Actually, I think a lot of people are mad *because* we intervene too much. We do political jiggering in the Middle East (keeping non-Democratic governments in power in the process) just so we can have oil. We're fucking up some foreign countries real good, and in 100 years, the oil will be gone anyway, and they'll still have civil chaos because of the stuff we're doing now. At that point, we're all (us and them) screwed.
FEWER not LESS. But kword is hardly a toy. I couldn't care less about Word documents. Nobody I know is stupid enough to send me Word documents. Aside from that, KWord is actually reasonably powerful, a good 80% product (ie. 80% of users find it powerful enough). It might not be right for people tied to closed content, but that doesn't make it a toy.
America invented the internet. >>>>>>>>> China invented fireworks. No fireworks for you! Bye bye fourth of July. Phoenicians invented the "English" alphabet, so you best stop writing! Arabs invented Algebra and the "English" system of numerals is Indian in origin. There goes math! In fact, 0 is a concept that originated in India, so you'll have to find another value to denote your IQ.
America uses the internet the most. During the late 90s, Internet traffic in North America more than doubled every six months. >>>> Europeans use cell phones the most, so I guess we should all adopt GSM. The Chinese eat rice the most (I'm from India, another rice-eating nation, so this isn't a racist comment:) so if you want to eat rice, you have to do it chopsticks!
Don't even get me started on the last one. World history is my little hobby, I'd have to intellectually beat the crap out of you...
However if the drive fails, you're toast. >>>>>>>> Heh heh, all my data is stored 650 miles away in a nice safe server. I've been in my dorm room for 2 months now, and I've already got two spare HDDs sitting in my desk drawer. I figure that if my main HD dies, I'll be up and running again, with all my data, within an hour.
laptop HDs are still 4800 rpm. They don't put out much heat at all. What really puts out heat are those mobile P4s. I can feel mine through my 3/4" wooden desk. You *definately* don't want to use P4 laptops on your lap, not if you ever want to have children anyway. That'd be a funny statistic to know. Are P4 laptop owners less fertile than the population as a whole? What is it, 50% drop in sperm count for each 10 degrees over normal temp?
The difference is that 2.5 GB of RedHat has tons more tools than the 1.5 GB of Windows XP. Give me 1.5 GB, I can fit the base install, KDE, KOffice, and a bunch of utilities, along with a full development environment.
Um, I don't really know what kind of people you know, but I know lot's of people who push near 100 GB. My 30 GB laptop drive was barely enough for Windows XP and a few games, and I'm not even much of a gamer. In Linux, my base system (bare-bones GNOME install, full KDE install, including KOffice) takes up 4 GB. Add in a relatively small number of MP3s (6GB) and some videos, etc, and my 30 GB drive is getting cramped, and I barely do anything! I've got one friend who's nearly filled his 100 GB HD with free bluegrass music! Imagine how much HD space a heavy gamer/MP3er uses!
Considering that the memory in the powerbook is SDR RAM, vs the RAM on the Inspiron which is DDR RAM, the Apple memory is quite expensive. Currently, DDR RAM is anywhere from 2x (for 128MB DIMMS) to 4x (for 512MB DIMMS) as expensive as SDR RAM.
I'll testify for CONS. I used it for an OS kernel at one point. OS kernels necessarily have totally obscure build systems. One issue in particular was running a second multi-file optimization pass with Intel C++. I was able to do configuration and compilation in less than 500 lines of Perl. I actually got carried away with adding features (like doing line-counts for files) because it was just so easy!
Many ways better than MSVC? With regards to standards complience gcc blows MSVC away.
1) MSVC doesn't even support partial template specialization! GCC is the only compiler to pass all the Boost library compiler tests, and is the reccomened compiler for Loki.
2) GCC doesn't have nearly as many "magic" keywords as MSVC (you have no idea how many people think that some of thos MSVC stuff is standard C) and the stuff that it does have is being depricated. That alone makes it better:)
Yeah, but they didn't have the technology to prove or disprove it. Stuff is only possible until it is within your technological ability to ascertain things one way or the other.
