Well, If I throw a ball to you your hands will hopefully be in the right place to catch the ball before it arrives.
That's because Newtons laws of gravity and motion are more-or-less perfect for ball catching. Where greater times, distances and speeds are involved Einstein's theory of relativity becomes more useful.
Scientists hedge their bets on those laws and previous observations of stars and galaxies being good enough to estimate where this are now and where they were.
This could all be in my head and everything's just a dream, I'll hedge my bets it isn't.
Or they could keep on with the old numbering system dev could work on 2.7 and backport to 2.6 once it's stable enough.
Maybe Linus doesn't have any idea of what to put into a 2.8 kernel so branching to 2.7 would be pointless.
Filing system devices would be nice so that soft links appear for device-driver bindings and you can install a new driver using cp or ln, and it would be nice if the kernel was split into drivers and kernel to giver comercial operatiors half a chance of making a binary driver that will install for at least a couple of kernel releases.
Yeh, the difference between 2.6.8 and 2.6.10 iss only a partial re-write of a driver when the memory allocations systems change in the kernel not a HUGE-scale re-write that a change from 2.4 to 2.6 would require.
So potentially if I'm providing an offsite backup service without any trademarks then the distribution part is OK, it's just the copyright part I have to worry about.
Count words, I just happen to have Michel Abrash's graphics programming black book on the table and it's got a good discussion on count words.
1: The data is static, so it fits you theory. 2: The C version is almost optimal 3: The java version is not comparable to the C version for real world tasks (I haven't bothered to re-write it to be the same as the c version to see if it's performance is better or not).
HeapSort why not quick sort? don't most languages come with quick sort built in and heap sort is almost identical to quick sort (variants on the binary merge).
Object methods... 1: the C 'object' and the java object are not the same, the java object provides a lot of extra functionality that may well be used in a real world application that has not been put in-place in the c version. 2: He didn't even use final in the Java version, how is using an almost static C 'object' equivalent to using a non-final java object in real world tasks.
That's the theory. In practice, this has not yet been achieved outside very tight constraints, such as where the data does not change very often.
Which accounts to 95% of my hard drive, so tight, but common.
Did you even READ the fucking site
No I just plucked It does say that start up time is included too, so that taints the benchmarks. out of thin air.
No longer the case. If you actually bother to look at the fucking benchmark So their fucking benchmarks now, there were nice benchmarks that showed you point earlier.
I read the FAQ, if it's wrong what does that tell me I should assume about the fucking benchmarks?
a profiling Just in time JVM can profile the code while it is running and then next time around recompile the code to be faster.
That site doesn't list how the tests were carried out except to say that there were, no details on any profiling or optimization attempts that any serious deployment would use.
It does say that start up time is included too, so that taints the benchmarks.
Despite this, the police turned up in force and told us we wouldn't be having a party. When questioned on what laws or powers they were using to shut us down (much to their dismay, the criminal Justice Act and the licensing laws do not cover free warehouse party's). The policeman in charge replied: "What are you, a fucking Barrister now...? You're not having a party because I SAY SO..!"
Numerous policemen then lined up in an attempt at preventing people getting into the warehouse. But when it became apparent that we weren't going to be intimidated into not having a party, they waded in and began indiscriminately beating party people with their truncheons. As a result of this action, alot of people received injuries, some of them quite serious. One woman was on the floor being beaten by at least 3 or 4 policemen and one man was held down over a railway line by two policemen who repeatedly smashed his face into a sleeper until his nose caved in.
' and easier to maintain code than today's so called high level languages.'
Java can be faster if you use a profiling JVM, (well unless you profile your C code and get gentoo to re-emerge with using the profile every other week)
this should also apply to.NET
Java is also much easier to maintain that C, because it's almost exactly the same as C with enough bits missing to let it run in a Virtual Machine and some extra rules inplace.
Of couse bad C is probably easier to maintain that bad Java and Bad C++. In which case you would have to start argueing.
Use OSS because all software is bad, it's just that our version of bad is easier to maintain....
your using the 'anti' claus, your not saying were better because were the best your saying. We use C isn't better because they try to sell a new compiler (who SUN and GCC?).
Not we use C it's better because it's simpler and more compact that many of todays languages that have lots of bolted on features that havent matured yet.
