What is COM/DCOM. Distributed Object Systems 4 COM/DCOM. COM vs Corba 1. COM vs. Corba 2. Multiple inheritance vs multiple interfaces

(1)

Distributed Object Systems – 4 –

COM/DCOM

Piet van Oostrum

Sept 18, 2008

DOS-4

What is COM/DCOM

I Component Object Model

I Components (distributed objects) à la Microsoft

I Mainly on Windows platforms

I Is used in large parts of Windows itself

I COM is local version, DCOM is distributed version

I Microsoft wants it to be replaced by .NET

I Originally developed to connect MS Word and Excel

I OLE: Object Linking and Embedding

I Later: generalisation: basic object communication mechanism could be used for other things: COM

I Mozilla’s XPCOM and UNO (OpenOffice.org) are similar

Piet van Oostrum DOS-4 1

DOS-4

COM vs Corba –1–

I Corba is platform independent

I COM is Microsoft-only

I Corba is multi-vendor

I COM comes with Windows.

I Corba must be installed separately

I Corba specifies how remote objects must be called (source code)

I COM defines a binary standard (implementation)

I Corba and COM have a similar structure (Services, etc.)

DOS-4

COM vs. Corba –2–

I Interfaces

I COM objects can have more than one interface

I Corba objects have a single interface

(Corba Component Model supports multiple interfaces) I Identity

I Corba objects are identified by an object reference

I COM objects have one or more interface pointers I Inheritance

I COM and Corba have interface inheritance

I COM has single inheritance

I Corba has multiple inheritance

Multiple inheritance vs multiple interfaces

Suppose you have an interface for printing and an interface for storage:

I Print interface:

I set_page(pagesize)

I build_pages(page_numbers)

I printit() I Storage interface:

I open_container(filename)

I read_object(obj_id)

I write_object(obj_id)

I Defining an object that supports both interfaces:

I In COM: the object implements both interfaces

I In Corba: inherit from both:

interface MyClass: Print, Storage

COM – What’s in it?

I COM is a mechanism to call objects (local or remote), compare Corba’s ORB

I Many services are defined that use COM (big parts of Windows O.S.)

I OLE Documents:

I Linking

I Embedding

I In-place activation

I (OLE) Automation (how to command an application from another application)

I Events . . .

(2)

DOS-4

Interfaces

I The interfaces to an object must be implemented in a specific binary way.

I This is derived from the way Visual C++ implements objects and methods.

I For COM objects this implementation must be used in all programming languages

I The interface can be described with (Microsoft) IDL

DOS-4

Visual C++ object

m4

method code

method table vtable

var1 var2

m1 m2 m3

I Object have a pointer to avtable(an array of method pointers)

I This is followed by the instance variables

I Often a pointer to the object is passed around (object*)

DOS-4

COM Interface

m4

code method

method table

vtable m1

m2 m3

I A COM interface is implemented as a pointer to a pointer to a list of function addresses (interface pointer)

I A COM object must implement at least one interface

I A COM object may implement more than one interface

I COM objects written in VC++ can use a pointer to themselves (this) as interface pointer

DOS-4

COM Interface - Subclasses

class A { m1(){}; m2(){}; m3(){}; m4(){}; } class B extends A { m5(){}; m6(){}; }

m6

vtable m1

m2 m3 m4

m5

vtable m1

m2 m3 m4

B A

I The vtable for a subinterface is just an extension (more methods at the end) of the superinterface

I so the subinterface pointer can be used where the superinterface is needed

Interface pointer Notation

Interface B Interface A

Object

client uses interface pointer:

Client Object

In the case of a remote object, the pointer refers to a proxy object.

There is no concept of object pointer

IUnknown

Every interface must have the following methods in the first 3 slots:

I QueryInterface (IID)

I For finding other interfaces I AddRef ()

I For counting how many references there are I Release ()

I For releasing a reference

I These three methods together form the interface IUnknown (Most people use names starting with ‘I’ for interfaces)

(3)

DOS-4

Identification

I A method from an interface is not identified by name, but by its position in the method table

I 0=QueryInterface

I 1=AddRef

I 2=Release, etc.

