Table Serialization |
|
Design decisions include
loadstring
(if so what style of table construction?), binary I/O, XML, JSON, YAML, HTML, protocol buffers, etc.?
table.foreach(t, print)
).
loadstring
, then a custom environment may be used, although this doesn't guard against infinite loops.)
Because of these different needs, there have been many implementations.
table.tostring
and table.print
[1] functions in its standard library (which can be used outside of Metalua). These focus on customizable pretty-printing, but do not preserve cycles or serialize functions. (See examples below.)
load
/loadstring
. Similar in purpose to JSON.parse()
vs. JavaScript eval()
.
This example doesn't focus on the readability of the serialized table. Instead, and contrary to other examples in this page, it conserves shared sub-structures. Here's an example of a test that would be passed by metalua serialization, but not by pretty-printers:
> x={ 1 } > x[2] = x > x[x] = 3 > x[3]={ 'indirect recursion', [x]=x } > y = { x, x } > x.y = y > assert (y[1] == y[2]) > s = serialize (x) > z = loadstring (s)() > assert (z.y[1] == z.y[2]) > =s local _={ } _[1]={ "indirect recursion" } _[2]={ false, false } _[3]={ 1, false, _[1], ["y"] = _[2] } _[3][2] = _[3] _[1][_[3]] = _[3] _[3][_[3]] = 3 _[2][1] = _[3] _[2][2] = _[3] return _[3] >
Sources for this serializer are available on the MetaLua repository: [2].
These functions are intended for pretty-printing rather than serialization: they don't preserve identity. They will terminate, though: if a table references itself, the inner occurrence will be printed as "[table: 0x12345678]" in order to avoid infinite recursion.
require "table2" require "string2" local u = {9} local t = {2, "3\"4", {5, 6}, x=function() end, [u]=u} table.print(t) --> { [{ 9 }] = { 9 }, x = function: 0x6a2870, 2, "3\"4", { 5, 6 } } table.print(t, 'nohash') --> { 2, "3\"4", { 5, 6 } } table.print(t, 'nohash', 10) --> { 2, -- "3\"4", -- { 5, 6 } } -- The `tag' field is particularly important in metalua, to represent tree-like structures. -- As such, it has got a special syntax, introduced by a back-quote "`", -- which is rendered by default by metalua's pretty printers. local t = {tag='Sum', 1, {tag='Product', 2, 3}, lines={10,11}} table.print(t) --> `Sum{ lines = { 10, 11 }, 1, `Product{ 2, 3 } } -- metalua tag syntax table.print(t, 'nohash') --> `Sum{ 1, `Product{ 2, 3 } } table.print(t, 'nohash', 10) -- metalua tag syntax --> `Sum{ 1, -- `Product{ 2, -- 3 } } -- tags syntax can be disabled: table.print(t, 'nohash', 'notag') --> { tag = "Sum", 1, { tag = "Product", 2, 3 } } -- tag syntax disabled -- Note: table.print(t, ...) is equivalent to print(table.tostring(t, ...)).
ISSUE: this should return a string rather than assume how the user wants to output the text.
-- Print anything - including nested tables function table_print (tt, indent, done) done = done or {} indent = indent or 0 if type(tt) == "table" then for key, value in pairs (tt) do io.write(string.rep (" ", indent)) -- indent it if type (value) == "table" and not done [value] then done [value] = true io.write(string.format("[%s] => table\n", tostring (key))); io.write(string.rep (" ", indent+4)) -- indent it io.write("(\n"); table_print (value, indent + 7, done) io.write(string.rep (" ", indent+4)) -- indent it io.write(")\n"); else io.write(string.format("[%s] => %s\n", tostring (key), tostring(value))) end end else io.write(tt .. "\n") end end
Universal tostring
function table_print (tt, indent, done) done = done or {} indent = indent or 0 if type(tt) == "table" then local sb = {} for key, value in pairs (tt) do table.insert(sb, string.rep (" ", indent)) -- indent it if type (value) == "table" and not done [value] then done [value] = true table.insert(sb, key .. " = {\n"); table.insert(sb, table_print (value, indent + 2, done)) table.insert(sb, string.rep (" ", indent)) -- indent it table.insert(sb, "}\n"); elseif "number" == type(key) then table.insert(sb, string.format("\"%s\"\n", tostring(value))) else table.insert(sb, string.format( "%s = \"%s\"\n", tostring (key), tostring(value))) end end return table.concat(sb) else return tt .. "\n" end end function to_string( tbl ) if "nil" == type( tbl ) then return tostring(nil) elseif "table" == type( tbl ) then return table_print(tbl) elseif "string" == type( tbl ) then return tbl else return tostring(tbl) end end
Example
print(to_string{ "Lua",user="Mariacher", {{co=coroutine.create(function() end),{number=12345.6789}}, func=function() end}, boolt=true} )
This prints
"Lua" { { { number = "12345.6789" } co = "thread: 0212B848" } func = "function: 01FC7C70" } boolt = "true" user = "Mariacher"
(the above code was originally from TableUtils)
Based on [PHP print_r]. Based on code by Nick Gammon, hacked by DracoBlue to fit to [PHP print_r]-Style.
