-- Simple JSON encoding and decoding in pure Lua. -- -- Copyright 2010-2016 Jeffrey Friedl -- http://regex.info/blog/ -- Latest version: http://regex.info/blog/lua/json -- -- This code is released under a Creative Commons CC-BY "Attribution" License: -- http://creativecommons.org/licenses/by/3.0/deed.en_US -- -- It can be used for any purpose so long as the copyright notice above, -- the web-page links above, and the 'AUTHOR_NOTE' string below are -- maintained. Enjoy. -- local VERSION = 20160728.17 -- version history at end of file local AUTHOR_NOTE = "-[ JSON.lua package by Jeffrey Friedl (http://regex.info/blog/lua/json) version 20160728.17 ]-" -- -- The 'AUTHOR_NOTE' variable exists so that information about the source -- of the package is maintained even in compiled versions. It's also -- included in OBJDEF below mostly to quiet warnings about unused variables. -- local OBJDEF = { VERSION = VERSION, AUTHOR_NOTE = AUTHOR_NOTE, } local default_pretty_indent = " " local default_pretty_options = { pretty = true, align_keys = false, indent = default_pretty_indent } local isArray = { __tostring = function() return "JSON array" end } isArray.__index = isArray local isObject = { __tostring = function() return "JSON object" end } isObject.__index = isObject function OBJDEF:newArray(tbl) return setmetatable(tbl or {}, isArray) end function OBJDEF:newObject(tbl) return setmetatable(tbl or {}, isObject) end local function getnum(op) return type(op) == 'number' and op or op.N end local isNumber = { __index = isNumber, __tostring = function(T) return T.S end, __unm = function(op) return getnum(op) end, __concat = function(op1, op2) return tostring(op1) .. tostring(op2) end, __add = function(op1, op2) return getnum(op1) + getnum(op2) end, __sub = function(op1, op2) return getnum(op1) - getnum(op2) end, __mul = function(op1, op2) return getnum(op1) * getnum(op2) end, __div = function(op1, op2) return getnum(op1) / getnum(op2) end, __mod = function(op1, op2) return getnum(op1) % getnum(op2) end, __pow = function(op1, op2) return getnum(op1) ^ getnum(op2) end, __lt = function(op1, op2) return getnum(op1) < getnum(op2) end, __eq = function(op1, op2) return getnum(op1) == getnum(op2) end, __le = function(op1, op2) return getnum(op1) <= getnum(op2) end, } function OBJDEF:asNumber(item) if getmetatable(item) == isNumber then -- it's already a JSON number object. return item elseif type(item) == 'table' and type(item.S) == 'string' and type(item.N) == 'number' then -- it's a number-object table that lost its metatable, so give it one return setmetatable(item, isNumber) else -- the normal situation... given a number or a string representation of a number.... local holder = { S = tostring(item), -- S is the representation of the number as a string, which remains precise N = tonumber(item), -- N is the number as a Lua number. } return setmetatable(holder, isNumber) end end local function unicode_codepoint_as_utf8(codepoint) -- -- codepoint is a number -- if codepoint <= 127 then return string.char(codepoint) elseif codepoint <= 2047 then -- -- 110yyyxx 10xxxxxx <-- useful notation from http://en.wikipedia.org/wiki/Utf8 -- local highpart = math.floor(codepoint / 0x40) local lowpart = codepoint - (0x40 * highpart) return string.char(0xC0 + highpart, 0x80 + lowpart) elseif codepoint <= 65535 then -- -- 1110yyyy 10yyyyxx 10xxxxxx -- local highpart = math.floor(codepoint / 0x1000) local remainder = codepoint - 0x1000 * highpart local midpart = math.floor(remainder / 0x40) local lowpart = remainder - 0x40 * midpart highpart = 0xE0 + highpart midpart = 0x80 + midpart lowpart = 0x80 + lowpart -- -- Check for an invalid character (thanks Andy R. at Adobe). -- See table 3.7, page 93, in http://www.unicode.org/versions/Unicode5.2.0/ch03.pdf#G28070 -- if ( highpart == 0xE0 and midpart < 0xA0 ) or ( highpart == 