How Would You Refocus Linux Development?

buddyglass writes "The majority of Slashdot readers are no doubt appreciative of Linux in the general sense, but I suspect we all have some application or aspect of the platform that we wish were more stable, performant, feature-rich, etc. So my question is: if you were able to devote a 'significant' number of resources (read: high-quality developers) to a particular app or area of the kernel, and were able to set the focus for those resources (stability, performance, new features, etc.), what application or kernel area would you attempt to improve, and what would aspect you focus on improving?"

Comments

Three things.

[ArcherB]

Better hardware support
Better performance
Maintain excellent reliability.

What else could you need?

Re:Three things.

[__aawkdb2598]

• Visual coherency and a refined GUI. Taste in UI's vary between people, but most linux GUIs that aren't very minimalist tend to suffer from wasted space.
• In interests of making linux more accessible, more configuration utilities that don't require specific knowledge and in-errant editing of configuration text files.

Re:Three things.

[The Mad Debugger]

1. Better GNOME usability for Ubuntu (with delivery of Bulletproof X and the GTK Xconfig ASAP, please)

Seriously, the desktop lacks stuff that has been in Windoze since '95. The kernel works pretty good. We have pluggable storage okay.. but there's still basic holes in the usability (like changing the res on the fly when I move my laptop in and out of my office) that just need to get fixed.

2. Spend whatever time is left over to make OOo faster and easier to use.

The MS Office import filters are so *almost* there, but this app really needs to close the usability gap with Office. I have a semi-decent machine running Ubuntu, and even with Java disabled, it still takes what seems like forever to open a simple document that someone emails to me.

I know these aren't *really* linux-specific, especially OOo, but it's what needs to happen to make linux a real, legitimate desktop force. I'm an easy sell, I love open source, but right now there are too many excuses for why this stuff isn't gettin' fixed, and not enough fixin' it, and right now I'm not telling my computer-illiterate friends that they should go order a Dell machine with Ubuntu preloaded.. I'm telling them to buy a Mac, so I don't have to tell them how to fix basic stuff.

Re:Three things.

[timeOday]

You know what I'd love more than further improvement in any of those areas? Comprehensive, well-written documentation.

Re:Three things.

[ArcherB]

Visual coherency and a refined GUI. Taste in UI's vary between people, but most linux GUIs that aren't very minimalist tend to suffer from wasted space.

Granted, this is important to the Linux community, but when I hear Linux development, I think kernel, modules, and organization (like what goes in /etc, what goes in /bin, what goes in /usr/bin, and so on). Things like KDE, Gnome and other window managers are merely applications as far as I'm concerned and should be considered no more Linux development than, say, Open Office. Of course, I don't mean to say your view is wrong in the least. I just considered the question more narrow than you did and wanted to explain why I didn't consider any X development as part of the question.
Also, since X relies on video hardware, I'd consider X and XGL/Compiz-Fusion/Beryl to be categorized under hardware support.

In interests of making linux more accessible, more configuration utilities that don't require specific knowledge and in-errant editing of configuration text files.

Good point, or better yet, make these files standard across distros so the same configuration utilities works as well on Gentoo as Ubuntu.

A standard method for installing applications across distros would be nice too. I forgot to mention that!

Re:Three things.

[EvanED]

Granted, this is important to the Linux community, but when I hear Linux development, I think kernel, modules, and organization

Then you didn't read the summary very carefully:
if you were able to devote a 'significant' number of resources (read: high-quality developers) to a particular app or area of the kernel

In other words, something that improves KDE, Gnome, X, etc. is a perfectly fine answer to this question.

Re:Three things.

[Chandon Seldon]

Would you do the same for Windows? Would you lump Windows Vista, Windows Server 2003, Windows CE, Windows Cluster Edition, and Windows Mobile all together into one product?

There aren't really more significant Linux distros than that, and they aren't really much more redundant by purpose than that either. Sure, there are a lot of hobbyist distros and LiveCD distros, but those aren't duplicated effort - those are hobbyists playing around.

Re:Three things.

[Yaztromo]

Why? Could you give them a naked box and a Vista DVD and have them get it up and running?

This is exactly the sort of attitude that I would first change about Linux: the idea that Linux needs to replicate Windows, and that it doesn't need to excel past Windows to win the hearts and minds of users.

