = 0
    while i <= strlenMinus2 do
        local x1, x2, x3 = string_byte(str, i, i+2)
        i = i + 3
        local cache = x1+x2*256+x3*65536
        local b1 = cache % 64
        cache = (cache - b1) / 64
        local b2 = cache % 64
        cache = (cache - b2) / 64
        local b3 = cache % 64
        local b4 = (cache - b3) / 64
        buffer_size = buffer_size + 1
        buffer[buffer_size] =
            _byte_to_6bit_char[b1].._byte_to_6bit_char[b2]
            .._byte_to_6bit_char[b3].._byte_to_6bit_char[b4]
    end

    local cache = 0
    local cache_bitlen = 0
    while i <= strlen do
        local x = string_byte(str, i, i)
        cache = cache + x * _pow2[cache_bitlen]
        cache_bitlen = cache_bitlen + 8
        i = i + 1
    end
    while cache_bitlen > 0 do
        local bit6 = cache % 64
        buffer_size = buffer_size + 1
        buffer[buffer_size] = _byte_to_6bit_char[bit6]
        cache = (cache - bit6) / 64
        cache_bitlen = cache_bitlen - 6
    end

    return table_concat(buffer)
end

--- Decode the printable string produced by LibDeflate:EncodeForPrint.
-- "str" will have its prefixed and trailing control characters or space
-- removed before it is decoded, so it is easier to use if "str" comes form
-- user copy and paste with some prefixed or trailing spaces.
-- Then decode fails if the string contains any characters cant be produced by
-- LibDeflate:EncodeForPrint. That means, decode fails if the string contains a
-- characters NOT one of 26 lowercase letters, 26 uppercase letters,
-- 10 numbers digits, left parenthese, or right parenthese.
-- @param str [string] The string to be decoded
-- @return [string/nil] The decoded string if succeeds. nil if fails.
function LibDeflate:DecodeForPrint(str)
    if type(str) ~= "string" then
        error(("Usage: LibDeflate:DecodeForPrint(str):"
            .." 'str' - string expected got '%s'."):format(type(str)), 2)
    end
    str = str:gsub("^[%c ]+", "")
    str = str:gsub("[%c ]+$", "")

    local strlen = #str
    if strlen == 1 then
        return nil
    end
    local strlenMinus3 = strlen - 3
    local i = 1
    local buffer = {}
    local buffer_size = 0
    while i <= strlenMinus3 do
        local x1, x2, x3, x4 = string_byte(str, i, i+3)
        x1 = _6bit_to_byte[x1]
        x2 = _6bit_to_byte[x2]
        x3 = _6bit_to_byte[x3]
        x4 = _6bit_to_byte[x4]
        if not (x1 and x2 and x3 and x4) then
            return nil
        end
        i = i + 4
        local cache = x1+x2*64+x3*4096+x4*262144
        local b1 = cache % 256
        cache = (cache - b1) / 256
        local b2 = cache % 256
        local b3 = (cache - b2) / 256
        buffer_size = buffer_size + 1
        buffer[buffer_size] =
            _byte_to_char[b1].._byte_to_char[b2].._byte_to_char[b3]
    end

    local cache  = 0
    local cache_bitlen = 0
    while i <= strlen do
        local x = string_byte(str, i, i)
        x =  _6bit_to_byte[x]
        if not x then
            return nil
        end
        cache = cache + x * _pow2[cache_bitlen]
        cache_bitlen = cache_bitlen + 6
        i = i + 1
    end

    while cache_bitlen >= 8 do
        local byte = cache % 256
        buffer_size = buffer_size + 1
        buffer[buffer_size] = _byte_to_char[byte]
        cache = (cache - byte) / 256
        cache_bitlen = cache_bitlen - 8
    end

    return table_concat(buffer)
end

local function InternalClearCache()
    _addon_channel_codec = nil
end

-- For test. Don't use the functions in this table for real application.
-- Stuffs in this table is subject to change.
LibDeflate.internals = {
    LoadStringToTable = LoadStringToTable,
    IsValidDictionary = IsValidDictionary,
    IsEqualAdler32 = IsEqualAdler32,
    _byte_to_6bit_char = _byte_to_6bit_char,
    _6bit_to_byte = _6bit_to_byte,
    InternalClearCache = InternalClearCache,
}

return Compression