The second post you posted was far less flameworthy than the first:) As for me not maintaining code, that's correct. I also don't edit other people's writing. I guess I'd be less happy with English if I did. I hope you realize that neither example suggests anything more than the majority of people cannot handle complexity and power, which was something we knew anyway. And neither example makes any comment on either C++ or English other than point out they are complex and powerful, which we also knew anyway.
As for using multiple languages, I agree. So far I've got Perl and Python down pretty well. I know Java, but refuse to use it, and get hives from languages like Lisp. Even then, I must admit that both have their uses. But C++ is still my language of choice, because in all honesty, a jack of all trades is a master of none. It's not just about using the right tool for the job, but a combination of the right tool and the skill with that tool. For most of the stuff I do, C++ is the right tool, so even for cases where C++ is just a mediocre choice in terms of tools, I tend to use it anyway because my experience with it outweighs any benifets another language might have. As for using C as a low-level language, I'd hold that C++ can be used as just a better C. I'm doing an OS kernel in C++ (as a hobby project) and I've found that templates in particular are very helpful (OS kernels are highly datastructure intensive), virtual functions and abstract base classes are perfect for modeling things like driver APIs, and exceptions are a nice way to keep error handling code out of "hot" code paths. If you take a look at real kernel code, you'll see that there are a lot of C++-isms. In the driver API example I gave above, often kernels use the exact same code a C++ compiler would generate for an abstract base class (table of function pointers). I've also used C++ for higher level tasks like scientific simulation. Using C++ and the STL allowed me to write the simulation extremely quickly (because I didn't have to worry about memory allocation and whatnot) while still getting the performance of low-level code.
Re:Satisfies "News for Nerds" OR "Stuff that matte
on
Design Patterns
·
· Score: 2
Actually, the way it is written, it clearly means that "News for Nerds" is "Stuff that matters."
The first 5 arguements basically boil down to this: You propose that C++ is less clean because of certain problematic features it has. My reasoning is that I couldn't care less how "clean" C++ is. C++ has all those features because it allows you to write code that uses the most elegant constructs available at a given time. The language itself is more powerful for that, just as natural languages (if I wasn't an engineering major, I'd be an English major:) are more powerful because of all the different mechanisms they include. It's sad that people abuse the power of C++ (just as people abuse English) but that's no reason to castrate the language, and (in contrast to the ideas of the original poster) no reason to declare C++ dead. In my opinion, the code you can write with a language is the most important thing, overall feature-set aside. C++ allows you (assuming the programmer isn't the bottleneck) to write clean code for a wide variety of situations. Other language (Java in particular) forces you into a dogma (OO in that case) that forces you to write sub-optimal code in many situations. The Java container classes, again, are a good example. The STL is clean, natural, and orthogonal (and type safe!). The Java container classes feel like they tried to shoe-horn a non-OO concept into an OO implementation.
Please don't use Java as a counter-example of useful high-level languages. Python, Perl, Scheme, CL and their ilk are where I'd go to compare. >>>>>>>>> I don't really consider Python, Perl, and Scheme to be in the same league as C++. First, C++ is by all means a lower level language, and more appropriate for systems programming tasks. Beyond that, C++ is a far more generic language. If you're working on a very specific task, Python, Perl, and Scheme might very well be extremely useful and productive. But as a base-level, widespread language, C++ fits much better. In other words, it's important to use the best tool for the job, but if I could only know one language, it would be C++.
7) If you take a look at the equivilent data structures in C, you'll realize how much better the STL is. C is a hard-core low-level language. It's the ultimate roll-your-own and about as low-level as you can ever get and still remain portable (that was, after all the goal). Libraries like glib (which have many of the things you're used to in STL) >>>>>>>>> glib has nothing on the STL. The original poster suggested ditching C++ and going back to C. I used the STL as an example of something where a C++ and C library try to do the exact same task, but the C++ library accomplishes the task (thanks to templates) in a much cleaner and more elegant way.