Linux distros need to strive to be better than Windows. They shouldn't be attempting to duplicate the Windows desktop (something early versions of both KDE and Gnome were, IMO, quite guilty of); they shouldn't be attempting to simply improve upon Windows; they should be attempting to create a unique, best-of-breed solution that users will actually be excited about running (like what Apple is doing with OS X).

So if you can't give the relatives a naked box and a Vista DVD and expect them to get through the install, then we should be aiming to allow them to do this with Linux. The goal of a non-server Linux distro shouldn't be to wind up with the same mundacity that Windows has, or a feature-to-feature parity while ignoring those features Windows lacks; it should attempt simply to be the best operating system it can be -- and if this means forging new ground, then that's what should be done. Linux will never win by trying to be Windows, or by having people say "Well, Windows isn't good at that, so we don't have to be good at that either" -- which is, in effect, the attitude you've just displayed.

Yaz.

Re:Three things.

[WNight]

Why should we strive to beat MS at some ridiculous game someone else thought of?

Either there's a simple installer (no buttons for advanced mode) and it wipes partitions without asking, or there's a complex one that asks too many questions. This is why there are multiple distros - not every one needs to be installed onto a bare metal machine by a newb without instructions.

Besides, Linux has *far* exceeded every MS windows installer. Boot Knoppix and install Debian or RedHat on a partition, while browsing the web, SSHed into your servers, watching a movie, using the included development tools.

Windows on the other hand, if you don't get it pre-installed on the machine requires you to reboot, answer a bunch of questions without help available, makes your machine unusable until finished, requires you to go find the patches, requires multiple reboots or an admin to make a slipstreamed install CD, etc...

Besides, if your parents need a machine they can drool over, buy a Mac, or at least pre-installed Ubuntu. They've turned off the fancy stuff to keep people from hurting themselves.

Re:Three things.

[RAMMS EIN]

``Linux distros need to strive to be better than Windows. They shouldn't be attempting to duplicate the Windows desktop (something early versions of both KDE and Gnome were, IMO, quite guilty of); they shouldn't be attempting to simply improve upon Windows; they should be attempting to create a unique, best-of-breed solution that users will actually be excited about running (like what Apple is doing with OS X).''

I think it does that, and I think _that_ is the real strength of Linux (and, actually, other free *nixes). As far as duplicating the commercial vendors goes, of course the open source world will always be behind. This is necessarily true. Still, people will be looking for the same functionality in the same places, and will complain when they don't find it. What they are missing is that the Linux world has so much *more* to offer besides duplicating the functionality of proprietary software in Free software.

Forget about MS Office for a second. Forget about the latest edition of your favorite Windows-only game for a second. What do you really want to do with your computer? Do you want to run these programs, or do you want to obtain the results that these programs allow you to obtain (i.e. getting your work done, or having fun, or both)? If it is the former, your _only_ option is to actually use these exact programs. But I think that isn't actually what you want. What you want is not actually using specific software, but the results of using the software. And this, Linux can give you in so many more ways than Windows can.

Want to write a paper? You can do that in MS Word (even on Linux). Or you could use OpenOffice.org Writer. And that's where most people will stop. But there is so much more! For example, you could write your paper in LaTeX using a text editor. This is a completely different approach, and, if you only know MS Word, it will require a significant effort to master it. However, that doesn't mean it's not a good idea. I use LaTeX, and I always get complimented on how good my papers look. And I didn't have to do anything to make it that way: I just typed the text, LaTeX made it beautiful. And got all the numbering and references right (something MS Word still gets wrong...wtf?) I use LaTeX because it makes my work _easier_.

Want to install software? You could download the source code and compile it. Or you could go to RPMSeek or whatever and search until you find the right package for your distribution, all its dependencies, and install them all. Or you could use a distro like Debian where you just use apt-get (or Synaptic, if you like GUIs) and it will fetch the package and all its dependencies _and_ provide you with upgrades together with the rest of your system. Ever update _all_ the software on your Windows machine with a single click or command?

Wonder why Linux users always seem to use the command line? Perhaps it is because there aren't any GUIs on Linux. Or perhaps there are GUIs, but not for what these users happen to be doing. Or it could just be that they use the command line because they find it convenient. Ever renamed a whole bunch of files? Thrown out old files, but not new ones, from a directory? Looked up the IP address associated with a certain hostname? Explained to someone how to navigate, through several screens, to a certain field where they had to make a change to some setting? Sometimes, the command line actually is quicker, more powerful, or simply easier.

