COM
A Slideshow of Epic Proportions
Sean Baxter
What is COM?
• A standard for interoperable software
• A runtime library providing low-level or hard-to-implement functionality
• A competitive, capitalist marketplace for selling functionality to application
developers
That Last Slide Was Pulled From Booboo’s
Rear End. What is COM, Really?
• A sacred covenant held by the world’s best developers. All software functionality is
encapsulated through a COM class (coclass) and exposed through COM interfaces
Blah Blah Blah Blah Blah Blah Blah Blah Blah
Booboo
Interface and Coclass Fundamentals
• An interface is a group of methods• A coclass is a group of interfaces
• All coclasses implement the IUnknown
interface
• All other interfaces derive IUnknown
• Lollipop diagrams describe coclasses
I U n k n o w n
O i n k
About Interfaces
• The client communicates with the server through interface pointers (think base class pointers)
• Methods implemented by the coclass are invoked through interface pointers
• Interface pointers are acquired through IUnknown
• Because all interfaces derive IUnknown, any interface pointer on an object can be acquired through any other interface pointer on the object • The method to acquire an interface pointer is
More About Interfaces
• Interfaces are equivalent to C++ abstract base classes
• Because there is no standard thunk mechanism, interfaces do not support multiple inheritance
• Two IUnknown pointers point to the same
object if their values are equal
• “Dispatch interfaces” are special case interfaces allowed by the COM
Interfaces and Virtual Tables
• An interface pointer points to the
corresponding slice of a COM object
• The first 4 bytes of this slice comprise a pointer to the interface’s virtual table
• Each slot in this virtual table is the address of an interface method
• A method implementation may be
p A d d R e f p R e l e a s e p F o o F u n c 1 p F o o F u n c 2 p F o o F u n c 3 p Q u e r y I n t e r f a c e
p A d d R e f p R e l e a s e p B a r F u n c 1 p B a r F u n c 2 p B a r F u n c 3 p Q u e r y I n t e r f a c e
p A d d R e f p R e l e a s e p Q u e r y I n t e r f a c e
I U n k n o w n
O i n k
I F o o I B a r
m _ p U n k V t b l m _ p F o o V t b l m _ p B a r V t b l
Q u e r y I n t e r f a c e ( ) A d d R e f ( ) R e l e a s e ( )
F o o F u n c 1 ( ) F o o F u n c 2 ( ) F o o F u n c 3 ( )
B a r F u n c 1 ( ) B a r F u n c 2 ( ) B a r F u n c 3 ( )
Reference Counting
• To regulate destruction of objects, COM requires that all coclasses implement a reference counter
• COM objects can have their methods
invoked from other threads, processes, and even machines. Reference Counting
Reference Counting Rules
• Each object is given a reference count of 1 upon instantiation
• IUnknown::AddRef increments the reference
count
• A successful IUnknown::QueryInterface call
increments the reference count
• IUnknown::Release decrements the reference
count
So Why All These Interface and
Reference Counting Formalities?
• One of the most amazing features of COM objects is location transparency: COM marshals interfaces between processes and remote computers
• Reference counting guarantees that a client cannot prematurely destroy a shared object
Speaking of Location Transparency...
• COM objects (servers) can be delivered in two module types: EXEs or DLLs (aka
OCXs)
• EXEs cannot run in-process • DLLs can only run in-process
An In-process Server
(same apartment)
I U n k n o w n
F o o
I F o o
p A d d R e f p R e l e a s e p S h a k e
p R a t t l e p R o l l p Q u e r y I n t e r f a c e
p A d d R e f p R e l e a s e p Q u e r y I n t e r f a c e
m _ p U n k V t b l
m _ p F o o V t b l
Q u e r y I n t e r f a c e ( )
A d d R e f ( )
R e le a s e ( )
S h a k e ( )
R a t t l e ( )
In-process Server
(different apartment)
S T A
C O M I n v i s i b l e W i n d o w
I U n k n o w n
O i n k _ S t u b
I F o o
I B a r I U n k n o w n
O i n k
I F o o I B a r I U n k n o w n
O i n k _ P r o x y
I F o o I B a r
C l i e n t C o d e
Local Server
I U n k n o w n
F o o _ P r o x y
I F o o
p A d d R e f _ P r o x y p R e l e a s e _ P r o x y p S h a k e _ P r o x y
p R a t t l e _ P r o x y p R o l l _ P r o x y p Q u e r y I n t e r f a c e _ P r o x y
p A d d R e f _ P r o x y p R e l e a s e _ P r o x y p Q u e r y I n t e r f a c e _ P r o x y
m _ p U n k V t b l m _ p F o o V t b l
Q u e r y I n t e r f a c e _ P r o x y ( ) A d d R e f _ P r o x y ( ) R e l e a s e _ P r o x y ( )
S h a k e _ P r o x y ( ) R a t t l e _ P r o x y ( ) R o l l _ P r o x y ( )
Q u e r y I n t e r f a c e _ S t u b ( )
S h a k e _ S t u b ( ) R a t t l e _ S t u b ( ) R o l l _ S t u b ( ) P r o c e s s
B o u n d a r y
I F o o
I U n k n o w n
F o o
m _ p U n k V t b l m _ p F o o V t b l
Remote Server
I U n k n o w n
F o o _ P r o x y
I F o o
p A d d R e f _ P r o x y p R e l e a s e _ P r o x y
p S h a k e _ P r o x y p R a t t l e _ P r o x y p R o l l _ P r o x y p Q u e r y I n t e r f a c e _ P r o x y
p A d d R e f _ P r o x y p R e l e a s e _ P r o x y p Q u e r y I n t e r f a c e _ P r o x y
m _ p U n k V t b l m _ p F o o V t b l
Q u e r y I n t e r f a c e _ P r o x y ( ) A d d R e f _ P r o x y ( ) R e l e a s e _ P r o x y ( )
S h a k e _ P r o x y ( ) R a t t l e _ P r o x y ( ) R o l l _ P r o x y ( )
Q u e r y I n t e r f a c e _ S t u b ( )
S h a k e _ S t u b ( ) R a t t l e _ S t u b ( ) R o l l _ S t u b ( )
I F o o
I U n k n o w n
F o o
m _ p U n k V t b l m _ p F o o V t b l
C l i e n t M a c h i n e
S e r v e r M a c h i n e
Things To Note
• All interface method calls are synchronous; a
remote call to a New York server must complete its round-trip before the client can continue execution • Asynchronous services are available through
Microsoft Message Queue (MSMQ) and ICallFactory (Windows 2000)
• Encryption, authentication, and impersonation services are available with remote server
Confused?