I think you took the arguement slightly out of context. I wasn't claiming that C++ is the be all end-all of languages. I was responding to a claim that C++ was a "bad" language, and that we should ditch it and go back to C. My claim is that C++ is a damn good language in its own right, and its complexity is justified by its expressiveness.
The kernel exports certain symbols that are available for driver modules to link to. These symbols are classified by license. Many symbols a driver can link to without having to be a GPL'ed driver. These symbols, presumably, are written by people who grant permission to link against non-GPLed code. Other symbols (for example the new work-queues stuff) are exported as so-called GPL_ONLY symbols, and linking against those in a non-GPL module is illegal. Thus, if you avoid using the GPL_ONLY symbols you can probably get away with a binary only module. The fact that this mechanism exists implies permission, so this aspect would be important to bring up to the lawyer.
Err, nowhere close. I'm running Gentoo custom compiled for my P4, with X reniced to -11, kdeinit (and thus all KDE applications) reniced to -10 (helps significantly to redraw times). I've also removed all unnecessary extensions from my XF86Config, and am using the lastest NVIDIA drivers. Now, KDE does better when the system is under load (like doing a background compile) where Win2k just breaks up, but for unloaded "crispness" Win2K is just sharper. XP, however, is another story, and is much slower than 2K. Even then, some things (like resizing Internet Explorer windows vs resizing Konqueror windows) are less crisp.
Bullshit. Both of them result in a call to graphic driver which , generally, is better optimized on Windows. >>>>>>>>> First, the graphics driver doesn't do all that much these days. For example, with Xft2 and sub pixel anti-aliasing, the graphics driver doesn't even handle blitting of text anymore. Second, it depends on the drivers. The NVIDIA drivers I use, for example, are just as good as the Windows version. That said, if it's a driver problem, it's not X's fault, is it?
They use it the way it was intended. Every fucking Qt call of say QPainter maps almost directly to Xlib function. What else would you have them to do ? >>>>>>>> Do you really believe that? Qt and GTK+ use X as basically a blit engine. They do all the drawing in pixmaps (in software) then blit the results to the screen via the X server. That's most definately *not* how X was designed to be used. There are also numerous issues about not using the protocol properly. Thus, I refer you to the recent thread on the X Render mailing list about XFree86's performance.
You'll take away C++ when you can pry it out of my cold, dead hands. I tend to be a mindless C++ advocate, so I disagree heartily with everything you say. But as for your complaints, I think Stroustrup heard them all before designing things into the language.
1) C++ isn't a high-level language, and it isn't a low-level language. It's both. And that's why it kicks ass.
2) Operator overloading is terribly useful. The whole idea of C++ is that you can implement user-defined types that act like built-in ones. Operator overloading is meant for mathematics-style user defined types (for scientific programs). If you're using operator overloading in a place where it doesn't make sense, then that's your own fault. I hate the Java way of doing things, where everything that could potentially hurt you is illegal. It's the programming language equivilant of Ralph Nader. And > are not just bitshifting operators. They serve double duty as streams operators. I have yet to see anybody ever confuse the two, because it's obvious in most cases that you're not bitshifting the integer 'cout' by "Hello World" or something of that sort. And like printf(), in all it's type unsafe, variable parameter glory is any better!
3) Multiple inheritence can also be useful. Mixin classes, for example, are real nice. Again, if you abuse it, it's your fault. That holds for drug abuse, why not language abuse?
4) There are four casting operators for a reason. You're not supposed to cast in C++. The casting operators are long and ugly and verbose to make the point that you're not supposed to do that. But if you do it, at least the four different casting operators make it clear exactly what you're trying to do.