0xED and midpart > 0x9F ) or ( highpart == 0xF0 and midpart < 0x90 ) or ( highpart == 0xF4 and midpart > 0x8F ) then return "?" else return string.char(highpart, midpart, lowpart) end else -- -- 11110zzz 10zzyyyy 10yyyyxx 10xxxxxx -- local highpart = math.floor(codepoint / 0x40000) local remainder = codepoint - 0x40000 * highpart local midA = math.floor(remainder / 0x1000) remainder = remainder - 0x1000 * midA local midB = math.floor(remainder / 0x40) local lowpart = remainder - 0x40 * midB return string.char(0xF0 + highpart, 0x80 + midA, 0x80 + midB, 0x80 + lowpart) end end function OBJDEF:onDecodeError(message, text, location, etc) if text then if location then message = string.format("%s at char %d of: %s", message, location, text) else message = string.format("%s: %s", message, text) end end if etc ~= nil then message = message .. " (" .. OBJDEF:encode(etc) .. ")" end if self.assert then self.assert(false, message) else assert(false, message) end end OBJDEF.onDecodeOfNilError = OBJDEF.onDecodeError OBJDEF.onDecodeOfHTMLError = OBJDEF.onDecodeError function OBJDEF:onEncodeError(message, etc) if etc ~= nil then message = message .. " (" .. OBJDEF:encode(etc) .. ")" end if self.assert then self.assert(false, message) else assert(false, message) end end local function grok_number(self, text, start, options) -- -- Grab the integer part -- local integer_part = text:match('^-?[1-9]%d*', start) or text:match("^-?0", start) if not integer_part then self:onDecodeError("expected number", text, start, options.etc) end local i = start + integer_part:len() -- -- Grab an optional decimal part -- local decimal_part = text:match('^%.%d+', i) or "" i = i + decimal_part:len() -- -- Grab an optional exponential part -- local exponent_part = text:match('^[eE][-+]?%d+', i) or "" i = i + exponent_part:len() local full_number_text = integer_part .. decimal_part .. exponent_part if options.decodeNumbersAsObjects then return OBJDEF:asNumber(full_number_text), i end -- -- If we're told to stringify under certain conditions, so do. -- We punt a bit when there's an exponent by just stringifying no matter what. -- I suppose we should really look to see whether the exponent is actually big enough one -- way or the other to trip stringification, but I'll be lazy about it until someone asks. -- if (options.decodeIntegerStringificationLength and (integer_part:len() >= options.decodeIntegerStringificationLength or exponent_part:len() > 0)) or (options.decodeDecimalStringificationLength and (decimal_part:len() >= options.decodeDecimalStringificationLength or exponent_part:len() > 0)) then return full_number_text, i -- this returns the exact string representation seen in the original JSON end local as_number = tonumber(full_number_text) if not as_number then self:onDecodeError("bad number", text, start, options.etc) end return as_number, i end local function grok_string(self, text, start, options) if text:sub(start,start) ~= '"' then self:onDecodeError("expected string's opening quote", text, start, options.etc) end local i = start + 1 -- +1 to bypass the initial quote local text_len = text:len() local VALUE = "" while i <= text_len do local c = text:sub(i,i) if c == '"' then return VALUE, i + 1 end if c ~= '\\' then VALUE = VALUE .. c i = i + 1 elseif text:match('^\\b', i) then VALUE = VALUE .. "\b" i = i + 2 elseif text:match('^\\f', i) then VALUE = VALUE .. "\f" i = i + 2 elseif text:match('^\\n', i) then VALUE = VALUE .. "\n" i = i + 2 elseif text:match('^\\r', i) then VALUE = VALUE .. "\r" i = i + 2 elseif text:match('^\\t', i) then VALUE = VALUE .. "\t" i = i + 2 else local hex = text:match('^\\u([0123456789aAbBcCdDeEfF][0123456789aAbBcCdDeEfF][0123456789aAbBcCdDeEfF][0123456789aAbBcCdDeEfF])', i) if hex then i = i + 6 -- bypass what we just read -- We have a Unicode codepoint. It could be standalone, or