Re: Spec vs. Implementation

On 2004-07-02 11:17:11 +0800 clearm <[email protected]> wrote:

> I think the best way to develop a window system these days is 
> to develop a
> specification instead of a single source tree implementation.  
> You'd simply
> publish the specification (like the X Window spec or Java 
> spec), and let
> anyone who wants to develop an implementation have at it.  The 
> advantage of
> this is:
> 
> 1)  Everyone can develop their own Y Server on whatever 
> hardware/OS they
> want.  All they have to do is make sure that they interface 
> with the client
> API properly.  So, if Joe Blow knows how to do hardware 
> acceleration on a
> Xenocorp Model 300X Ultracomputer then by all means he can 
> create his own
> fully optimized Model 300X Y Server without having to worry 
> about having to
> target SDL, DirectX, OpenGL, GGI, DirectFB, SVGALIB, iPAQ, 
> Palmpilot, etc.
> in the same source tree.
> 
> 2)  In the same vane as point #1 above, it simplifies the 
> reference
> implementation (i.e. the current source tree) because it 
> doesn't have to be
> portable.  We don't need a gazillion abstraction layers in the 
> reference
> implementation to account for every single possible hardware/OS
> configuration.
> 
> 3)  They can write the Y server in whatever programming 
> language they want.
> It can be C, C++, Lisp, even friggin perl if they want.  So 
> long as a client
> can connect to it and send Y messages, it doesn't matter what 
> language the
> server is programmed in.
> 
> 4)  There can be more developers working/exploring the 
> project.  With one
> source tree, only a few people can submit changes without 
> screwing it up.
> You don't want everyone and their grandmother dipping into the 
> source and
> changing things.  But if there are multiple small groups 
> implementing their
> own Y servers they can do as they please.
> 
> 5)  Makes it easier to develop.  I guess this is just a 
> rehash, but if
> everyone can use their own language and coding style; develop 
> and their own
> pace; and not have to flip through a huge, over-abstracted, 
> single base of
> code; then it will be easier for them to work on the project.  
> For example,
> when I look at a function called DrawWindow() I want to see 
> code that puts a
> grey box on the screen.  I don't want to see two or three 
> layers of
> abstraction, pointers to user-defined functions, structures 
> that hold
> who-knows-what information, and other garbage.
> 
> Whaddya think?
> 
> Mack
> 
> 
> 

I like that idea. Then all Mark has to do is make a reference 
implementation.