Where Am I?
COM Owns You
So what does all this nonsense about location transparency and reference counting have to do with anything?
WRONG!
They
are
COM.
MS
Marketecture
:
Common Misperceptions
• OLE is a system of linking and embedding data into a document (hence the acronym “Object Link and Embedding”)
• ActiveX, like Java, is a system for building web page applets
• Microsoft Transaction Server (MTS) processes business data from the internet
The Journey of COM
• Dynamic Data Exchange (DDE) was introduced while the Earth was still cooling.
• DDE was unsavory. MS unveils OLE 1.0 after the villagers riot
• OLE 1.0 is a big success but hard to program. MS unveils OLE 2.0 for Windows 3.1: COM is born (it’s 16bit)
The Journey of COM (II)
• To celebrate Windows 95, MS blesses COM with the apartment model
• 1996 sees Windows NT 4. DCOM brings network and multithreading enhancements to COM
• Microsoft Transaction Server (MTS) becomes
available for NT Server. This doesn’t process web business data. It’s a COM runtime environment which allows components to run more efficiently (solving scalability difficulties often faced in
The Journey of COM (III)
• OLE-DB provides consistency when accessing data from a variety of providers
• MS makes web browsers a priority (IE3) and
simultaneously coins the term “ActiveX,” adding to the COM marketecture. Like OLE, ActiveX is a
synonym for COM.
• Microsoft Message Queue (MSMQ) delivers
asynchronous (and vastly more reliable) method calls • Visual C++ 5.0 ships with ATL 2.1. OLE Control
The Journey of COM (IV)
• Windows 2000 ships in 2004. Microsoft introduces COM+ acronym: COM+ is simply COM on Win2k • Win2k includes many COM updates including a
third threading model (the Neutral Threaded
Apartment) and integration of MTS into the COM runtimes
• COM+ 2.0 delivers legendary COM virtual machine
• Windows NT 6 device contexts replaced by DirectDraw surfaces. Classic OLE Control
COMprehension
• COM is a set of many, many technologies. • COM is really Microsoft’s software
engineering strategy
• Computer people (you) are betting the bank on Microsoft
COM Defined
A single sentence can’t define COM,
because COM is such a monumental and omnipresent set of technologies.
So what is a COM technology? Any
COM in Everyday Life
The Month View is from the Microsoft Windows Common
Property Pages
Events
Standard Interfaces
• Programming sucks. Best to fall back on platform-provided functionality
• A container/client can interact with any
server/control simply by implementing the required site interfaces
• Application developers who license your components are already familiar with the interfaces
What Are Some COM-Based
Technologies?
• OLE Controls • OLE-DB
• ADO • DAO • DirectX • TAPI 3.0
• Exchange Server • Exchange Client • Active Scripting
• CDO 2.0 • MAPI
• SQL-DMO • SQL-NS
• Microsoft Repository • Broadcast Architecture • Still Image API
Still More COM Technologies!
• Active Desktop • IIS
• Microsoft Agent
• Microsoft Java SDK • NetMeeting
• NetShow
• Microsoft Wallet • Active Directory
• Active Accessibility • Clustering Service
• Fax Services • MSMQ
• RAS
• Synchronization Manager • Structured Storage
• MTS • MMC
The Glory of COM
• COM eliminates thread synchronization and IPC/RPC confusion
• COM enables real polymorphism: language
transparency and location transparency • COM mandates a consistency among
component interfaces to reduce learning curves • COM allows versionless updates
DLL Server Mechanisms
• DllRegisterServer and DllUnregisterServer manipulate the registry
• DllCanUnloadNow tells COM when the module is not being used
• DllGetClassObject provides the client with the class factory of the required COM
EXE Server Mechanisms
• A /RegServer argument registers the EXE • A /UnregServer argument un-registers the
EXE
• CoRegisterClassObject registers a class factory for clients to connect to
COM Object Instantiation
• CoGetClassObject (wrapped by
CoCreateInstance and CoCreateInstanceEx) launches a module and retrieves its class
factory
• CoGetClassObjectFromURL downloads a module from an URL, decompresses it,
registers it, and invokes CoGetClassObject on it
COM Security
• CoInitializeSecurity sets the authentication requirements for the current process
• The COSERVERINFO structure specifies the machine name and authentication info for
clients connecting to remote objects
Books of COMpetency
• The C++ Programming Language (Bjarne Stroustrup)
• Programming Windows (Charles Petzold) • Advanced Windows (Jeffrey Richter)
• Design Patterns (Gamma, Helm, Johnson, Vlissides)
Hoorah For Reading!
• Effective COM (Box, Brown, Ewald, Sells) • Inside OLE (Kraig Brockschmidt)
• Professional ATL COM Programming (Richard Grimes)
• Microsoft Systems Journal (www.microsoft.com/msj)
May the COM Be With You