gmserialization/README.md

GM Serialization

Serialization helpers for the Gajumaru.

For an overview of the static serializer, see this document.

Build

$ rebar3 compile

Test

$ rebar3 eunit

Dynamic encoding

The module gmser_dyn offers dynamic encoding support, encoding most 'regular' Erlang data types into an internal RLP representation.

Main API:

• encode(term()) -> iolist()

• encode_typed(template(), term()) -> iolist()

• decode(iolist()) -> term()

• serialize(term()) -> binary()

• serialize_typed(template(), term()) -> binary()

• deserialize(binary()) -> term()

In the examples below, we use the decode functions, to illustrate how the type information is represented. The fully serialized form is produced by the serialize functions.

The basic types supported by the encoder are:

• integer() (anyint, code: 246)
• neg_integer() (negint, code: 247)
• non_neg_integer() (int , code: 248)
• binary() (binary, code: 249)
• boolean() (bool , code: 250)
• list() (list , code: 251)
• map() (map , code: 252)
• tuple() (tuple , code: 253)
• gmser_id:id() (id , code: 254)
• atom() (label , code: 255)

(The range of codes is chosen because the gmser_chain_objects codes range from 10 to 200, and also to stay within 1 byte.)

When encoding map types, the map elements are first sorted.

When specifying a map type for template-driven encoding, use the #{items => [{Key, ValueType} | {opt, Key, ValueType}]} construct. The key names are included in the encoding, and are match against the item specs during decoding. If the key names don't match, the decoding fails, unless for an {opt, K, V} item, in which case that item spec is skipped.

T = #{items => [{a,int},{opt,b,int},{c,int}]}
E1 = gmser_dyn:encode_typed(T, #{a => 1, b => 2, c => 3}) ->
    [<<0>>,<<1>>,[<<252>>,
     [[[<<255>>,<<97>>],[<<248>>,<<1>>]],
     [[<<255>>,<<98>>],[<<248>>,<<2>>]],
     [[<<255>>,<<99>>],[<<248>>,<<3>>]]]]]
E2 = gmser_dyn:encode_typed(T, #{a => 1, c => 3}) ->
    [<<0>>,<<1>>,[<<252>>,
     [[[<<255>>,<<97>>],[<<248>>,<<1>>]],
     [[<<255>>,<<99>>],[<<248>>,<<3>>]]]]]
gmser_dyn:decode_typed(T,E2) ->
    #{c => 3,a => 1}

Labels

Labels correspond to (existing) atoms in Erlang. Decoding of a label results in a call to binary_to_existing_atom/2, so will fail if the corresponding atom does not already exist.

This behavior can be modified using the option #{missing_labels => fail | create | convert}, where fail is the default, as described above, convert means that missing atoms are converted to binaries, and create means that the atom is created dynamically.

The option can be passed e.g.:

gmser_dyn:deserialize(Binary, gmser_dyn:set_opts(#{missing_labels => convert}))

or

gmser_dyn:deserialize(Binary, gmser_dyn:set_opts(#{missing_labels => convert}, Types))

By calling gmser_dyn:register_types/1, after having added options to the type map, the options can be made to take effect automatically.

It's possible to cache labels for more compact encoding. Note that when caching labels, the same cache mapping needs to be used on the decoder side.

Labels are encoded as [<<255>>, << AtomToBinary/binary >>]. If a cached label is used, the encoding becomes [<<255>, [Ix]], where Ix is the integer-encoded index value of the cached label.

Examples

Dynamically encoded objects have the basic structure [<<0>>,V,Obj], where V is the integer-coded version, and Obj is the top-level encoding on the form [Tag,Data].

E = fun(T) -> io:fwrite("~w~n", [gmser_dyn:encode(T)]) end.

E(17)        -> [<<0>>,<<1>>,[<<248>>,<<17>>]]
E(<<"abc">>) -> [<<0>>,<<1>>,[<<249>>,<<97,98,99>>]]
E(true)      -> [<<0>>,<<1>>,[<<250>>,<<1>>]]
E(false)     -> [<<0>>,<<1>>,[<<250>>,<<0>>]]
E([1,2])     -> [<<0>>,<<1>>,[<<251>>,[[<<248>>,<<1>>],[<<248>>,<<2>>]]]]
E({1,2})     -> [<<0>>,<<1>>,[<<253>>,[[<<248>>,<<1>>],[<<248>>,<<2>>]]]]
E(#{a=>1, b=>2}) ->
  [<<0>>,<<1>>,[<<252>>,[[[<<255>>,<<97>>],[<<248>>,<<1>>]],[[<<255>>,<<98>>],[<<248>>,<<2>>]]]]]
E(gmser_id:create(account,<<1:256>>)) ->
  [<<0>>,<<1>>,[<<254>>,<<1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1>>]]

Note that tuples and list are encoded the same way, except for the initial type tag. Maps are encoded as [<Map>, [KV1, KV2, ...]], where [KV1, KV2, ...] is the sorted list of key-value tuples from map:to_list(Map), but with the tuple type tag omitted.

Template-driven encoding

Templates can be provided to the encoder by either naming an already registered type, or by passing a template directly. In both cases, the encoder will enforce the type information in the template.