And templates are mana from heaven. Templates give you the efficiency (yeah, I said efficiency, take that, Java-heads!) of coding custom data structures for each type, while letting the compiler do the actual dirty work. For data structures, the STL beats the Java container classes to a pulp. If you take a look at the equivilent data structures in C, you'll realize how much better the STL is.
You've obviously never used BeOS. I'm running KDE 3.1 on kernel 2.5.44 on a 2 Ghz P4 with 640MB of RAM and a 64MB GeForce4. I love it dearly, but BeOS on my antique P2-300 was a good deal faster, and Win2K on my Duron 750 is still faster.
X is actually faster than the GDI for most things. Don't believe me? Benchmark it yourself? I've benchmarked it before, and for stuff like line drawing or bitmap blitting, X stands up well even to DirectX. The main problem is that the toolkits (ahem, Qt) don't really use X all that well. That said, a lot of problem is due to the fact that a properly fast KDE/GNOME desktop needs a lot of proper configuration. My KDE 3.1 desktop, for example, is as fast as WinXP for most things (expect app startup speed, but glibc 2.3 and prelinking should fix that) but it required some custom compiling and renice tricks to get it that way.
It's not really a problem with either C++ or Qt, it's a problem with the compilers. In particular, with how compilers implement templates. For example, let's say you have the files one.cpp, two.cpp, and three.cpp. If you use std::string for each one, then gcc will compile the string class in every file that uses it. So one.o, two.o, and three.o will all have the parts of the string class that they use (ie, most of it). The linker then discards the redundant bits when the object files are linked together. Now for heavily used template classes (the STL, for example) you basically have gcc compile large parts of a template class library for each source file, then have the linker throw them away at link time. Now the alternative is already part of the C++ '98 standard. You have this thing called the export keyword. Unfortunately, none of the big compilers (g++, Intel C++, VisualC++) support it yet.
Slashdot Mirror
Actually, I think a lot of people are mad *because* we intervene too much. We do political jiggering in the Middle East (keeping non-Democratic governments in power in the process) just so we can have oil. We're fucking up some foreign countries real good, and in 100 years, the oil will be gone anyway, and they'll still have civil chaos because of the stuff we're doing now. At that point, we're all (us and them) screwed.
FEWER not LESS. But kword is hardly a toy. I couldn't care less about Word documents. Nobody I know is stupid enough to send me Word documents. Aside from that, KWord is actually reasonably powerful, a good 80% product (ie. 80% of users find it powerful enough). It might not be right for people tied to closed content, but that doesn't make it a toy.
First ASCII, now this...
:) so if you want to eat rice, you have to do it chopsticks!
America invented the internet.
>>>>>>>>>
China invented fireworks. No fireworks for you! Bye bye fourth of July. Phoenicians invented the "English" alphabet, so you best stop writing! Arabs invented Algebra and the "English" system of numerals is Indian in origin. There goes math! In fact, 0 is a concept that originated in India, so you'll have to find another value to denote your IQ.
America uses the internet the most. During the late 90s, Internet traffic in North America more than doubled every six months.
>>>>
Europeans use cell phones the most, so I guess we should all adopt GSM. The Chinese eat rice the most (I'm from India, another rice-eating nation, so this isn't a racist comment
Don't even get me started on the last one. World history is my little hobby, I'd have to intellectually beat the crap out of you...
And not having to power that DVD drive saves enough juice that you can actually watch that full-length movie!
However if the drive fails, you're toast.
>>>>>>>>
Heh heh, all my data is stored 650 miles away in a nice safe server. I've been in my dorm room for 2 months now, and I've already got two spare HDDs sitting in my desk drawer. I figure that if my main HD dies, I'll be up and running again, with all my data, within an hour.
laptop HDs are still 4800 rpm. They don't put out much heat at all. What really puts out heat are those mobile P4s. I can feel mine through my 3/4" wooden desk. You *definately* don't want to use P4 laptops on your lap, not if you ever want to have children anyway. That'd be a funny statistic to know. Are P4 laptop owners less fertile than the population as a whole? What is it, 50% drop in sperm count for each 10 degrees over normal temp?