if in the proper range and -- followed by another in a specific range, it'll be a two-code surrogate pair. local codepoint = tonumber(hex, 16) if codepoint >= 0xD800 and codepoint <= 0xDBFF then -- it's a hi surrogate... see whether we have a following low local lo_surrogate = text:match('^\\u([dD][cdefCDEF][0123456789aAbBcCdDeEfF][0123456789aAbBcCdDeEfF])', i) if lo_surrogate then i = i + 6 -- bypass the low surrogate we just read codepoint = 0x2400 + (codepoint - 0xD800) * 0x400 + tonumber(lo_surrogate, 16) else -- not a proper low, so we'll just leave the first codepoint as is and spit it out. end end VALUE = VALUE .. unicode_codepoint_as_utf8(codepoint) else -- just pass through what's escaped VALUE = VALUE .. text:match('^\\(.)', i) i = i + 2 end end end self:onDecodeError("unclosed string", text, start, options.etc) end local function skip_whitespace(text, start) local _, match_end = text:find("^[ \n\r\t]+", start) -- [http://www.ietf.org/rfc/rfc4627.txt] Section 2 if match_end then return match_end + 1 else return start end end local grok_one -- assigned later local function grok_object(self, text, start, options) if text:sub(start,start) ~= '{' then self:onDecodeError("expected '{'", text, start, options.etc) end local i = skip_whitespace(text, start + 1) -- +1 to skip the '{' local VALUE = self.strictTypes and self:newObject { } or { } if text:sub(i,i) == '}' then return VALUE, i + 1 end local text_len = text:len() while i <= text_len do local key, new_i = grok_string(self, text, i, options) i = skip_whitespace(text, new_i) if text:sub(i, i) ~= ':' then self:onDecodeError("expected colon", text, i, options.etc) end i = skip_whitespace(text, i + 1) local new_val, new_i = grok_one(self, text, i, options) VALUE[key] = new_val -- -- Expect now either '}' to end things, or a ',' to allow us to continue. -- i = skip_whitespace(text, new_i) local c = text:sub(i,i) if c == '}' then return VALUE, i + 1 end if text:sub(i, i) ~= ',' then self:onDecodeError("expected comma or '}'", text, i, options.etc) end i = skip_whitespace(text, i + 1) end self:onDecodeError("unclosed '{'", text, start, options.etc) end local function grok_array(self, text, start, options) if text:sub(start,start) ~= '[' then self:onDecodeError("expected '['", text, start, options.etc) end local i = skip_whitespace(text, start + 1) -- +1 to skip the '[' local VALUE = self.strictTypes and self:newArray { } or { } if text:sub(i,i) == ']' then return VALUE, i + 1 end local VALUE_INDEX = 1 local text_len = text:len() while i <= text_len do local val, new_i = grok_one(self, text, i, options) -- can't table.insert(VALUE, val) here because it's a no-op if val is nil VALUE[VALUE_INDEX] = val VALUE_INDEX = VALUE_INDEX + 1 i = skip_whitespace(text, new_i) -- -- Expect now either ']' to end things, or a ',' to allow us to continue. -- local c = text:sub(i,i) if c == ']' then return VALUE, i + 1 end if text:sub(i, i) ~= ',' then self:onDecodeError("expected comma or '['", text, i, options.etc) end i = skip_whitespace(text, i + 1) end self:onDecodeError("unclosed '['", text, start, options.etc) end grok_one = function(self, text, start, options) -- Skip any whitespace start = skip_whitespace(text, start) if start > text:len() then self:onDecodeError("unexpected end of string", text, nil, options.etc) end if text:find('^"', start) then return grok_string(self, text, start, options.etc) elseif text:find('^[-0123456789 ]', start) then return grok_number(self, text, start, options) elseif text:find('^%{', start) then return grok_object(self, text, start, options) elseif text:find('^%[', start) then return grok_array(self, text, start, options) elseif text:find('^true', start) then return true, start + 4 elseif text:find('^false', start) then return false, start + 5 elseif text:find('^null', start) then return nil, start + 4 else self:onDecodeError("can't parse JSON", text, start, options.etc) end end function OBJDEF:decode(text, etc, options) -- -- If the user didn't pass in a table of decode options, make an empty one. -- if type(options) ~= 'table' then options = {} end -- -- If they passed in an 'etc' argument, stuff it into the options. -- (If not, any 'etc' field in the options they passed in remains to be used) -- if etc ~= nil then options.etc = etc end if type(self) ~= 'table' or self.