If the template has been registered, the encoder uses the registered type specification to drive the encoding. The code of the registered template is embedded in the encoded output:

gmser_dyn:encode_typed({int,int,int}, {1,2,3}) ->
    [<<0>>,<<1>>,[<<253>>,
     [[<<248>>,<<1>>],[<<248>>,<<2>>],[<<248>>,<<3>>]]]]

Types = gmser_dyn_types:add_type(t3,1013,{int,int,int}).
gmser_dyn:encode_typed(t3, {1,2,3}, Types) ->
    [<<0>>,<<1>>,[[<<3,245>>,<<253>>],
     [[<<248>>,<<1>>],[<<248>>,<<2>>],[<<248>>,<<3>>]]]]

Note that the original <<253>> type code is wrapped as [<<3,245>>,<<253>>], where <<3,245>> corresponds to the custom code 1013.

Using the default option #{strict => true}, the decoder will extract the custom type spec, and validate the encoded data against it. If the custom code is missing, the decoder aborts. Using #{strict => false}, the custom code is used if it exists, but otherwise, it's ignored, and the encoded data is decoded using the dynamic type info.

Alternative types

The dynamic encoder supports a few additions to the gmserialization template language: any, #{list => Type}, #{alt => [AltTypes]} and #{switch => [AltTypes]}.

any

The any type doesn't have an associated code, but enforces dynamic encoding.

list

The original list type notation expects a key-value list, e.g.

[{name, binary}, {age, int}]

EL = gmser_dyn:encode_typed([{name,binary},{age,int}], [{name,<<"Ulf">>},{age,29}]) ->
    [<<0>>,<<1>>,[<<251>>,
     [[<<253>>,[[<<255>>,<<110,97,109,101>>],[<<249>>,<<85,108,102>>]]],
      [<<253>>,[[<<255>>,<<97,103,101>>],[<<248>>,<<29>>]]]]]]

Note that the encoding explicitly lays out a [{Key, Value}] structure, all dynamically typed. This means it can be dynamically decoded without templates.

gmser_dyn:decode(EL).
[{name,<<"Ulf">>},{age,29}]

In order to specify something like Erlang's [integer()] type, we can use the following:

gmser_dyn:encode_typed(#{list => int}, [1,2,3,4]) ->
    [<<0>>,<<1>>,[<<251>>,
     [[<<248>>,<<1>>],[<<248>>,<<2>>],[<<248>>,<<3>>],[<<248>>,<<4>>]]]]

alt

The #{alt => [Type]} construct also enforces dynamic encoding, and will try to encode as each type in the list, in the specified order, until one matches.

gmser_dyn:encode_typed(#{alt => [negint,int]}, 5) -> [<<0>>,<<1>>,[<<247>>,<<5>>]]
gmser_dyn:encode_typed(#{alt => [negint,int]}, 5) -> [<<0>>,<<1>>,[<<248>>,<<5>>]]

gmser_dyn:encode_typed(anyint,-5) -> [<<0>>,<<1>>,[<<246>>,[<<247>>,<<5>>]]]
gmser_dyn:encode_typed(anyint,5)  -> [<<0>>,<<1>>,[<<246>>,[<<248>>,<<5>>]]]

switch

The switch type allows for encoding a 'tagged' object, where the tag determines the type.

E1 = gmser_dyn:encode_typed(#{switch => #{name => binary, age => int}}, #{age => 29}) ->
    [<<0>>,<<1>>,[<<252>>,[[[<<255>>,<<97,103,101>>],[<<248>>,<<29>>]]]]]
gmser_dyn:decode_typed(#{switch => #{name => binary, age => int}}, E1) ->
    #{age => 29}
E2 = gmser_dyn:encode_typed(#{switch => #{name => binary, age => int}}, #{name => <<"Ulf">>}) ->
    [<<0>>,<<1>>,[<<252>>,[[[<<255>>,<<110,97,109,101>>],[<<249>>,<<85,108,102>>]]]]]
gmser_dyn:decode_typed(#{switch => #{name => binary, age => int}}, E1) ->
    #{name => <<"Ulf">>}

A practical use of switch would be in a protocol schema:

t_msg(_) ->
    #{switch => #{ call         => t_call
                 , reply        => t_reply
                 , notification => t_notification }}.

t_call(_) ->
    #{items => [ {id, anyint}
               , {req, t_req} ]}.

t_reply(_) ->
    #{alt => [#{items => [ {id, anyint}
                         , {result, t_result} ]},
              #{items => [ {id, anyint}
                         , {code, anyint}
                         , {message, binary} ]}
             ]}.

In this scenario, messages are 'taggged' as 1-element maps, e.g.:

async_request(Msg) ->
    Id = erlang:unique_integer(),
    gmmp_cp:to_server(
      whereis(gmmp_core_connector),
      #{call => #{ id => Id
                 , req => Msg }}),
    Id.

Notes

Note that anyint is a standard type. The static serializer supports only positive integers (int), as negative numbers are forbidden on-chain. For dynamic encoding e.g. in messaging protocols, handling negative numbers can be useful, so the negint type was added as a dynamic type. To encode a full-range integer, the alt construct is needed.

(Floats are not supported, as they are non-deterministic. Rationals and fixed-point numbers could easily be handled as high-level types, e.g. as {int,int}.)