The difference is that 2.5 GB of RedHat has tons more tools than the 1.5 GB of Windows XP. Give me 1.5 GB, I can fit the base install, KDE, KOffice, and a bunch of utilities, along with a full development environment.
Even the multiple operation advantage is diminishing thanks to Tagged Command Queueing, which is implemented in modern ATA drives.
Um, I don't really know what kind of people you know, but I know lot's of people who push near 100 GB. My 30 GB laptop drive was barely enough for Windows XP and a few games, and I'm not even much of a gamer. In Linux, my base system (bare-bones GNOME install, full KDE install, including KOffice) takes up 4 GB. Add in a relatively small number of MP3s (6GB) and some videos, etc, and my 30 GB drive is getting cramped, and I barely do anything! I've got one friend who's nearly filled his 100 GB HD with free bluegrass music! Imagine how much HD space a heavy gamer/MP3er uses!
Considering that the memory in the powerbook is SDR RAM, vs the RAM on the Inspiron which is DDR RAM, the Apple memory is quite expensive. Currently, DDR RAM is anywhere from 2x (for 128MB DIMMS) to 4x (for 512MB DIMMS) as expensive as SDR RAM.
I'll testify for CONS. I used it for an OS kernel at one point. OS kernels necessarily have totally obscure build systems. One issue in particular was running a second multi-file optimization pass with Intel C++. I was able to do configuration and compilation in less than 500 lines of Perl. I actually got carried away with adding features (like doing line-counts for files) because it was just so easy!
glibc is actually notoriously hard to port. Currently, only Linux and Hurd are officially supported, along with Windows support from Cygwin.
Many ways better than MSVC? With regards to standards complience gcc blows MSVC away.
:)
1) MSVC doesn't even support partial template specialization! GCC is the only compiler to pass all the Boost library compiler tests, and is the reccomened compiler for Loki.
2) GCC doesn't have nearly as many "magic" keywords as MSVC (you have no idea how many people think that some of thos MSVC stuff is standard C) and the stuff that it does have is being depricated. That alone makes it better
I don't have modpoints, but if I did, this guy would get it! Somebody, mod this guy up!
Yeah, but they didn't have the technology to prove or disprove it. Stuff is only possible until it is within your technological ability to ascertain things one way or the other.
The second post you posted was far less flameworthy than the first :) As for me not maintaining code, that's correct. I also don't edit other people's writing. I guess I'd be less happy with English if I did. I hope you realize that neither example suggests anything more than the majority of people cannot handle complexity and power, which was something we knew anyway. And neither example makes any comment on either C++ or English other than point out they are complex and powerful, which we also knew anyway.
As for using multiple languages, I agree. So far I've got Perl and Python down pretty well. I know Java, but refuse to use it, and get hives from languages like Lisp. Even then, I must admit that both have their uses. But C++ is still my language of choice, because in all honesty, a jack of all trades is a master of none. It's not just about using the right tool for the job, but a combination of the right tool and the skill with that tool. For most of the stuff I do, C++ is the right tool, so even for cases where C++ is just a mediocre choice in terms of tools, I tend to use it anyway because my experience with it outweighs any benifets another language might have. As for using C as a low-level language, I'd hold that C++ can be used as just a better C. I'm doing an OS kernel in C++ (as a hobby project) and I've found that templates in particular are very helpful (OS kernels are highly datastructure intensive), virtual functions and abstract base classes are perfect for modeling things like driver APIs, and exceptions are a nice way to keep error handling code out of "hot" code paths. If you take a look at real kernel code, you'll see that there are a lot of C++-isms. In the driver API example I gave above, often kernels use the exact same code a C++ compiler would generate for an abstract base class (table of function pointers). I've also used C++ for higher level tasks like scientific simulation. Using C++ and the STL allowed me to write the simulation extremely quickly (because I didn't have to worry about memory allocation and whatnot) while still getting the performance of low-level code.