Ever had the same operating system you work with on your desktop run on an embedded device (say, a wireless router)? Ever upgraded from one Windows release to another, updating all your installed applications in the process, with a single command? Ever experimented with different windowing systems, for example, ones that start up in a blink of the eye, that can be controlled solely using the keyboard, or that didn't have any overlapping windows? Ever trimmed down the options a program gave you to just the ones you used, by commenting them out in the source code and recompiling? Ever ad

Re:Three things.

[David Gerard]

The winning aspect of Ubuntu is: if something's too hard for a newbie, that's considered a reportable bug that should be fixed. This attitude makes all the difference.

Re:Three things.

[siride]

Hmm, this is very insightful. I hope somebody figures this out. Oh wait...Linux distros have been doing this for years. How long have you been using Gentoo or LFS? If you had touched a modern Linux distro, you'd find that this is exactly what they do and have done for a number of years.

I'd use a combination of convex

[Jeremiah Cornelius]

and concave lenses, with a relatively low refractive index and arranged in an optimum series for magnification of subtle surface-details, at quite a close range - say between 200 and 400 mm.

Thanks. I'll be here all week.

Make everything "Just Work"

[vonFinkelstien]

Find out all the things at take too many clicks, or require editing text files and make them "Just Work" in a simple and easy way.

Re:Make everything "Just Work"

[greenguy]

Bingo. This is what keeps me from recommending Linux (more enthusiastically) to my friends and co-workers. I find myself saying things like "They've come a long way on wireless... but they still have a ways to go." Same thing for hibernation. And don't get me started on installing -- I know what "make" does, but I've been at it for several years now. God forbid I try to get a Python app going. (Yes. I do know about the install front-ends on Ubuntu, SuSe, Fedora, etc.]

You want to be 31337? Great, more power to you. Some people have work to do, and aren't interested in matching skillz with you.

I'm aware this is boring shit to focus on. But that's the stuff I want to see.

Might I Suggest...

[bealzabobs_youruncle]

that anyone who thinks that CLI usage is not a feature of Linux think again? This topic is 12 minutes old and three post have already suggested we bury the command line; part of what makes Linux so fast, flexible and customizable is access to virtually every setting from a text editor. This is not something that needs to be changed, instead change your mindset that this is not Windows.

If you are looking for a completely GUI drive *nix I would say OS X is your best bet (yes, I know you can use the CLI in OS X, but you never have to unless you so desire).

Re:Might I Suggest...

[jeevesbond]

This topic is 12 minutes old and three post have already suggested we bury the command line; part of what makes Linux so fast, flexible and customizable is access to virtually every setting from a text editor.

Whilst I totally agree with what follows after the semi-colon in this sentence am not so sure about the part prior to it. All we're seeing is that people do not want to be forced into changing settings--am assuming, on their desktop machines--using the command line. This does not mean we should 'bury' the command line, or stop using text files to hold settings! In fact you've made my point for me:

If you are looking for a completely GUI drive *nix I would say OS X is your best bet (yes, I know you can use the CLI in OS X, but you never have to unless you so desire).

Aye, there's the rub! The user should be able to choose between a GUI configuration interface or editing a text file: everyone's a winner! Also a GUI should be able to read/write text configuration files whilst handling seperate user changes to those files gracefully.

In fact I'd spend a lot of the money on getting everyone (or as many projects as I could) to agree to a configuration file format that could easily be interpreted by an application. A one-size-fits-all library could be written to get the settings from file into memory and back again, then it would just be a matter of organising that data into a front-end that's meaningful for the user. The real joy is that with a standard file format, and library to support it, a rudimentary GUI for a new application could be created in minutes.

This is not something that needs to be changed, instead change your mindset that this is not Windows.

This is a very conservative viewpoint, why can things not change? Why can't we have the best of both worlds, with both GUI configuration tools and text files?

Re:Might I Suggest...

[Chandon Seldon]

for every thing that I can do with the GUI, I want a way to access that with the command line and visa verse.

This is false. You don't actually want a way to do everything with the GUI that can be scripted. For exceptionally esoteric features that are only of interest to programmers, sysadmins, and CLI-aware power users it's perfectly reasonable to have them only be accessable though command line options or a non-GUI config file.

