FreeLists / haiku-appserver / [haiku-appserver] Re: getting a clone of the current Accelerant

[Stephan Aßmus <superstippi@xxxxxx>]

On 2005-12-01 at 18:39:16 [+0100], Axel Dörfler <axeld@xxxxxxxxxxxxxxxx> 
wrote:
> Stephan Aßmus <superstippi@xxxxxx> wrote:
> > > /boot/optional/sample-code/drivers/graphics/test_harness
> > When I compiled and ran it, it froze up my system in a way that
> > reminded me
> > of the good old C=64 cracked games days... or more recently the days
> > of a
> > more instable Haiku app_server. :-) But I think this is exactly what
> > I
> > need. Cool, thanks! I'm thinking about making a "perfect" simulation
> > of my
> > new clipping stuff, involving BDirectWindow and direct banging of the
> > Accelerant :-) completely avoiding any app_server communication.
> > Unless
> > someone tells me to take it easy.
> 
> I don't know if it's worth the effort, when the app_server test
> environment already features most of this. Besides, what exactly do you
> want to test with this?

That's exactly the point, the app_server test environment _doesn't_ feature 
this. And one important difference is that Painter doesn't write to the 
graphics frame buffer like it does under Haiku. There is a noticable 
performance difference which I would like to investigate sometimes in the 
future. Second problem: I cannot test if my threading model works out, 
because after all, each WindowLayer thread is blocking on the BWindow lock 
of the test environment when trying to draw. Third problem: I cannot tell 
how much the app_server communication drains on the overall performance, so 
it is hard to compare my results with R5. When moving or resizing windows 
in the simulation, it looks like it's definitely eating more CPU than 
native window movement under R5 (which involves much heaview redraws 
actually).

For now, maybe you're right though, and I'd better finish with this and 
start integrating with the real app_server. Before I do this, I wanted to 
simulate animations from the client side and scrolling. It should be 
complete otherwise. Before I integrate, maybe we can have a look into the 
code together?

Best regards,
-Stephan