Actually, the way it is written, it clearly means that "News for Nerds" is "Stuff that matters."
The first 5 arguements basically boil down to this: You propose that C++ is less clean because of certain problematic features it has. My reasoning is that I couldn't care less how "clean" C++ is. C++ has all those features because it allows you to write code that uses the most elegant constructs available at a given time. The language itself is more powerful for that, just as natural languages (if I wasn't an engineering major, I'd be an English major :) are more powerful because of all the different mechanisms they include. It's sad that people abuse the power of C++ (just as people abuse English) but that's no reason to castrate the language, and (in contrast to the ideas of the original poster) no reason to declare C++ dead. In my opinion, the code you can write with a language is the most important thing, overall feature-set aside. C++ allows you (assuming the programmer isn't the bottleneck) to write clean code for a wide variety of situations. Other language (Java in particular) forces you into a dogma (OO in that case) that forces you to write sub-optimal code in many situations. The Java container classes, again, are a good example. The STL is clean, natural, and orthogonal (and type safe!). The Java container classes feel like they tried to shoe-horn a non-OO concept into an OO implementation.
Please don't use Java as a counter-example of useful high-level languages. Python, Perl, Scheme, CL and their ilk are where I'd go to compare.
>>>>>>>>>
I don't really consider Python, Perl, and Scheme to be in the same league as C++. First, C++ is by all means a lower level language, and more appropriate for systems programming tasks. Beyond that, C++ is a far more generic language. If you're working on a very specific task, Python, Perl, and Scheme might very well be extremely useful and productive. But as a base-level, widespread language, C++ fits much better. In other words, it's important to use the best tool for the job, but if I could only know one language, it would be C++.
7) If you take a look at the equivilent data structures in C, you'll realize how much better the STL is. C is a hard-core low-level language. It's the ultimate roll-your-own and about as low-level as you can ever get and still remain portable (that was, after all the goal). Libraries like glib (which have many of the things you're used to in STL)
>>>>>>>>>
glib has nothing on the STL. The original poster suggested ditching C++ and going back to C. I used the STL as an example of something where a C++ and C library try to do the exact same task, but the C++ library accomplishes the task (thanks to templates) in a much cleaner and more elegant way.
I think you took the arguement slightly out of context. I wasn't claiming that C++ is the be all end-all of languages. I was responding to a claim that C++ was a "bad" language, and that we should ditch it and go back to C. My claim is that C++ is a damn good language in its own right, and its complexity is justified by its expressiveness.
The kernel exports certain symbols that are available for driver modules to link to. These symbols are classified by license. Many symbols a driver can link to without having to be a GPL'ed driver. These symbols, presumably, are written by people who grant permission to link against non-GPLed code. Other symbols (for example the new work-queues stuff) are exported as so-called GPL_ONLY symbols, and linking against those in a non-GPL module is illegal. Thus, if you avoid using the GPL_ONLY symbols you can probably get away with a binary only module. The fact that this mechanism exists implies permission, so this aspect would be important to bring up to the lawyer.
Err, nowhere close. I'm running Gentoo custom compiled for my P4, with X reniced to -11, kdeinit (and thus all KDE applications) reniced to -10 (helps significantly to redraw times). I've also removed all unnecessary extensions from my XF86Config, and am using the lastest NVIDIA drivers. Now, KDE does better when the system is under load (like doing a background compile) where Win2k just breaks up, but for unloaded "crispness" Win2K is just sharper. XP, however, is another story, and is much slower than 2K. Even then, some things (like resizing Internet Explorer windows vs resizing Konqueror windows) are less crisp.
Bullshit. Both of them result in a call to graphic driver which , generally, is better optimized on Windows.