Remember, positive energy activates constant elevation. Be positive about OSS not negative about everything else.
obviously you know nothing about the British government, all there laws seem to stem from random debate.
Personally I think if they should be taxing anyone it should be the companies that use our personal information. 10% of the transaction cost as an extra tax should bring in a nice revenue and make companies think 1.5 times before using your personal information.
It only counts if the property doesn't belong to you.
breaking and entering
'housebreaking: trespassing for an unlawful purpose; illegal entrance into premises with criminal intent '
'The gaining of unauthorized, illegal access to another's premises, as by forcing a lock.'
'The act of gaining passage into and entering another's property (as a building or vehicle) without privilege or by force; also The crime of breaking and entering see also burglary'
'Breaking and entering is defined as the crime of illegally entering a residence or other enclosed property using any amount of force (even pushing open an unlocked door). If criminal intent can be established, breaking and entering can be considered a burglary, a felony in most U.S. jurisdictions. If there is no criminal intent, breaking and entering can be considered trespassing, which is usually a misdemeanor crime.'
All they had to do was merge the 'search' and the 'search dns' inputs and it would be great. as it is there are two inputs, one called 'search' that umm searches, and one called go that umm... searches in a different way.
One 50Char box that searches unless you type in a domain name, or news:// etc.. or two 20 char boxes, the choice isn't yours.
'This is one of those things that people keep repeating as if that makes it true. X isn't slow,' Did I say X? no. anyhow for the sake of your argument windows slow, X slower why do you think they wrote the xDamage and xComposite extensions?
'If you start adding tons of animation and complex effects to the GUI, it's going to start requiring scads of memory and processing time even if it's all done in 2D.'
Yes, Mr I want a chroma colour green screen monitor for xmas.
' You think vector graphics are fast on a CPU? You think font glyphs don't take up space in main memory now?' What are you going on about, I think that vectors should be real-time, not static pre-rendered pixelmaps.
'Lets see your PIII try to render it in 2D on the CPU. I'll bet my six year old Geforce2 MX handles it a lot better. And while your CPU is choking under the burden, my GPU will be happily using its antialiasing hardware to make the whole thing look great, text and all.'
my PIII has a green screen and it runs 1990's software. If you want to be stuck in the 90's with you FSAA then feel free, the rest of us can move with the times. I can have all the effects in the world and still run at 50fps with a low CPU overhead while your left with 6-8 updates a second taking 5% of your CPU. Hmm.. I wonder what they'll be showing off at expo this year.
IANAL, so if you sue me, I'll get professional help.
I claim:
1. A virus detection system for detecting if a computer file is infected by a virus, the file having a plurality of potential virus entry points, the system comprising:
A virus checking system, to check a file that could have a virus embedded in it at any location. I've interpreted plurality as any, and not one of many
an engine for controlling operation of the virus detection system responsive to instructions stored in an intermediate language, the instructions adapted to examine the plurality of potential virus entry points and post for emulating ones of the plurality of potential virus entry points exhibiting characteristics indicating a possible virus;
An automated process for sending in one go a sequence of commands to the virus checking system, that then (repeat) goes on to check the 'plurality' of entry points.(/repeat), where A virus may be embeded.
a scanning module coupled to the engine for scanning regions of the virtual memory for a signature of the virus responsive to the engine and the emulating module, wherein presence of the virus signature in a scanned region indicates that the file is infected by the virus.
something that check 'virtual' memory for a 'virus' when told to by the above automated process (who knows what the emulating module is?), when it find a virus if somehow magically knows what 'the file' is supposed to mean indicates [to the user?] that the file is infected.
Ok, virtual memory is 434 and the file is, 100. But I couldn't find the definitions anywhere.
Looks like a poorly though out virus checker that can take multiple commands in one instruction from a separate process that knows where the virus may be embedded, that also checks virtual memory for the virus via a 'module'
It has been an obvious way to check for viruses ever since I knew about them (well before virus checker were around). It just wasn't a practical way at the time.
Yes. I don't think you have any idea how 3D rendering works, or if you do it's completely screwed up. I can write windowing systems, WYSYWIG editors, and 3D graphics engines that will run on a 268 with a few MB of ram.
