Serpent

Lua serializer and pretty printer.

Features

• Human readable:
  • Provides single-line and multi-line output.
  • Nested tables are properly indented in the multi-line output.
  • Numerical keys are listed first.
  • Keys are (optionally) sorted alphanumerically.
  • Array part skips keys ({'a', 'b'} instead of {[1] = 'a', [2] = 'b'}).
  • nil values are included when expected ({1, nil, 3} instead of {1, [3]=3}).
  • Keys use short notation ({foo = 'foo'} instead of {['foo'] = 'foo'}).
  • Shared references and self-references are marked in the output.
• Machine readable: provides reliable deserialization using loadstring().
• Supports deeply nested tables.
• Supports tables with self-references.
• Keeps shared tables and functions shared after de/serialization.
• Supports function serialization using string.dump().
• Supports serialization of global functions.
• Supports __tostring and __serialize metamethods.
• Escapes new-line \010 and end-of-file control \026 characters in strings.
• Configurable with options and custom formatters.

Usage

local serpent = require("serpent")
local a = {1, nil, 3, x=1, ['true'] = 2, [not true]=3}
a[a] = a -- self-reference with a table as key and value

print(serpent.dump(a)) -- full serialization
print(serpent.line(a)) -- single line, no self-ref section
print(serpent.block(a)) -- multi-line indented, no self-ref section

local fun, err = loadstring(serpent.dump(a))
if err then error(err) end
local copy = fun()

-- or using serpent.load:
local ok, copy = serpent.load(serpent.dump(a))
print(ok and copy[3] == a[3])

Functions

Serpent provides three functions that are shortcuts to the same internal function, but set different options by default:

• dump(a[, {...}]) -- full serialization; sets name, compact and sparse options;
• line(a[, {...}]) -- single line pretty printing, no self-ref section; sets sortkeys and comment options;
• block(a[, {...}]) -- multi-line indented pretty printing, no self-ref section; sets indent, sortkeys, and comment options.

Note that line and block functions return pretty-printed data structures and if you want to deserialize them, you need to add return before running them through loadstring. For example: loadstring('return '..require('mobdebug').line("foo"))() == "foo".

While you can use loadstring or load functions to load serialized fragments, Serpent also provides load function that adds safety checks and reports an error if there is any executable code in the fragment.

• ok, res = serpent.load(str[, {safe = true}]) -- loads serialized fragment; you need to pass {safe = false} as the second value if you want to turn safety checks off.

Similar to pcall and loadstring calls, load returns status as the first value and the result or the error message as the second value.

Options

• indent (string) -- indentation; triggers long multi-line output.
• comment (true/false/maxlevel) -- provide stringified value in a comment (up to maxlevel of depth).
• sortkeys (true/false/function) -- sort keys.
• sparse (true/false) -- force sparse encoding (no nil filling based on #t).
• compact (true/false) -- remove spaces.
• fatal (true/False) -- raise fatal error on non-serilizable values.
• fixradix (true/False) -- change radix character set depending on locale to decimal dot.
• nocode (true/False) -- disable bytecode serialization for easy comparison.
• nohuge (true/False) -- disable checking numbers against undefined and huge values.
• maxlevel (number) -- specify max level up to which to expand nested tables.
• maxnum (number) -- specify max number of elements in a table.
• maxlength (number) -- specify max length for all table elements.
• metatostring (True/false) -- use __tostring metamethod when serializing tables (v0.29); set to false to disable and serialize the table as is, even when __tostring is present.
• numformat (string; "%.17g") -- specify format for numeric values as shortest possible round-trippable double (v0.30). Use "%.16g" for better readability and "%.17g" (the default value) to preserve floating point precision.
• valignore (table) -- allows to specify a list of values to ignore (as keys).
• keyallow (table) -- allows to specify the list of keys to be serialized. Any keys not in this list are not included in final output (as keys).
• keyignore (table) -- allows to specity the list of keys to ignore in serialization.
• valtypeignore (table) -- allows to specify a list of value types to ignore (as keys).
• custom (function) -- provide custom output for tables.
• name (string) -- name; triggers full serialization with self-ref section.

These options can be provided as a second parameter to Serpent functions.

block(a, {fatal = true})
line(a, {nocode = true, valignore = {[arrayToIgnore] = true}})
function todiff(a) return dump(a, {nocode = true, indent = ' '}) end

Serpent functions set these options to different default values:

• dump sets compact and sparse to true;
• line sets sortkeys and comment to true;
• block sets sortkeys and comment to true and indent to ' '.

