home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sys.next.advocacy
- Path: sparky!uunet!microsoft!hexnut!edwardj
- From: edwardj@microsoft.com (Edward Jung)
- Subject: Re: Weight of NT OS-level "objects"?
- Message-ID: <1993Jan04.002835.1342@microsoft.com>
- Date: 04 Jan 93 00:28:35 GMT
- Organization: Strategy Division, Microsoft Corp.
- References: <1992Dec31.234458.6859@fnbc.com>
- Keywords: NT,Smalltalk,Windows,Microsoft,Object Oriented
- Lines: 50
-
- In the newsgroup comp.sys.next.advocacy <1992Dec31.234458.6859@fnbc.com> you wri
- te:
- >Will OO languages shipped for NT be able to base objects on NT objects, or
- >will objects in a running app exist in a whole different name space? I
-
- NT objects are all kernel objects, so they are generally not used
- by user-level processes (which generally don't need -- and for safety
- and security should not have -- the ability to define kernel objects)
- The NT object manager can be extended to handle more object types
- (although there are many limitations to this), but it is not expected
- that this is the way developers get OO support. Even subsystems (i.e.
- Win32, Win16, OS/2, POSIX) do not define new NT kernel objects since
- they are user-level code (similar to Mach 3.0's architecture, but
- not NeXT's, which uses a variation of an earlier Mach). NT objects
- are very low overhead (hey, they're in the kernel!), and in many
- ways are thus also low functionality (they do not, for example,
- transparently distribute across networks).
-
- In some sense, microkernel architectures are not the way to TRULY
- support objects from "top to bottom". In fact, virtually all of
- today's chip architectures don't support objects efficiently. Any
- object system running on a modern kernel of traditional design
- such as Mach, NT, OS/2, etc. is generally layered on top of
- traditional kernel services which are in turn on top of traditional
- chip architectures. There exist research object-oriented systems
- (like Choices) that are "ground-up" OO, and a number of failed
- chip architectures that support OO better. Many IBM chip
- architectures are better at supporting objects than others;
- these are not found in workstations or personal computers.
-
- Microsoft defined a basic object model and released it in a
- set of interfaces called "Object Linking and Embedding" (OLE).
- This object model is primative and only contains the minimal
- definition required to enable linking and embedding functionality.
- The next version (widely available in beta) also includes
- storage and external programmability functionality. A future
- systems product will extend this model for transparent
- distribution and "plug and play" protocols, and base the entire
- suite of OS services on this model.
-
- A point of fact that is lost on many people is that OLE is
- the object model that has the widest use today by applications.
- It is neither a class library nor a language-dependent model
- (although there are class libraries that "make it easier"
- shipped by various other parties). And, I'll admit, it has
- an ugly API and no interesting tools. But we're working on it ;-\
-
- --
- Edward Jung, Software Architect edwardj@microsoft.com
- Advanced Systems, Microsoft Corp.
-