That's how most programs work. Even Windows with WSH and the Registry. Hell, even Windows video games have options that can only be accessed by editing some config.ini file somewhere. If you tried to fit all that stuff in the GUI, you'd never be able to find any of it.

One universal install method...

[mikesd81]

That just freaking works. I never understood why every distro can't just use the same install method. Whatever it may be, rpm, apt, yast whatever. And I don't mean the 3 step make install method. Wouldn't it be great to go grab a package from freshmeat or sourceforge and...oh look that's the package type I need....

WINE

[jstomel]

Admit it, wine sucks and there are lots of programs that will never be ported. I want wine to be integrated and almost invisible, like the Classic interface in OSX.

Open Source 3D and better interactivity

[Omnifarious]

The first of those two is self-explanatory. High-quality, high-performance Free 3D drivers for good hardware.

The second...

I want some (not all) kernel developers to stop using throughput based metrics to measure performance, and instead use a metric based on interactive performance. I have a suggestion for such a metric...

The time between user input and the user input having a noticeable affect on an output device like a display. And I don't think this time should be as short as possible, though that's a good goal. The time should be as consistent as possible while remaining short. I propose a metric that measures this latency and plots the standard deviation of the latency and uses that as the main metric with that average latency being a secondary metric.

Most linux problems are "desktop" issues...

[bytesmythe]

1) Wifi networking
    Too many wifi cards (especially the Broadcom chipsets) are painfully difficult to get working correctly. WPA2 encryption support is flaky. Wired and wifi should switch gracefully.

2) Better sound support
    There are too many conflicting ways of producing sound, some of which dislike working together. Midi support should be built-in. Currently, it's a pain to install. Hopefully KDE's Phonon subsystem will help here.

3) Better a/v
    Too many movies have unsynced audio and video. Also, many codecs are unsupported. Yes, I know they're proprietary, but I don't really care. Ubuntu is making codec installation easier, but frequently the codecs only work with some particular backend. (For instance, even with mp3 support installed for gxine, Amarok (a KDE app) still needs to install it's own. The desktop environment should provide a generic way for apps play audio, and if a KDE app is running under a Gnome environment, it should be able to "just work".) Don't forget the wonderful closed-source
graphics card drivers!

4) Easier windowing subsystem
    No one should have to edit xorg.conf to get anything working. Fortunately, the next release of X windows is supposed to finally do away with this by adding dynamic configuration with xrandr. Also, it will be nice when CompizFusion is more robust. Lots of people really like the eye-candy, and I find some of the features useful.

5) Applications
    It should be easier to keep applications up-to-date. I love Ubuntu, but it drives me nuts seeing bug fixes or major enhancements to applications that I can't easily obtain because either the OS updates don't include application upgrades, or the OS repositories are simply not adequately maintained. I don't want to have to
litter my package manager with repositories, or manually install packages just to keep my apps updated.

6) Laptop support
    Suspend and resume don't always work very well. Some laptops don't come back, and frequently networking
is messed up.

Thourough context sensitive help...

[AmazingRuss]

Seems like everytime I click the help menu, I get some skeleton outline, if anything at all. I don't mind googling around for the information, but if usage is going to grow outside of the techie segment, the help systems are going to have to catch up with Windows 95 era chm files, at least. I'm not talking about technically, but rather actually having some useful content in the systems. I understand that writing documentation is no fun, so I don't hold out much hope for this.

Sure would be nice though.

3D

[G3ckoG33k]

While it has been said before, I believe 3D is the path. Please, please, provide support on par with Windows for any 3D graphics hardware, whether inside a computer or a console.

That is the path to success on the desktop. Today, I cannot even run OpenGL apps or any 3D apps on the lastest and greatest 3D graphics hardware from AMD (formerly ATI), the Radeon 2900XT. Why? There are no drivers. They have focused entirely on Windows, and consider Linux a niche market not worth the effort. Because of that, my family do not have a Linux only machine, which is also why I dual boot. The Radeon 2900XT support may well come to Linux, "when it is ready".

Please, take 3D support in Linux more seriously, whether you are hardware manufacturer or a software developer.

The Elektra Project

[Thaidog]

http://elektra.g4ii.com/Main_Page

I think it's at least worth trying such an implementation. Ok... now bring on the "It Windows again" haters...