I An (abstract) interface is identified by a 128-bit (usually random-like) number:

I GUID (Globally Unique Identifier) or

I UUID (Universally Unique Identifier)

I UUID is a DCE term.

I There are utilities to generate GUID’s

DOS-4

IUnknown

I QueryInterface on an object with a UUID for an interface as parameter returns an interface pointer for that interface (and the same object) if it exists.

I QueryInterface could return any interface pointer that has the proper slots (e.g. a subinterface)

I AddRef and Release together form a simple garbage collection mechanism (refcount): When the refcount for all interfaces of an object together reaches 0, the object can delete itself

I These three methods can be used with any interface

DOS-4

Object identity

I There are no object pointers

I Any interface pointer for an object can be used to identify the object

I Two different interface pointers could identify the same object

I Object identity:

I A QueryInterface call for IUnknown for the same object should always give the same interface pointer

I regardless of the interface upon which it is called.

DOS-4

MIDL

I Microsofts IDL is derived from OSF DCE’s IDL.

I MIDL is similar to Corba IDL but uses a different syntax

I It also specifies the UUID’s

I Interfaces can be grouped in classes and classes in libraries

I An interface can be used in many classes

I Classes are used to create objects

I Classes and libraries are also identified by GUID’s

I Class GUID’s (CLSID) are stored in the Windows registry with information how to run the code that implements the objects

Example IDL

[object, uuid(3CFDB283-CCC5-11D0-BA0B-00A0C90DF8BC) ] interface IStat: IUnknown {

import "unknwn.idl";

HRESULT init();

HRESULT addItem([in] double item);

HRESULT getAvg([out] double *avg);

HRESULT getNum([out] long *num);

HRESULT getAll([out] double *avg, [out] long *num);

};

Conventionally a HRESULT return code signifies if the operation succeeded or not (there are no exceptions).

Method results are returned through out parameters.

MIDL types

I boolean – 8 bits I byte – 8 bits.

I char – 8 bits.

I double – 64-bit floating point number.

I float – 32-bit floating point number.

I hyper – 64-bit integer.

I int – 32-bit integer.

I long – 32-bit integer.

I short – 16-bit integer.

I small – 8-bit integer.

I wchar_t – 16-bit wide characters.

(4)

DOS-4

Examples of structs and arrays

#define MAX_INDEX 10

typedef char ATYPE[MAX_INDEX];

typedef short BTYPE[]; // Equivalent to [*];

typedef long CTYPE[*][10]; // [][10]

typedef float DTYPE[0..10]; // Equivalent to [11]

typedef float ETYPE[0..(MAX_INDEX)];

typedef struct

{ unsigned short size;

unsigned short length;

[size_is(size), length_is(length)] char string[*];

} counted_string;

DOS-4

Use of dynamic arrays

HRESULT MyFunction(

[in, out] short * pSize,

[in, out, string, size_is(*pSize)] char a[0..*]

);

I pSize is the size of the array, given as separate parameter.

I IDL tells that this parameter really is the size of the array (needed for marshalling the proper number of elements)

I size_is is the number of elements reserved, length_is is the actual number of elements present if different

DOS-4

Unions

Encapsulated union (with a discriminator) typedef union S1_TYPE

switch (long l1) U1_TYPE { case 1024:

float f1;

case 2048:

double d2;

};

S1_TYPE will be a struct in C/C++ with members the long l1 and the union with name U1_TYPE.

There are also other notations.

DOS-4

Where are objects running?

I In a DLL:

I The DLL is loaded if necessary (first occurrence) and the client gets the real pointer to the object interface (in C++ could be the object pointer).

I Runs in the same process as the client

I This is calledinprocserver

I Method call is as fast as ‘normal’

I In another process on the same machine

I The process is started if necessary

I In the client process a proxy object is created with the same interface

I The proxy object uses RPC mechanisms to the real object

I This is calledlocalserver

Where are objects running?