__index ~= OBJDEF then OBJDEF:onDecodeError("JSON:decode must be called in method format", nil, nil, options.etc) end if text == nil then self:onDecodeOfNilError(string.format("nil passed to JSON:decode()"), nil, nil, options.etc) elseif type(text) ~= 'string' then self:onDecodeError(string.format("expected string argument to JSON:decode(), got %s", type(text)), nil, nil, options.etc) end if text:match('^%s*$') then return nil end if text:match('^%s*<') then -- Can't be JSON... we'll assume it's HTML self:onDecodeOfHTMLError(string.format("html passed to JSON:decode()"), text, nil, options.etc) end -- -- Ensure that it's not UTF-32 or UTF-16. -- Those are perfectly valid encodings for JSON (as per RFC 4627 section 3), -- but this package can't handle them. -- if text:sub(1,1):byte() == 0 or (text:len() >= 2 and text:sub(2,2):byte() == 0) then self:onDecodeError("JSON package groks only UTF-8, sorry", text, nil, options.etc) end -- -- apply global options -- if options.decodeNumbersAsObjects == nil then options.decodeNumbersAsObjects = self.decodeNumbersAsObjects end if options.decodeIntegerStringificationLength == nil then options.decodeIntegerStringificationLength = self.decodeIntegerStringificationLength end if options.decodeDecimalStringificationLength == nil then options.decodeDecimalStringificationLength = self.decodeDecimalStringificationLength end local success, value = pcall(grok_one, self, text, 1, options) if success then return value else -- if JSON:onDecodeError() didn't abort out of the pcall, we'll have received the error message here as "value", so pass it along as an assert. if self.assert then self.assert(false, value) else assert(false, value) end -- and if we're still here, return a nil and throw the error message on as a second arg return nil, value end end local function backslash_replacement_function(c) if c == "\n" then return "\\n" elseif c == "\r" then return "\\r" elseif c == "\t" then return "\\t" elseif c == "\b" then return "\\b" elseif c == "\f" then return "\\f" elseif c == '"' then return '\\"' elseif c == '\\' then return '\\\\' else return string.format("\\u%04x", c:byte()) end end local chars_to_be_escaped_in_JSON_string = '[' .. '"' -- class sub-pattern to match a double quote .. '%\\' -- class sub-pattern to match a backslash .. '%z' -- class sub-pattern to match a null .. '\001' .. '-' .. '\031' -- class sub-pattern to match control characters .. ']' local LINE_SEPARATOR_as_utf8 = unicode_codepoint_as_utf8(0x2028) local PARAGRAPH_SEPARATOR_as_utf8 = unicode_codepoint_as_utf8(0x2029) local function json_string_literal(value, options) local newval = value:gsub(chars_to_be_escaped_in_JSON_string, backslash_replacement_function) if options.stringsAreUtf8 then -- -- This feels really ugly to just look into a string for the sequence of bytes that we know to be a particular utf8 character, -- but utf8 was designed purposefully to make this kind of thing possible. Still, feels dirty. -- I'd rather decode the byte stream into a character stream, but it's not technically needed so -- not technically worth it. -- newval = newval:gsub(LINE_SEPARATOR_as_utf8, '\\u2028'):gsub(PARAGRAPH_SEPARATOR_as_utf8,'\\u2029') end return '"' .. newval .. '"' end local function object_or_array(self, T, etc) -- -- We need to inspect all the keys... if there are any strings, we'll convert to a JSON -- object. If there are only numbers, it's a JSON array. -- -- If we'll be converting to a JSON object, we'll want to sort the keys so that the -- end result is deterministic. -- local string_keys = { } local number_keys = { } local number_keys_must_be_strings = false local maximum_number_key for key in pairs(T) do if type(key) == 'string' then table.insert(string_keys, key) elseif type(key) == 'number' then table.insert(number_keys, key) if key <= 0 or key >= math.huge then number_keys_must_be_strings = true elseif not maximum_number_key or key > maximum_number_key then maximum_number_key = key end else self:onEncodeError("can't encode table with a key of type " .. type(key), etc) end end if #string_keys == 0 and not number_keys_must_be_strings then -- -- An empty table, or a numeric-only array -- if #number_keys > 0 then return nil, maximum_number_key -- an array elseif tostring(T) == "JSON array" then return nil elseif tostring(T) == "JSON object" then return { } else -- have to guess, so we'll pick array, since empty arrays are likely more common than empty objects return nil end end table.sort(string_keys) local map if #number_keys > 0 then -- -- If we're here then we have either mixed string/number keys, or numbers inappropriate for a JSON array -- It's not ideal, but we'll turn the numbers into strings so that we can at least create a JSON object. -- if self.noKeyConversion then self:onEncodeError("a table with both numeric and string keys could be an object or array; aborting", etc) end -- -- Have to make a shallow copy of the source table so we can remap the numeric keys to be strings -- map = { } for key, val in pairs(T) do map[key] = val end table.sort(number_keys) -- -- Throw numeric keys in there as strings -- for _, number_key in ipairs(number_keys) do local string_key = tostring(number_key) if map[string_key] == nil then table.insert(string_keys , string_key) map[string_key] = T[number_key] else self:onEncodeError("conflict converting table with mixed-type keys into a JSON object: key " .. number_key .. " exists both as a string and a number.", etc) end end end return string_keys, nil, map end -- -- Encode -- -- 'options' is nil, or a table with possible keys: -- -- pretty -- If true, return a pretty-printed version. -- -- indent -- A string (usually of spaces) used to indent each nested level. -- -- align_keys -- If true, align all the keys when formatting a table. -- -- null -- If this exists with a string value, table elements with this value are output as JSON null. -- -- stringsAreUtf8 -- If true, consider Lua strings not as a sequence of bytes, but as a sequence of UTF-8 characters. -- (Currently, the only practical effect of setting this option is that Unicode LINE and PARAGRAPH -- separators, if found in a string, are encoded with a JSON escape instead of as raw UTF-8. -- The JSON is valid either way, but encoding this way, apparently, allows the resulting JSON -- to also be valid Java.) -- -- local encode_value -- must predeclare because it calls itself function encode_value(self, value, parents, etc, options, indent, for_key) -- -- keys in a JSON object can never be null, so we don't even consider options.null when converting a key value -- if value == nil or (not for_key and options and options.null and value == options.null) then return 'null' elseif type(value) == 'string' then return json_string_literal(value, options) elseif type(value) == 'number' then if value ~= value then -- -- NaN (Not a Number). -- JSON has no NaN, so we have to fudge the best we can. This should really be a package option. -- return "null" elseif value >= math.huge then -- -- Positive infinity. JSON has no INF, so we have to fudge the best we can. This should -- really be a package option. Note: at least with some implementations, positive infinity -- is both ">= math.huge" and "<= -math.huge", which makes no sense but that's how it is. -- Negative infinity is properly "<= -math.huge". So, we must be sure to check the ">=" -- case first. -- return "1e+9999" elseif value <= -math.huge then -- -- Negative infinity. -- JSON has no INF, so we have to fudge the best we can. This should really be a package option. -- return "-1e+9999" else return tostring(value) end elseif type(value) == 'boolean' then return tostring(value) elseif type(value) ~= 'table' then self:onEncodeError("can't convert " .. type(value) .. " to JSON", etc) elseif getmetatable(value) == isNumber then return tostring(value) else -- -- A table to be converted to either a JSON object or array. -- local T = value if type(options) ~= 'table' then options = {} end if type(indent) ~= 'string' then indent = "" end if parents[T] then self:onEncodeError("table " .. tostring(T) .. " is a child of itself", etc) else parents[T] = true end local result_value local object_keys, maximum_number_key, map = object_or_array(self, T, etc) if maximum_number_key then -- -- An array... -- local ITEMS = { } for i = 1, maximum_number_key do table.insert(ITEMS, encode_value(self, T[i], parents, etc, options, indent)) end if options.pretty then result_value = "[ " .. table.concat(ITEMS, ", ") .. " ]" else result_value = "[" .. table.concat(ITEMS, ",") .. "]" end elseif object_keys then -- -- An object -- local TT = map or T if options.pretty then local KEYS = { } local max_key_length = 0 for _, key in ipairs(object_keys) do local encoded = encode_value(self, tostring(key), parents, etc, options, indent, true) if options.align_keys then max_key_length = math.max(max_key_length, #encoded) end table.insert(KEYS, encoded) end local key_indent = indent .. tostring(options.indent or "") local subtable_indent = key_indent .. string.rep(" ", max_key_length) .. (options.align_keys and " " or "") local FORMAT = "%s%" .. string.format("%d", max_key_length) .. "s: %s" local COMBINED_PARTS = { } for i, key in ipairs(object_keys) do local encoded_val = encode_value(self, TT[key], parents, etc, options, subtable_indent) table.insert(COMBINED_PARTS, string.format(FORMAT, key_indent, KEYS[i], encoded_val)) end result_value = "{\n" .. table.concat(COMBINED_PARTS, ",\n") .. "\n" .. indent .. "}" else local PARTS = { } for _, key in ipairs(object_keys) do local encoded_val = encode_value(self, TT[key], parents, etc, options, indent) local encoded_key = encode_value(self, tostring(key), parents, etc, options, indent, true) table.insert(PARTS, string.format("%s:%s", encoded_key, encoded_val)) end result_value = "{" .. table.concat(PARTS, ",") .. "}" end else -- -- An empty array/object... we'll treat it as an array, though it should really be an option -- result_value = "[]" end parents[T] = false return result_value end end function OBJDEF:encode(value, etc, options) if type(self) ~= 'table' or self.__index ~= OBJDEF then OBJDEF:onEncodeError("JSON:encode must be called in method format", etc) end -- -- If the user didn't pass in a table of decode options, make an empty one. -- if type(options) ~= 'table' then options = {} end return encode_value(self, value, {}, etc, options) end function OBJDEF:encode_pretty(value, etc, options) if type(self) ~= 'table' or self.__index ~= OBJDEF then OBJDEF:onEncodeError("JSON:encode_pretty must be called in method format", etc) end -- -- If the user didn't pass in a table of decode options, use the default pretty ones -- if type(options) ~= 'table' then options = default_pretty_options end return encode_value(self, value, {}, etc, options) end function OBJDEF.__tostring() return "JSON encode/decode package" end OBJDEF.__index = OBJDEF function OBJDEF:new(args) local new = { } if args then for key, val in pairs(args) do new[key] = val end end return setmetatable(new, OBJDEF) end return OBJDEF:new()