Example: print_r{ 5,3,{5,3} } -->
[1] => 5 [2] => 3 [3] => Table { [1] => 5 [2] => 3 }
Compatibility: Lua 5.0 and 5.1
function print_r (t, indent, done) done = done or {} indent = indent or '' local nextIndent -- Storage for next indentation value for key, value in pairs (t) do if type (value) == "table" and not done [value] then nextIndent = nextIndent or (indent .. string.rep(' ',string.len(tostring (key))+2)) -- Shortcut conditional allocation done [value] = true print (indent .. "[" .. tostring (key) .. "] => Table {"); print (nextIndent .. "{"); print_r (value, nextIndent .. string.rep(' ',2), done) print (nextIndent .. "}"); else print (indent .. "[" .. tostring (key) .. "] => " .. tostring (value).."") end end end function print_r (t, indent) -- alt version, abuse to http://richard.warburton.it local indent=indent or '' for key,value in pairs(t) do io.write(indent,'[',tostring(key),']') if type(value)=="table" then io.write(':\n') print_r(value,indent..'\t') else io.write(' = ',tostring(value),'\n') end end end -- alt version2, handles cycles, functions, booleans, etc -- - abuse to http://richard.warburton.it -- output almost identical to print(table.show(t)) below. function print_r (t, name, indent) local tableList = {} function table_r (t, name, indent, full) local serial=string.len(full) == 0 and name or type(name)~="number" and '["'..tostring(name)..'"]' or '['..name..']' io.write(indent,serial,' = ') if type(t) == "table" then if tableList[t] ~= nil then io.write('{}; -- ',tableList[t],' (self reference)\n') else tableList[t]=full..serial if next(t) then -- Table not empty io.write('{\n') for key,value in pairs(t) do table_r(value,key,indent..'\t',full..serial) end io.write(indent,'};\n') else io.write('{};\n') end end else io.write(type(t)~="number" and type(t)~="boolean" and '"'..tostring(t)..'"' or tostring(t),';\n') end end table_r(t,name or '__unnamed__',indent or '','') end
Here is a more complete version of print_r:
Sorry for the length!
--[[ Author: Julio Manuel Fernandez-Diaz Date: January 12, 2007 (For Lua 5.1) Modified slightly by RiciLake to avoid the unnecessary table traversal in tablecount() Formats tables with cycles recursively to any depth. The output is returned as a string. References to other tables are shown as values. Self references are indicated. The string returned is "Lua code", which can be procesed (in the case in which indent is composed by spaces or "--"). Userdata and function keys and values are shown as strings, which logically are exactly not equivalent to the original code. This routine can serve for pretty formating tables with proper indentations, apart from printing them: print(table.show(t, "t")) -- a typical use Heavily based on "Saving tables with cycles", PIL2, p. 113. Arguments: t is the table. name is the name of the table (optional) indent is a first indentation (optional). --]] function table.show(t, name, indent) local cart -- a container local autoref -- for self references --[[ counts the number of elements in a table local function tablecount(t) local n = 0 for _, _ in pairs(t) do n = n+1 end return n end ]] -- (RiciLake) returns true if the table is empty local function isemptytable(t) return next(t) == nil end local function basicSerialize (o) local so = tostring(o) if type(o) == "function" then local info = debug.getinfo(o, "S") -- info.name is nil because o is not a calling level if info.what == "C" then return string.format("%q", so .. ", C function") else -- the information is defined through lines return string.format("%q", so .. ", defined in (" .. info.linedefined .. "-" .. info.lastlinedefined .. ")" .. info.source) end elseif type(o) == "number" or type(o) == "boolean" then return so else return string.format("%q", so) end end local function addtocart (value, name, indent, saved, field) indent = indent or "" saved = saved or {} field = field or name cart = cart .. indent .. field if type(value) ~= "table" then cart = cart .. " = " .. basicSerialize(value) .. ";\n" else if saved[value] then cart = cart .. " = {}; -- " .. saved[value] .. " (self