JSON Types for Server Methods in DataSnap 2010

by Aug 19, 2009


I am very happy I was granted permission to talk about the new features present in Delphi 2010 concerning database area.

DataSnap 2010 (DS2010) extended the list of server method parameter types with the full suite of JSON types.

It is now possible to pass a TJSONObject instance for example from a client process to a server as input or output parameter and receive a TJSONValue back as an output parameter.

The unit where JSON types are implemented is DBXJSON. In the paragraphs below will be talking about creating JSON objects, writing server methods with JSON parameters and using JSON objects to marshal in and out user types.

Creating JSON objects

JSON objects are regular objects created out of types defined in DBXJSON unit. One can create such objects either instantiating specific types or through parsing of a byte array representing a JSON value according to specs.

Types such as TJSONObject, TJSONArray, TJSONNumber, TJSONString, TJSONTrue, TJSONFalse, TJSONull are derived from TJSONValue and can be instantiated. TJSONPair is a type that can be only added to TJSONObject but it is not in itself a TJSONValue.

The code below creates a TJSONObject with a {"Hello":"World"} object:

obj: TJSONObject;
obj := TJSONObject.Create;
obj.AddPair(TJSONPair.Create('Hello', 'World'));

A JSON pair is formed by a string and a value. The TJSONPair has several constructors that will allow the creation of a pair less verbose. For example, in the example above is assumes that the value is actually a JSON string and creates the appropriate object.

A JSON value can be also created from a byte array using the two TJSONObject class functions ParseJSONValue. They both have similar signatures, one assumes that it has to parse the entire array to create a consistent JSON object while the second expects to use only a part of it for that.

class function ParseJSONValue(const Data: TBytes; const Offset: Integer): TJSONValue; overload; static;
class function ParseJSONValue(const Data: TBytes; const Offset: Integerconst Count: Integer): TJSONValue; overload; static;

By default a JSON object assumes ownership of its elements. Freeing the root object will cascade the release of all JSON pairs and/or values forming that particular instance. Each JSON object has an Owned property that provides ownership hints when an object is destructed. The default value, true, provides the information to an eventual container that the instance can be released together with the container itself. All objects with a false property will be spared assuming that they are owned by other resources (data stores, parameters, fields). This was done to ease the construction of complex objects based on existing ones but owned by separate entities (save memory and CPU).

Each object can be cloned through Clone function. This feature can be handy when server method input parameter need to be stored for longer than method execution itself.

ToString provides a Unicode JSON representation. This should not be used as input for the JSON parser as its content representation is not parser friendly; use ToBytes instead. The following code snippet will demonstrate how to obtain a standard byte representation of a JSON value:

SetLength(ToBytes, JSONValue.EstimatedByteSize);
SetLength(ToBytes, JSONString.ToBytes(ToBytes, 0));

Value property provides the Unicode representation for a TJSONString or a TJSONNumber but it has no use on other types.

Server method with JSON parameters

Using JSON parameter types for server methods is as natural as using them for any method or function; input, var, out and return parameters are accepted.

The following server method uses all possible combination:

function JSONPing(data: TJSONObject; var ok: TJSONValue): TJSONObject;

The DS2010 owns those parameters hence they will be destroyed immediately after method execution.
One cannot return an object that doesn't own; here is where Clone method can be useful: if there is a need for any of the input parameters to be preserved outside the scope of the server method then it needs to be cloned.

It may be possible to keep bits and pieces of them through Owned, but be advised that you are doing it at your own risk.

Handling JSON objects over the client side

Clients can invoke server methods by using the proxy code generated for the server methods or directly using DBXConnection.

The proxy method signature is identical with the server method:

function JSONPing(data: TJSONObject; var ok: TJSONValue): TJSONObject;

The parameter life cycle is controlled by the proxy instance owner flag used for client creation:

constructor Create(ADBXConnection: TDBXConnection; AInstanceOwner: Booleanoverload;

When using the DBXConnection directly, the parameter setter function SetJSONValue accepts a Boolean flag that will control the eventual destruction of its instance upon execution.

FJSONPingCommand.Parameters[0].Value.SetJSONValue(data, true

Marshaling user types into JSON objects

DBXJSONReflect unit provides the tools needed to marshal all user types into and from JSON values.

The marshaling technology is implemented by TTypeMarshaller class is based on traversing user objects and emitting specialized events for each action; an example would be when a new object is about to be visited OnTypeStart is called. The events implementation is done through a specialization of TConverter generic class.

The conversion is left at the latitude of the user; an user type can be marshaled into any representation such as XML or text or binary. Out of the box we provide JSON conversion through TJSONConverter class that can be coupled with TJSONMarshal class.

It is important to be able to restore the source content of the user object from the marshaled data. Due to the fact that some support classes can be quite complex and some types are not well supported by current RTTI runtime converters can be added. We currently support two categories:

  • Field: when the field is encountered on a user type the provided converter is used instead of the default one

  • Type: when a field of that type is encountered the converter is used instead of the default one

Each category has 4 types: convert the field value into another object, a string, an array of objects or an array of strings. For example a set or a record can be transformed on an array of strings.

For each user-defined converter a reverter is required to restore the field value. For example, an array of strings can be restored into a record's content or a set based on field's name or type.

To restore the user object the marshaled data are traversed this time only to generate a deep copy of the source data. Out of the box we provide TJSONUnMarshal that will traverse a JSON object and rebuild the user object.

We will provide a code snippet that will illustrate the marshaling of user objects. The user object types are:

TAddress = record
FStreet: String FCity: String FCode: String FCountry: String FDescription: TStringList;

TPerson = class
FName: string FHeight: integer FAddress: TAddress;
FSex: char FRetired: boolean FQualifications: TStringList;
FChildren: array of TPerson;
FNumbers: set of 1..10;
constructor Create;
destructor Destroy; override
procedure AddChild(kid: TPerson);

We are using record, sets, string lists all combined to describe a TPerson with fields like name or children list.