If you want to know how to write a WYSYWIG editor in a few Mb than I'll tell you, I'll also tell you why as soon as you get a little more complex you use tens of MB and even more complex and you start hitting hundreds.
Here's a simple walk through or a rendering system.
1: all your textures are pixel maps, pixel MyTexture[usize][vsize]. (not no mip-mapping, volume textures or cube, just simple textures)
2: You then create a number polygons to render on screen, this is for something simple like Unreal Tournament or X,
3: at the vertex of each polygon you put a offset into a texture.
4: The GPU then rasterises the polygons using a linear interpolate of the co-ordinates given for each vertex of the polygon.
Something like Unreal or Quake will use concave polygons and clip them prior to rendering, this reduces the work required in 4: mipmaps will also be generated to allow for better scaling this increases the memory requirements, multi-texturing will be used to increase the apparent detail level of the textures, and for things like deciles alpha-blending will be uses to generate effects such as light from a blast from a weapon, and cubic or trilinear interpolating will be used instead of linear to create shading effects on the textures.
Now, for more modern games you can use 'vertex shaders', they allow programmable effects to be applied along a polygon edge, an can crate effects such as underwater.
Then there are pixel shaders, they are similar to vertex shaders but are calculated per-pixel, they allow things like toon-shading, bump mapping, per-pixel lighting, true phong, and procedural textures, the GPU will all sever on card only 'virtual' textures to be used as a scratch pad for per-pixel shading.
Got that?
Now, how do windowing systems work.
1: Each windows is given a square clipping region on the screen, the window boundaries. Each window has a pixel map
2: The application updates it's pixel map, and notifies the windowing system.
3: the pixel maps are then composed and displayed.
Now, because giving each window it's own pixel map requires a lot of memory most windowing system cheat and use a shared pixel map and clip based on the windows clipping region, this requires the windowing system to inform a 'hidden' window to redraw it's self when a windows above it moves.
Each window is build up from a number of elements, icons (pixel maps), fonts (often vectors) and widgets (line drawn or pixel maps).
On today's systems all the elements are rendered by the application and presented as pixel maps. Fonts are turned into pixel maps and cached for speed, because processing vectors is hard.
On tomorrow's system the fonts will be sent to the graphics card, the widgets will be rendered using vectors on the graphics card.
Yes, it's easy for today's graphics card to deal with our current systems, which are low quality and optimised for the current windowing systems rendered on the CPU, but Apple, Microsoft, and Gnome are already moving away from separating the application functionality from the visual display allowing more and more work to be performed on the GPU instead of the CPU, requiring a GPU with more memory capable of manipulating the windows without having to hog the system bus, CPU and main memory.
'You're making the totally unreasonable assumption that the video card would keep everything in its own memory, all the time.' No I'm not.
'UIs are fantastically simple compared to games. They rarely contain more than a few thousand screen elements, only a small fraction of which are active at any time.' No there not.
'Have you noticed that XWindows (as it stands now) can easily handle dozens of windows without depleting main memory'
Yes, and it's slow as a dog, what's the point of using the GPU if's your going to have a slow windowing system. What was it they say about history and repeating it?
Try turning back buffers on in X (if you driver supports it which I doubt it does) man startx. then see how much memory it uses.
I should have known better than to be concise in my post, but it looks like I'm going to have to explain how on GPU windowing is different to the comparatively crappy system X uses.
1: Really what is the point of saying 'the graphics card can handle the windowing system' when your still hogging the system bus and system memory. You may as well stick with X.
For each visible fully visible window you need a pixel map, so when showing those windows in 3d, or scaled down you potentially need a full screen pixel map,the MAC cheats and uses small pixel maps for it's scaling.
2: Unless we do all the clipping calculations on the CPU (see what's the point in 1) then were going to have to store unclipped regions on the GPU. You may not think you notice the performance lag, but I didn't notice the lack of colours on a black and white TV, you will notice it once you've seen it rendered at 50fsp.
3: . UI elements rarely change, most of them aren't visible, most of them aren't animated, Yes, I like your thinking, lets use the same crappy UI designs that we have today, build in a world when the GPU could manage the windowing system.