>>>>>>>>>
First, the graphics driver doesn't do all that much these days. For example, with Xft2 and sub pixel anti-aliasing, the graphics driver doesn't even handle blitting of text anymore. Second, it depends on the drivers. The NVIDIA drivers I use, for example, are just as good as the Windows version. That said, if it's a driver problem, it's not X's fault, is it?
They use it the way it was intended.
Every fucking Qt call of say QPainter maps almost directly to Xlib function.
What else would you have them to do ?
>>>>>>>>
Do you really believe that? Qt and GTK+ use X as basically a blit engine. They do all the drawing in pixmaps (in software) then blit the results to the screen via the X server. That's most definately *not* how X was designed to be used. There are also numerous issues about not using the protocol properly. Thus, I refer you to the recent thread on the X Render mailing list about XFree86's performance.
You'll take away C++ when you can pry it out of my cold, dead hands. I tend to be a mindless C++ advocate, so I disagree heartily with everything you say. But as for your complaints, I think Stroustrup heard them all before designing things into the language.
1) C++ isn't a high-level language, and it isn't a low-level language. It's both. And that's why it kicks ass.
2) Operator overloading is terribly useful. The whole idea of C++ is that you can implement user-defined types that act like built-in ones. Operator overloading is meant for mathematics-style user defined types (for scientific programs). If you're using operator overloading in a place where it doesn't make sense, then that's your own fault. I hate the Java way of doing things, where everything that could potentially hurt you is illegal. It's the programming language equivilant of Ralph Nader. And > are not just bitshifting operators. They serve double duty as streams operators. I have yet to see anybody ever confuse the two, because it's obvious in most cases that you're not bitshifting the integer 'cout' by "Hello World" or something of that sort. And like printf(), in all it's type unsafe, variable parameter glory is any better!
3) Multiple inheritence can also be useful. Mixin classes, for example, are real nice. Again, if you abuse it, it's your fault. That holds for drug abuse, why not language abuse?
4) There are four casting operators for a reason. You're not supposed to cast in C++. The casting operators are long and ugly and verbose to make the point that you're not supposed to do that. But if you do it, at least the four different casting operators make it clear exactly what you're trying to do.
And templates are mana from heaven. Templates give you the efficiency (yeah, I said efficiency, take that, Java-heads!) of coding custom data structures for each type, while letting the compiler do the actual dirty work. For data structures, the STL beats the Java container classes to a pulp. If you take a look at the equivilent data structures in C, you'll realize how much better the STL is.
You've obviously never used BeOS. I'm running KDE 3.1 on kernel 2.5.44 on a 2 Ghz P4 with 640MB of RAM and a 64MB GeForce4. I love it dearly, but BeOS on my antique P2-300 was a good deal faster, and Win2K on my Duron 750 is still faster.
X is actually faster than the GDI for most things. Don't believe me? Benchmark it yourself? I've benchmarked it before, and for stuff like line drawing or bitmap blitting, X stands up well even to DirectX. The main problem is that the toolkits (ahem, Qt) don't really use X all that well. That said, a lot of problem is due to the fact that a properly fast KDE/GNOME desktop needs a lot of proper configuration. My KDE 3.1 desktop, for example, is as fast as WinXP for most things (expect app startup speed, but glibc 2.3 and prelinking should fix that) but it required some custom compiling and renice tricks to get it that way.
It's not really a problem with either C++ or Qt, it's a problem with the compilers. In particular, with how compilers implement templates. For example, let's say you have the files one.cpp, two.cpp, and three.cpp. If you use std::string for each one, then gcc will compile the string class in every file that uses it. So one.o, two.o, and three.o will all have the parts of the string class that they use (ie, most of it). The linker then discards the redundant bits when the object files are linked together. Now for heavily used template classes (the STL, for example) you basically have gcc compile large parts of a template class library for each source file, then have the linker throw them away at link time. Now the alternative is already part of the C++ '98 standard. You have this thing called the export keyword. Unfortunately, none of the big compilers (g++, Intel C++, VisualC++) support it yet.