nanogui: Suggestion for adding nanowm into graphic server!
Subject:
Re: [nanogui] Suggestion for adding nanowm into graphic server!
From:
"wang" ####@####.####
Date:
6 Aug 2002 01:37:06 -0000
Message-Id:
>The UNIX socket system in the Linux kernel is fast - real fast. I doubt
>greatly that you will get any noticeable increase in speed by
>integrating the window manager - especially on an embedded system thats
>only running 2 or 3 concurrent processes anyway, because your tall poll
>in terms of speed is (and always will be) the cost of repainting the
>screen. The time it takes for a message to be sent across the socket to
>the server can be measured in microseconds. The time it takes for a
>complex window to completely draw is measured in milliseconds. These
>are the facts of life - as we have always preferred portability over
>optimization.
Alex also mentioned that the speed of drawing window is the primary
reason of slow speed ,and give these suggestions:
* Make the WM do the window move calls at a fixed rate asynchronously to
the incoming mouse movement events in order to limit the number of
GrMoveWindow() calls per second.
* Implement a new Exposure event which returns an exposure region
(rectangle list) instead of a single bounding box rectangle, and modify
the window background code and some of the demo apps to make use of the
new region exposure event. That would stop them from wasting lots of
time continously redrawing the entire area of the window that is being
dragged over them when they only really needed to redraw the two small
rectangles that have been exposed.
Although I don't completely agree with you , you suggestion make
me wanna implement Alex's suggestion first.
I do not think socket is so fast as you said ,in fact I am thinking
about how to bypass the kernel when the clients communicate with the
server.
wang
####@####.####
2002-08-06