Look at Internet pages, they are becoming more animated especially with flash web sites, stop holding the UI in your slow, paged out, poorly clipped world and get into the 22nd century.
UI's are simple from a 3D standpoint (but then so is Quake2), but UI's aren't all about 3D, lets play what if!
What if the GPU could render all you fonts for you, like that idea?
What if the GPU rendered all you Widgets for you, with full per-pixel lighting, and tactile effects, like that Idea?
What if a myriad of 3D effects could be applied to the windows, much more than project looking glass.
Go on, render my per-pixel lit purple fury GUI, where the fur moves as the mouse moves over it on your Quake2 rendering graphics card.
You could also try the complex GUI where the graphical components of the application are rendered on with the GPU, my application is Doom3, make my GUI work on your Quake2 graphics card.
Why should I even bother porting a crap windowing system to the GPU? Gnome are moving to a vector based windowing system, so is Microsoft, so are Apple, just wait until tomorrows GPUs arrive with truetype and SVG rendering on chip.
Like I said 'UI's aren't about how many polygons you can blast out, but how well you can handle vectors/fonts and buffers.' and 'UI's are not games, in some ways (vectors) they are far more complex.', but maybe it was a little too concise for you to IO.
Slashdot Mirror
Well, If I throw a ball to you your hands will hopefully be in the right place to catch the ball before it arrives.
That's because Newtons laws of gravity and motion are more-or-less perfect for ball catching. Where greater times, distances and speeds are involved Einstein's theory of relativity becomes more useful.
Scientists hedge their bets on those laws and previous observations of stars and galaxies being good enough to estimate where this are now and where they were.
This could all be in my head and everything's just a dream, I'll hedge my bets it isn't.
pretty big 'maintance' patch.
Or they could keep on with the old numbering system dev could work on 2.7 and backport to 2.6 once it's stable enough.
Maybe Linus doesn't have any idea of what to put into a 2.8 kernel so branching to 2.7 would be pointless.
Filing system devices would be nice so that soft links appear for device-driver bindings and you can install a new driver using cp or ln, and it would be nice if the kernel was split into drivers and kernel to giver comercial operatiors half a chance of making a binary driver that will install for at least a couple of kernel releases.
Yeh, the difference between 2.6.8 and 2.6.10 iss only a partial re-write of a driver when the memory allocations systems change in the kernel not a HUGE-scale re-write that a change from 2.4 to 2.6 would require.
So potentially if I'm providing an offsite backup service without any trademarks then the distribution part is OK, it's just the copyright part I have to worry about.
more importantly,
you are not allowed to distribute a product without the copyright holders permission. Even if it is a fully paid up product.
e.g. Tesco (kinda wanabe wallmart) imported some Levis jeans direct from the producer and was forced to stop selling them.
Ok, here's some analysis.
Count words, I just happen to have Michel Abrash's graphics programming black book on the table and it's got a good discussion on count words.
1: The data is static, so it fits you theory.
2: The C version is almost optimal
3: The java version is not comparable to the C version for real world tasks (I haven't bothered to re-write it to be the same as the c version to see if it's performance is better or not).
HeapSort
why not quick sort? don't most languages come with quick sort built in and heap sort is almost identical to quick sort (variants on the binary merge).
Object methods...
1: the C 'object' and the java object are not the same, the java object provides a lot of extra functionality that may well be used in a real world application that has not been put in-place in the c version.
2: He didn't even use final in the Java version, how is using an almost static C 'object' equivalent to using a non-final java object in real world tasks.
I could go on.
That's the theory. In practice, this has not yet been achieved outside very tight constraints, such as where the data does not change very often.
Which accounts to 95% of my hard drive, so tight, but common.
Did you even READ the fucking site
No I just plucked It does say that start up time is included too, so that taints the benchmarks. out of thin air.
No longer the case. If you actually bother to look at the fucking benchmark
So their fucking benchmarks now, there were nice benchmarks that showed you point earlier.
I read the FAQ, if it's wrong what does that tell me I should assume about the fucking benchmarks?
a profiling Just in time JVM can profile the code while it is running and then next time around recompile the code to be faster.
That site doesn't list how the tests were carried out except to say that there were, no details on any profiling or optimization attempts that any serious deployment would use.