I In another machine

I Is used similar to local server

I RPC is over the network

I In the registry there is information about inproc or local server

I Also contains the location of the DLL and/or .exe file

I The client can specify inproc or local server

I For DCOM if the object is not found on the local machine, COM asks other machines if the object is there

I Remote objects must be separately registered in the registry

How to create (or get) objects

I Use the class object

I The COM runtime system (Service Control Manager) can find the class object, given a CLSID, from the registry

I It will also start a server or load a DLL if necessary

I The Class object should be a singleton (only one instance)

I It should provide methods to create objects or find existing ones.

I The function CoGetClassObject gets the class object, given a CLSID, interface ID and inproc/local/remote

I It returns an interface pointer for the class object

(5)

DOS-4

Integrated method

I Use CoCreateInstance

I This creates a class object and calls its CreateInstance method to get an object (instance)

I Must implement interface IFactory

I There is also an interface IFactory2 which supports licensing

I CoCreateInstance parameters:

I class id

I an Outer pointer (see later)

I inproc/local/remote

I interface id (GUID) for the object

DOS-4

Client code

double value;

CoInitialize(NULL); // initialize COM

CoCreateInstance(CLSID_CStat, NULL, CLSCTX_SERVER, IID_IStat, (void**) &pIStat);

pIStat->addItem(value);

pIStat->QueryInterface(IID_IStat2,

(void**) &pIStat2);

pIStat->Release();

pIStat2->count(value);

pIStat2->Release();

CoUninitialize();

DOS-4

Client side

(3) (5)

class factory

object

COM library COM runtime system

COM library Client

(6)

[1] [3]

[4]

[2]

[5]

(1) (2)

(4)

Server

1: CocreateInstance, 2: server started,

3: class object created, 4: IFactory interface called, 5: object created, 6: object interface pointer returned, [1],[2],[3]: calls to pIStat,

[4],[5]: calls to pIStat2

DOS-4

Implementation

Every object must implement the IUnknown methods:

#define STDMETHODIMP __stdcall HRESULT

STDMETHODIMP Stat::QueryInterface(REFIID riid, void** ppv) { if (riid == IID_IUnknown || riid == IID_IStat)

*ppv = (IStat*) this;

// if more interfaces are supported // else if (riid == ...

else { *ppv = NULL; return E_NOINTERFACE; } AddRef(); return S_OK;

}

In Visual C++ there are different function calling mechanisms.

For COM objects __stdcall must be used.

RefCount

int refcnt;

HRESULT

STDMETHODIMP Stat::AddRef() { refcnt++;

}HRESULT

STDMETHODIMP Stat::Release() { refcnt--;

if (refcnt == 0) { } ...

}

Class Factory

STDMETHODIMP HRESULT

CClassFactory::CreateInstance(LPUNKNOWN p, REFIID riid, void** ppv) { IStat* punk = (IStat*) new Stat();

HRESULT hr = punk->QueryInterface(riid, ppv);

return hr;

}

The CreateInstance method creates a new Stat object and returns its interface pointer for the interface iid.

QueyInterface will do the required AddRef.

(6)

DOS-4

Our own Methods

HRESULT

STDMETHODIMP Stat::addItem (double d) { sum += d;

n++;return S_OK;

} HRESULT

STDMETHODIMP Stat::getAvg (double* d) {

*d = sum/n; // should check n!=0 return S_OK;

}

DOS-4

Server code

void main() { ...

hr = CoInitializeEx(NULL, COINIT_MULTITHREADED);

CClassFactory* fact = new CClassFactory;

hr = CoRegisterClassObject(CLSID_CStat, fact, CLSCTX_SERVER, REGCLS_MULTIPLEUSE,

&dwRegister);

...

register the class object

DOS-4

Finally

I COM is optimized for local calls (especiallyinproc).

I Is a very popular mechanism on the MS Windows operating system

I DCOM (distributed version) has never become popular

I Is now gradually superseded by .NET

I Mainly used in C, C++ and Visual Basic

I It has a lot of complicated issues

I Java support is minimal

I No multi-platform support