reference)\n" autoref = autoref .. name .. " = " .. saved[value] .. ";\n" else saved[value] = name --if tablecount(value) == 0 then if isemptytable(value) then cart = cart .. " = {};\n" else cart = cart .. " = {\n" for k, v in pairs(value) do k = basicSerialize(k) local fname = string.format("%s[%s]", name, k) field = string.format("[%s]", k) -- three spaces between levels addtocart(v, fname, indent .. " ", saved, field) end cart = cart .. indent .. "};\n" end end end end name = name or "__unnamed__" if type(t) ~= "table" then return name .. " = " .. basicSerialize(t) end cart, autoref = "", "" addtocart(t, name, indent) return cart .. autoref end
A test:
----------------------------------------------------------- --- testing table.show t = {1, {2, 3, 4}, default = {"a", "b", d = {12, "w"}, e = 14}} t.g = t.default print("-----------------------------------") print(table.show(t)) -- shows __unnamed__ table tt = {1, h = {["p-q"] = "a", b = "e", c = {color = 3, name = "abc"}}, 2} f = table.show tt[f] = "OK" print("-----------------------------------") print(table.show(tt, "tt", "--oo-- ")) -- shows some initial 'indent' t.m = {} t.g.a = {} t.g.a.c = t t.tt = tt.new t.show = table.show print("-----------------------------------") print(table.show(t, "t")) -- most typical use print("-----------------------------------") print(table.show(math.tan, "tan")) -- not a table is OK print("-----------------------------------") s = "a string" print(table.show(s, "s")) -- not a table is OK
The output:
----------------------------------- __unnamed__ = { [1] = 1; [2] = { [1] = 2; [2] = 3; [3] = 4; }; ["default"] = { [1] = "a"; [2] = "b"; ["e"] = 14; ["d"] = { [1] = 12; [2] = "w"; }; }; ["g"] = {}; -- __unnamed__["default"] (self reference) }; __unnamed__["g"] = __unnamed__["default"]; ----------------------------------- --oo-- tt = { --oo-- [1] = 1; --oo-- [2] = 2; --oo-- ["function: 0x8070e20, defined in (28-99)@newprint_r.lua"] = "OK"; --oo-- ["h"] = { --oo-- ["b"] = "e"; --oo-- ["c"] = { --oo-- ["color"] = 3; --oo-- ["name"] = "abc"; --oo-- }; --oo-- ["p-q"] = "a"; --oo-- }; --oo-- }; ----------------------------------- t = { [1] = 1; [2] = { [1] = 2; [2] = 3; [3] = 4; }; ["m"] = {}; ["show"] = "function: 0x8070e20, defined in (28-99)@newprint_r.lua"; ["g"] = { [1] = "a"; [2] = "b"; ["e"] = 14; ["d"] = { [1] = 12; [2] = "w"; }; ["a"] = { ["c"] = {}; -- t (self reference) }; }; ["default"] = {}; -- t["g"] (self reference) }; t["g"]["a"]["c"] = t; t["default"] = t["g"]; ----------------------------------- tan = "function: 0x806f758, C function" ----------------------------------- s = "a string"
(the above code originally existed in MakingLuaLikePhp)
x = {1, 2, 3} x[x]=x print(table.show(x)) --[[output: __unnamed__ = { [1] = 1; [2] = 2; [3] = 3; ["table: 0x695f08"] = {}; -- __unnamed__ (self reference) }; __unnamed__["table: 0x695f08"] = __unnamed__; --]]
local val_to_str; do -- Cached function references (for performance). local byte = string.byte local find = string.find local match = string.match local gsub = string.gsub local format = string.format local insert = table.insert local sort = table.sort local concat = table.concat -- For escaping string values. local str_escape_map = { ['\a'] = '\\a', ['\b'] = '\\b', ['\t'] = '\\t', ['\n'] = '\\n', ['\v'] = '\\v', ['\f'] = '\\f', ['\r'] = '\\r', ['\\'] = '\\\\' } local str_escape_replace = function(c) return str_escape_map[c] or format('\\%03d', byte(c)) end -- Keys are comparable only if the same type, otherwise just sort them by type. local types_order, ref_types_order = { ['number'] = 0, ['boolean'] = 1, ['string'] = 2, ['table'] = 3, ['function'] = 4 }, 5 end local function compare_keys(k1, k2) local t1, t2 = type(k1), type(k2) if t1 ~= t2 then -- not the same type return (types_order[t1] or ref_types_order) < (types_order[t2] or ref_types_order) elseif t1 == 'boolean' then -- comparing booleans return not k1 -- Sort false before true. elseif t1 == 'number' or t1 == 'string' then -- comparing numbers (including NaNs or infinites) or strings return k1 < k2 -- Keys with