It does say that start up time is included too, so that taints the benchmarks.
oh, and he didn't even try jrockit
here's a link, no pickes though. (I've got some nice ones of people being hit by the police and blood going everywhere!).
http://tash.gn.apc.org/bristol1.htm
Despite this, the police turned up in force and told us we wouldn't be having a party. When questioned on what laws or powers they were using to shut us down (much to their dismay, the criminal Justice Act and the licensing laws do not cover free warehouse party's). The policeman in charge replied: "What are you, a fucking Barrister now...? You're not having a party because I SAY SO..!"
Numerous policemen then lined up in an attempt at preventing people getting into the warehouse. But when it became apparent that we weren't going to be intimidated into not having a party, they waded in and began indiscriminately beating party people with their truncheons. As a result of this action, alot of people received injuries, some of them quite serious. One woman was on the floor being beaten by at least 3 or 4 policemen and one man was held down over a railway line by two policemen who repeatedly smashed his face into a sleeper until his nose caved in.
That's agrivated tresspass not tresspass.
I've picked up litter infront of enough CS cas firing riot poilce to know the difference.
We've had hundreds of police, helecopters, dogs, CS gas and horses.
join the free parties, it's a riot every weekend.
' and easier to maintain code than today's so called high level languages.'
.NET
Java can be faster if you use a profiling JVM, (well unless you profile your C code and get gentoo to re-emerge with using the profile every other week)
this should also apply to
Java is also much easier to maintain that C, because it's almost exactly the same as C with enough bits missing to let it run in a Virtual Machine and some extra rules inplace.
Of couse bad C is probably easier to maintain that bad Java and Bad C++.
In which case you would have to start argueing.
Use OSS because all software is bad, it's just that our version of bad is easier to maintain....
your using the 'anti' claus, your not saying were better because were the best your saying.
We use C isn't better because they try to sell a new compiler (who SUN and GCC?).
Not we use C it's better because it's simpler and more compact that many of todays languages that have lots of bolted on features that havent matured yet.
Remember, positive energy activates constant elevation. Be positive about OSS not negative about everything else.
obviously you know nothing about the British government, all there laws seem to stem from random debate.
Personally I think if they should be taxing anyone it should be the companies that use our personal information. 10% of the transaction cost as an extra tax should bring in a nice revenue and make companies think 1.5 times before using your personal information.
how else can a post man deliver mail to your door if he has to trespass to get there?
trespass is a civil offence not a criminal one.
it may affect any information gained by trespass being admissible in a court of law though.
IANAL, but I listen and learn.
It only counts if the property doesn't belong to you.
breaking and entering
'housebreaking: trespassing for an unlawful purpose; illegal entrance into premises with criminal intent '
'The gaining of unauthorized, illegal access to another's premises, as by forcing a lock.'
'The act of gaining passage into and entering another's property (as a building or vehicle) without privilege or by force; also The crime of breaking and entering see also burglary'
'Breaking and entering is defined as the crime of illegally entering a residence or other enclosed property using any amount of force (even pushing open an unlocked door). If criminal intent can be established, breaking and entering can be considered a burglary, a felony in most U.S. jurisdictions. If there is no criminal intent, breaking and entering can be considered trespassing, which is usually a misdemeanor crime.'
All they had to do was merge the 'search' and the 'search dns' inputs and it would be great.
as it is there are two inputs, one called 'search' that umm searches, and one called go that umm... searches in a different way.
One 50Char box that searches unless you type in a domain name, or news:// etc.. or two 20 char boxes, the choice isn't yours.
Take a look
Wow, just notices the extra shopping, searches in a different way box, good thing it's bigger than the other two on a consumer product.
Well, I don't have any viruses, so it can't be too bad.
'This is one of those things that people keep repeating as if that makes it true. X isn't slow,'
Did I say X? no.
anyhow for the sake of your argument windows slow, X slower why do you think they wrote the xDamage and xComposite extensions?
'If you start adding tons of animation and complex effects to the GUI, it's going to start requiring scads of memory and processing time even if it's all done in 2D.'
Yes, Mr I want a chroma colour green screen monitor for xmas.
' You think vector graphics are fast on a CPU? You think font glyphs don't take up space in main memory now?'
