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, Value}]} construct.

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.

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 omits inner type tags (still inserting the top-level tag), leading to some compression of the output. This also means that the serialized term cannot be decoded without the same schema information on the decoder side.

In the case of a directly provided template, all type information is inserted, such that the serialized term can be decoded without any added type information. The template types are still enforced during encoding.

ET = fun(Type,Term) -> io:fwrite("~w~n", [gmser_dyn:encode_typed(Type,Term)]) end.

ET([{int,int}], [{1,2}]) -> [<<0>>,<<1>>,[<<251>>,[[[<<248>>,<<1>>],[<<248>>,<<2>>]]]]]

gmser_dyn:register_type(1000,lt2i,[{int,int}]).
ET(lt2i, [{1,2}]) -> [<<0>>,<<1>>,[<<3,232>>,[[<<1>>,<<2>>]]]]

Alternative types

The dynamic encoder supports two additions to the gmserialization template language: any and #{alt => [AltTypes]}.

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

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>>]]]

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}.)