Skip to content

Latest commit

 

History

History
165 lines (120 loc) · 4.69 KB

serialization.md

File metadata and controls

165 lines (120 loc) · 4.69 KB
Raw

Serialization

In order that ADL data structures serialized in one language can be deserialized in other, the mapping to serialized values must be consistent. Hence, ADL is a formal specification for serialized data.

Currently json serialization is implemented, as this gives best interoperability with browser based code. It is envisaged that in future binary serialization will also be supported. Note also that json is also the format used for literals in the ADL language itself.

Each ADL primitive and declaration is serialized as specified in the rules below. The serialization for any ADL type can hence be determined by the repeated application of these rules.

Specification

Primitives

Type JSON Value
Int8,Int16,Int32 number
Word8,Word16,Word32 number
Int64,Word64 number 1
Bool true or false
Void null
Float,Double number
String string
Bytes string containing base 64 encoded data
Vector<T> json array containing serialized values of type T
StringMap<T> A json object containing values of type T
Nullable<T> A serialized value T or null

Declarations

Structs

An ADL struct has the obvious json representation - name a json object where each field has the name of the ADL field, and corresponding value is the serialized representation the ADL fields type, ie

struct F
{
    Int32 field1;
    Vector<String> field2;
}

would allow the json value:

{
    "field1" : 42,
    "field2" : ["the","day","is","done"]
}

On deserialization, fields with declared defaults are allowed to be missing. Missing fields are populated with the default value.

Unions

An ADL union is a sum type - only one discriminator can be active at any time. The json serialisation depends on the type of discriminator. If the associated type is Void then the union is serialised as the discriminator name (ie a json string). For any other type, the union value is serialised as a json object with a single name /value pair: the name of the active discriminator, and it's serialized value.

Hence

union F
{
    Void empty;
    Int32 field1;
    Vector<String> field2;
}

could have any of the following serialised values:

"empty"

{ "field1" : 42 }

{ "field2" : ["the","day","is","done"] }

Newtypes

An ADL newtype has exactly the same serialization as the underlying type ie:

newtype ScopedName = Vector<String>;

would have a representation like

["org","adl","ast"]

Generics

Only monomophic values can be constructed, and hence actually be serialized. The json serialized values match the serialization of the generic type with type parameters fully substituted. ie with a definition

union Maybe<T>
{
   T just;
   Void nothing;
};

The following would be valid json values for a type Maybe<Vector<String>>:

{ "nothing" : null }

{ "just" : ["Sydney","Melbourne","Darwin"] }

Custom Field Names

The field names in structs and unions can be annotated with SerializedName to set the json field name to differ from the name in the generated code. Hence

struct Point
{
    @SerializedName "x"
    Double xvalue;
    @SerializedName "y"
    Double yvalue;
};

would allow the json value { "x" : 5, "y" : 7 }.

Backwards Compatibility

The json serialization scheme permits limited backwards compatibility. Fields can be added to structs provided the default value for those fields ensures compatibility. Discriminators may also be added to unions, provided that the default descriminator is not changed.

The custom type machinery can be used to implement more generalized backwards compatibility behaviour.

Footnotes

  1. It's not clear that a json number is the best representation for 64 bit values. The [json specification][jsonspec] doesn't specify a precision for numbers, but javascript implementations typically represent all numbers as a double, which lacks the precision to represent all Int64 values. [jsonspec]:http://www.json.org/ ↩