What are you going on about, I think that vectors should be real-time, not static pre-rendered pixelmaps.
'Lets see your PIII try to render it in 2D on the CPU. I'll bet my six year old Geforce2 MX handles it a lot better. And while your CPU is choking under the burden, my GPU will be happily using its antialiasing hardware to make the whole thing look great, text and all.'
my PIII has a green screen and it runs 1990's software. If you want to be stuck in the 90's with you FSAA then feel free, the rest of us can move with the times. I can have all the effects in the world and still run at 50fps with a low CPU overhead while your left with 6-8 updates a second taking 5% of your CPU. Hmm.. I wonder what they'll be showing off at expo this year.
IANAL, so if you sue me, I'll get professional help.
I claim:
1. A virus detection system for detecting if a computer file is infected by a virus, the file having a plurality of potential virus entry points, the system comprising:
A virus checking system, to check a file that could have a virus embedded in it at any location. I've interpreted plurality as any, and not one of many
an engine for controlling operation of the virus detection system responsive to instructions stored in an intermediate language, the instructions adapted to examine the plurality of potential virus entry points and post for emulating ones of the plurality of potential virus entry points exhibiting characteristics indicating a possible virus;
An automated process for sending in one go a sequence of commands to the virus checking system, that then (repeat) goes on to check the 'plurality' of entry points.(/repeat), where A virus may be embeded.
a scanning module coupled to the engine for scanning regions of the virtual memory for a signature of the virus responsive to the engine and the emulating module, wherein presence of the virus signature in a scanned region indicates that the file is infected by the virus.
something that check 'virtual' memory for a 'virus' when told to by the above automated process (who knows what the emulating module is?), when it find a virus if somehow magically knows what 'the file' is supposed to mean indicates [to the user?] that the file is infected.
Ok, virtual memory is 434 and the file is, 100.
But I couldn't find the definitions anywhere.
Looks like a poorly though out virus checker that can take multiple commands in one instruction from a separate process that knows where the virus may be embedded, that also checks virtual memory for the virus via a 'module'
It has been an obvious way to check for viruses ever since I knew about them (well before virus checker were around).
It just wasn't a practical way at the time.
It was also not shown that the Government did not know that Iraq could not launch WMS in 45-minutes.
So, should the BBC be 'reprimanded' for thinking the Government wasn't stupid?
Would you prefer a stupid government or one that lied?
Lets say I create a few seperate systems, one scans a stream of data for a match say a header on a tape.
I'll call this grep --file
another system finds blocks of data 'signatures' that I want to match.
I'll call this grep -o
A third lists all the files on my filing system.
I'll call this file.
So, symantic has patented a system where grep --file , grep -o and file cannot be used together if your looking for virus signatures.
(maybe with a little winedump and gzip thrown in for good measure).
There's something called 'Abstraction,
Filtration, Comparison',, applied in this case you would end up with nothing.
Yes. I don't think you have any idea how 3D rendering works, or if you do it's completely screwed up.
I can write windowing systems, WYSYWIG editors, and 3D graphics engines that will run on a 268 with a few MB of ram.
If you want to know how to write a WYSYWIG editor in a few Mb than I'll tell you, I'll also tell you why as soon as you get a little more complex you use tens of MB and even more complex and you start hitting hundreds.
Here's a simple walk through or a rendering system.
1: all your textures are pixel maps,
pixel MyTexture[usize][vsize].
(not no mip-mapping, volume textures or cube, just simple textures)
2: You then create a number polygons to render on screen, this is for something simple like Unreal Tournament or X,
3: at the vertex of each polygon you put a offset into a texture.
4: The GPU then rasterises the polygons using a linear interpolate of the co-ordinates given for each vertex of the polygon.
Something like Unreal or Quake will use concave polygons and clip them prior to rendering, this reduces the work required in 4: mipmaps will also be generated to allow for better scaling this increases the memory requirements, multi-texturing will be used to increase the apparent detail level of the textures, and for things like deciles alpha-blending will be uses to generate effects such as light from a blast from a weapon, and cubic or trilinear interpolating will be used instead of linear to create shading effects on the textures.
Now, for more modern games you can use 'vertex shaders', they allow programmable effects to be applied along a polygon edge, an can crate effects such as underwater.