Metatables with __tostring and __serialize methods

If a table or a userdata value has __tostring or __serialize method, the method will be used to serialize the value. If __serialize method is present, it will be called with the value as a parameter. if __serialize method is not present, but __tostring is, then tostring will be called with the value as a parameter. In both cases, the result will be serialized, so __serialize method can return a table, that will be serialize and replace the original value.

Sorting

A custom sort function can be provided to sort the contents of tables. The function takes 2 parameters, the first being the table (a list) with the keys, the second the original table. It should modify the first table in-place, and return nothing. For example, the following call will apply a sort function identical to the standard sort, except that it will not distinguish between lower- and uppercase.

local mysort  = function(k, o) -- k=keys, o=original table
  local maxn, to = 12, {number = 'a', string = 'b'}
  local function padnum(d) return ("%0"..maxn.."d"):format(d) end
  local sort = function(a,b)
    -- this -vvvvvvvvvv- is needed to sort array keys first
    return ((k[a] and 0 or to[type(a)] or 'z')..(tostring(a):gsub("%d+",padnum))):upper()
         < ((k[b] and 0 or to[type(b)] or 'z')..(tostring(b):gsub("%d+",padnum))):upper()
  end
  table.sort(k, sort)
end

local content = { some = 1, input = 2, To = 3, serialize = 4 }
local result = require('serpent').block(content, {sortkeys = mysort})

Formatters

Serpent supports a way to provide a custom formatter that allows to fully customize the output. The formatter takes four values:

• tag -- the name of the current element with '=' or an empty string in case of array index,
• head -- an opening table bracket { and associated indentation and newline (if any),
• body -- table elements concatenated into a string using commas and indentation/newlines (if any),
• tail -- a closing table bracket } and associated indentation and newline (if any), and
• level -- the current level.

For example, the following call will apply `Foo{bar} notation to its output (used by Metalua to display ASTs):

print((require "serpent").block(ast, {comment = false, custom =
  function(tag,head,body,tail)
    local out = head..body..tail
    if tag:find('^lineinfo') then
      out = out:gsub("\n%s+", "") -- collapse lineinfo to one line
    elseif tag == '' then
      body = body:gsub('%s*lineinfo = [^\n]+', '')
      local _,_,atag = body:find('tag = "(%w+)"%s*$')
      if atag then
        out = "`"..atag..head.. body:gsub('%s*tag = "%w+"%s*$', '')..tail
        out = out:gsub("\n%s+", ""):gsub(",}","}")
      else out = head..body..tail end
    end
    return tag..out
  end}))

Limitations

• Doesn't handle userdata (except filehandles in io.* table).
• Threads, function upvalues/environments, and metatables are not serialized.

Performance

A simple performance test against serialize.lua from metalua, pretty.write from Penlight, and tserialize.lua from lua-nucleo is included in t/bench.lua.

These are the results from one of the runs:

• nucleo (1000): 0.256s
• metalua (1000): 0.177s
• serpent (1000): 0.22s
• serpent (1000): 0.161s -- no comments, no string escapes, no math.huge check
• penlight (1000): 0.132s

Serpent does additional processing to escape \010 and \026 characters in strings (to address http://lua-users.org/lists/lua-l/2007-07/msg00362.html, which is already fixed in Lua 5.2) and to check all numbers for math.huge. The seconds number excludes this processing to put it on an equal footing with other modules that skip these checks (nucleo still checks for math.huge).

Author

Paul Kulchenko (paul@kulchenko.com)

License

See LICENSE file.

History

v0.30 (Sep 01 2017)

• Updated pairs to avoid using __pairs in Lua 5.2+.
• Added metatostring option to disable __tostring processing during serialization.
• Added level to the parameters of the custom function (closes #25).
• Added maxlength option to limit the space taken by table elements.
• Optimized serialization of functions when nocode option is specified.
• Added protection from __serialize check failing (pkulchenko/ZeroBraneStudio#732).
• Added keyignore option for the serializer.
• Added check for environments that may not have 'default' tables as tables.
• Added numeric format to preserve floating point precision (closes #17).
• Added using debug.metatable when available.
• Improved handling of failures in __tostring (pkulchenko/ZeroBraneStudio#446).

v0.28 (May 06 2015)

• Switched to a method proposed by @SoniEx2 to disallow function