For this particular example we will consider the FNumbers field transient and we will not marshal it in order to illustrate the warnings that are generated during the process.

The marshal classes used for the user objects and for the JSON equivalent form are declared and instantiated as such:

m: TJSONMarshal;
unm: TJSONUnMarshal;
m := TJSONMarshal.Create(TJSONConverter.Create);
unm := TJSONUnMarshal.Create;

The JSON marshal instance (m) is created with the out-of-the box JSON converter. To restore the user object TJSONUnMarshal instance (unm) is part of the unit.

Dealing with special cases, such as string lists, arrays and records is done through three user defined converters:

m.RegisterConverter(TPerson, 'FChildren', function(Data: TObject; Field: String): TListOfObjects
obj: TPerson;
I: Integer;
SetLength(Result, Length(TPerson(Data).FChildren));
I := Low(Result);
for obj in TPerson(Data).FChildren do
Result[I] := obj;
end end m.RegisterConverter(TStringList, function(Data: TObject): TListOfStrings
i, count: integer begin
count := TStringList(Data).Count;
SetLength(Result, count);
for I := 0 to count - 1 do
Result[i] := TStringList(Data)[i];
end m.RegisterConverter(TPerson, 'FAddress', function(Data: TObject; Field: String): TListOfStrings
Person: TPerson;
I: Integer Count: Integer begin
Person := TPerson(Data);
if Person.FAddress.FDescription <> nil then
Count := Person.FAddress.FDescription.Count
Count := 0;
SetLength(Result, Count + 4);
Result[0] := Person.FAddress.FStreet;
Result[1] := Person.FAddress.FCity;
Result[2] := Person.FAddress.FCode;
Result[3] := Person.FAddress.FCountry;
for I := 0 to Count - 1 do
Result[4+I] := Person.FAddress.FDescription[I];

The first converter is a field converter that is used in the context of marshaling a TPerson's FChildren field. The list of children objects is returned as an array; these objects will be individually marshaled later.

The second converter is a type converter and it will be used for all TStringList fields; it simply returns an array of strings.

Finally, the last converter is again a field converter that deals with a record type; it transforms it into an array of strings with the record's FDescription content is appended to it.

In a similar manner, the reverters are registered with the unm instance:

unm.RegisterReverter(TPerson, 'FChildren', procedure(Data: TObject; Field: String var
obj: TObject;
I: Integer;
SetLength(TPerson(Data).FChildren, Length(Args));
I := Low(TPerson(Data).FChildren);
for obj in Args do
TPerson(Data).FChildren[I] := TPerson(obj);
end unm.RegisterReverter(TStringList, function(Data: TListOfStrings): TObject
StrList: TStringList;
Str: string begin
StrList := TStringList.Create;
for Str in Data do
Result := StrList;
end unm.RegisterReverter(TPerson, 'FAddress', procedure(Data: TObject; Field: String; Args: TListOfStrings)
Person: TPerson;
I: Integer;
Person := TPerson(Data);
if Person.FAddress.FDescription <> nil then
else if Length(Args) > 4 then
Person.FAddress.FDescription := TStringList.Create;

Person.FAddress.FStreet := Args[0];
Person.FAddress.FCity := Args[1];
Person.FAddress.FCode := Args[2];
Person.FAddress.FCountry := args[3];
for I := 4 to Length(Args) - 1 do

Let's instantiate a couple of user objects and see what the outcome is:

person, kid: TPerson;
obj: TObject;
person := TPerson.Create;
person.FName := 'John Doe';
person.FHeight := 167;
person.FSex := 'M';
person.FRetired := false;
person.FQualifications.Add('Delphi developer');
person.FQualifications.Add('Chess player');
person.FAddress.FStreet := '62 Peter St';
person.FAddress.FCity := 'TO';
person.FAddress.FCode := '1334566';
person.FAddress.FDescription.Add('Driving directions: exit 84 on highway 66');
person.FAddress.FDescription.Add('Entry code: 31415');

kid := TPerson.Create;
kid.FName := 'Jane Doe';

v := m.Marshal(person);

obj := unm.Unmarshal(v);
assert( obj is TPerson );
assert( TPerson(obj).FAddress.FDescription nil assert( TPerson(obj).FAddress.FDescription.Count > 0 );
assert( TPerson(obj).FAddress.FDescription[0] = 'Driving directions: exit 84 on highway 66', 'Description');

The output is displayed into the form's memo1 component. For convenience and ease of use we formatted a little and presented it below:

"fields":{"FName":"John Doe",
"FAddress":["62 Peter St",
"Driving directions: exit 84 on highway 66",
"Entry code: 31415"],
"FQualifications":["Delphi developer",
"Chess player"],
"fields":{"FName":"Jane Doe",

Transient fields will be signaled as warnings. You can check if the JSON representation contains 100% of the user data by checking the HasWarnings function of the marshaller instance.

if m.HasWarnings then
for I := 0 to Length(m.Warnings) - 1 do
memo1.Lines.Add('Transient: ' + m.Warnings[I].FieldName);


JSON introduction to DataSnap 2010 opens this technology to

  • User Objects: server side data can now be accessed by client processes

  • Business Logic: server side rules interact with client side through callbacks

  • Third party applications: Web based applications based on JavaScript can smoothly interact with server methods.

We illustrated how aspects of the goals above can be implemented using JSON types. They are easy to handle, extensible and they can be used to stream structured data to third party applications.

For more information please visit Embarcadero RAD Studio 2010

JSON reflection code can be seen here.