Then there are pixel shaders, they are similar to vertex shaders but are calculated per-pixel, they allow things like toon-shading, bump mapping, per-pixel lighting, true phong, and procedural textures, the GPU will all sever on card only 'virtual' textures to be used as a scratch pad for per-pixel shading.
Got that?
Now, how do windowing systems work.
1: Each windows is given a square clipping region on the screen, the window boundaries.
Each window has a pixel map
2: The application updates it's pixel map, and notifies the windowing system.
3: the pixel maps are then composed and displayed.
Now, because giving each window it's own pixel map requires a lot of memory most windowing system cheat and use a shared pixel map and clip based on the windows clipping region, this requires the windowing system to inform a 'hidden' window to redraw it's self when a windows above it moves.
Each window is build up from a number of elements, icons (pixel maps), fonts (often vectors) and widgets (line drawn or pixel maps).
On today's systems all the elements are rendered by the application and presented as pixel maps. Fonts are turned into pixel maps and cached for speed, because processing vectors is hard.
On tomorrow's system the fonts will be sent to the graphics card, the widgets will be rendered using vectors on the graphics card.
Yes, it's easy for today's graphics card to deal with our current systems, which are low quality and optimised for the current windowing systems rendered on the CPU, but Apple, Microsoft, and Gnome are already moving away from separating the application functionality from the visual display allowing more and more work to be performed on the GPU instead of the CPU, requiring a GPU with more memory capable of manipulating the windows without having to hog the system bus, CPU and main memory.
'You're making the totally unreasonable assumption that the video card would keep everything in its own memory, all the time.'
No I'm not.
'UIs are fantastically simple compared to games. They rarely contain more than a few thousand screen elements, only a small fraction of which are active at any time.'
No there not.
'Have you noticed that XWindows (as it stands now) can easily handle dozens of windows without depleting main memory'
Yes, and it's slow as a dog, what's the point of using the GPU if's your going to have a slow windowing system. What was it they say about history and repeating it?
Try turning back buffers on in X (if you driver supports it which I doubt it does) man startx.
then see how much memory it uses.
I should have known better than to be concise in my post, but it looks like I'm going to have to explain how on GPU windowing is different to the comparatively crappy system X uses.
1: Really what is the point of saying 'the graphics card can handle the windowing system' when your still hogging the system bus and system memory. You may as well stick with X.
For each visible fully visible window you need a pixel map, so when showing those windows in 3d, or scaled down you potentially need a full screen pixel map,the MAC cheats and uses small pixel maps for it's scaling.
2: Unless we do all the clipping calculations on the CPU (see what's the point in 1) then were going to have to store unclipped regions on the GPU. You may not think you notice the performance lag, but I didn't notice the lack of colours on a black and white TV, you will notice it once you've seen it rendered at 50fsp.
3: . UI elements rarely change, most of them aren't visible, most of them aren't animated, Yes, I like your thinking, lets use the same crappy UI designs that we have today, build in a world when the GPU could manage the windowing system.
Look at Internet pages, they are becoming more animated especially with flash web sites, stop holding the UI in your slow, paged out, poorly clipped world and get into the 22nd century.
UI's are simple from a 3D standpoint (but then so is Quake2), but UI's aren't all about 3D, lets play what if!
What if the GPU could render all you fonts for you, like that idea?
What if the GPU rendered all you Widgets for you, with full per-pixel lighting, and tactile effects, like that Idea?
What if a myriad of 3D effects could be applied to the windows, much more than project looking glass.
Go on, render my per-pixel lit purple fury GUI, where the fur moves as the mouse moves over it on your Quake2 rendering graphics card.
You could also try the complex GUI where the graphical components of the application are rendered on with the GPU, my application is Doom3, make my GUI work on your Quake2 graphics card.
Why should I even bother porting a crap windowing system to the GPU? Gnome are moving to a vector based windowing system, so is Microsoft, so are Apple, just wait until tomorrows GPUs arrive with truetype and SVG rendering on chip.
Like I said 'UI's aren't about how many polygons you can blast out, but how well you can handle vectors/fonts and buffers.' and 'UI's are not games, in some ways (vectors) they are far more complex.', but maybe it was a little too concise for you to IO.