the same comparable type. else -- comparing references (including tables, functions, userdata, threads...) return tostring(k1) < tostring(k2) -- may be the Lua engine adds some comparable info end end -- String keys matching valid identifiers that are reserved by Lua. local reserved_keys = { ['and'] = 1, ['break'] = 1, ['do'] = 1, ['else'] = 1, ['elseif'] = 1, ['end'] = 1, ['false'] = 1, ['for'] = 1, ['function'] = 1, ['if'] = 1, ['in'] = 1, ['local'] = 1, ['nil'] = 1, ['not'] = 1, ['or'] = 1, ['repeat'] = 1, ['return'] = 1, ['then'] = 1, ['true'] = 1, ['until'] = 1, ['while'] = 1 } -- Main function. val_to_str = function(val, options) -- Decode and cache the options. local include_mt = options and options.include_mt local prettyprint = options and options.prettyprint local asciionly = options and options.asciionly -- Precompute the output formats depending on options. local open = prettyprint and '{ ' or '{' local equals = prettyprint and ' = ' or '=' local comma = prettyprint and ', ' or ',' local close = prettyprint and ' }' or '}' -- What to escape: C0 controls, the backslash, and optionally non-ASCII bytes. local str_escape_pattern = asciionly and '[%z\001-\031\\\127-\255]' or '[%z\001-\031\\\127]' -- Indexed references (mapped to ids), and counters per ref type. local ref_ids, ref_counts = {}, {} -- Helper needed to detect recursive tables and avoid infinite loops. local function visit(ref) local typ = type(ref) if typ == 'number' or typ == 'boolean' then return tostring(ref) elseif typ == 'string' then if find(ref, "'") then str_escape_map['"'] = '\\"' return '"' .. gsub(ref, str_escape_pattern, str_escape_replace) .. '"' else str_escape_map['"'] = '"' return "'" .. gsub(ref, str_escape_pattern, str_escape_replace) .. "'" end elseif typ == 'table' then local id = ref_ids[ref] if id then return ':' .. typ .. '#' .. id .. ':' end id = (ref_counts[typ] or 0) + 1; ref_ids[ref], ref_counts[typ] = id, id -- First dump keys that are in sequence. local result, sequenced, keys = {}, {}, {} for i, val in ipairs(ref) do insert(result, visit(val)) sequenced[i] = true end -- Then dump other keys out of sequence, in a stable order. for key, _ in pairs(ref) do if not sequenced[key] then insert(keys, key) end end sequenced = nil -- Free the temp table no longer needed. -- Sorting keys (of any type) is needed for stable comparison of results. sort(keys, compare_keys) for _, key in ipairs(keys) do insert(result, (type(key) == 'string' and not reserved_keys[key] and match(key, '^[%a_][%d%a_]*$') and key or '[' .. visit(key) .. ']') .. equals .. visit(ref[key])) end keys = nil -- Free the temp table no longer needed. -- Finally dump the metatable (with pseudo-key '[]'), if there's one. if include_mt then ref = getmetatable(ref) if ref then insert(result, '[]' .. equals .. visit(ref)) end end -- Pack the result string. -- TODO: improve pretty-printing with newlines/indentation return open .. concat(result, comma) .. close elseif typ ~= 'nil' then -- other reference types (function, userdata, etc.) local id = ref_ids[ref] if not id then id = (ref_counts[typ] or 0) + 1; ref_ids[ref], ref_counts[ref] = id, id end return ':' .. typ .. '#' .. id .. ':' else return 'nil' end end return visit(val) end end
Example usage:
local s = val_to_str(anyvalue) local s = val_to_str(anyvalue, {include_mt=1, prettyprint=1}) -- you can specify optional flags
Notes:
{ }
pairs, around the equals sign =
in key/value pairs, and after the comma ,
in table constructors, and no newlines at all.
[] =
and cyclic reference is permitted (this works with a table attaching itself as its metatable).
:function#1:
, where the ordinal is the occurence number where an object of that type occurs in the return string. There's no reliable way to identify them, unless the "debug.info()" function is available and provides some tracing identifier (it is not in secured installation of Lua, like Scribunto where the only available debugging feature is a modified "info.traceback()"), you could dump the function environment if getfenv() is available (it is not in Scribunto).