Unified packaging system...

[644bd346996]

...with good support for building from source. We obviously need a standard package format with robust support for complex dependencies. The build-from-source part is also really important: right now, no distro (not even Gentoo!) makes it easy to, for example, compile your own Mesa libraries and have them used by the pre-compiled X server. Right now, you can pull a project from cvs/svn and do a make install. But it will overwrite the version from the package and break dependencies. This greatly raises the barrier of entry for testing new code, making the "open source" aspect of Linux software far more accessible.

Once we have a unified packaging system, the meaning of a "linux distro" will change. There will be a lot more sharing of work for the base system, and separate distros will really become sets of config files with just a few changes from the upstream code. Kubuntu is a great example of this: it is a low-maintenance specialization of Ubuntu.

Publicity and Minor Improvements

[thePsychologist]

1) PDF support. Almost all PDF readers on Linux except for Adobe's product have difficulties with large PDF documents. What's with the "LOADING" message that takes forever? Adobe Reader looks horrible (inconsistent with the native GUI). There isn't a single PDF reader besides Adobe Reader that supports subpixel rendering which makes the font rendering hurt my eyes.

2) MIDI support

3) A "configuration manager" that knows most of the contents of the /etc directory and has three windows: a list of text config files, a window that displays the file, and a window with a paragraph or two of explanations and examples on how to change the file.

4) More active development of Fluxbox. It could use more features like shading on mouse wheel scroll and multiple backgrounds for each workspace.

5) A publicity website for Linux! This is probably the most important thing the Linux community could do. Features are nice, but who cares if no one uses them? The website would contain among other things:

-Step by step guide and interactive application to help people select a distribution
-Explanation of all major window manager/desktop environments, again to help people select.
-List of most mature Linux apps with description, screenshots, reviews, and commentary by users
-Discussion forums
-Latest on Linux section: demos of CompizFusion, new apps, tips and tricks, etc.
-Section specifically for articles on switching from Windows difficulties
-User friendly, designed primarily for noobs
-Linux store with quality Linux clothing
-Professional design

Have my list all ready :-)

[Burz]

1. An easy, approachable Hardware Compatibility lookup website. It would consolidate all the compatibility info from kernel & X11 devs, major distros, OEMs and also allow end-users to add their input. FWIW, the HCL at linuxquestions.org is an interesting start but nowhere near exhaustive or current enough to empower Linux users (not hackers) to confidently purchase new equipment.

1a. A certification program for drivers that allows products which meet criteria to bear a special Linux compatibility trademark emblem.

2. Fix the sound architecture. Blocking of sound output still occurs after many years of ALSA. There is no GOOD reason why Harriet shouldn't hear her softphone ring or calendar alarms just because a minimized web page contains a Flash object. Telling her to muck about in the CLI, to buy a pricier multi-channel soundcard, or to learn about sound servers and juggle them is beyond the pale.

3. Create an excellent default IDE for the LSB Desktop environment. The IDE will be geared to target the LSB Desktop spec by default, with desktop applications as the focus. Something you would write a video editor or DVD burner with, not so much a video card or disk driver. GORM on steroids: If it doesn't inspire budding application developers like XCode and Visual Studio then Linux will not inspire application developers to write. Linux will not benefit from many more systems developers at this point because its the apps that matter: The apps sell the platform.

3a. Well-rounded API documentation for the LSB Desktop target, ala MSDN or Apple Developer Connection, eventually integrated with IDE.

4. Enable app developers to become as independent as possible, such that distro managers do not insert themselves between the developers and their users. Distros ought to distribute OS software, and for the most part stay the F*ck away from controlling installation of particular applications. High-level package managers like APT, YUM, etc. should stick to managing (or mangling) the OS dependency tree and leave apps the hell alone! Provide dependency targets in the OS repo like "LSB Desktop", and only one or two others like "Java 6". Then, accept that all the extra stuff you supply on top of LSB is ONLY extra, and will get used when and if the user decides in specific cases.

4a. Ensure those budding app developers can easily share their work with friends and customers. Make appdirs like on OS X and Gobo Linux a standard. Dear God, please.

5. Hacker culture works extremely poorly for application software today. Fund efforts to spread the discipline of user-centered product development. Teach FOSS developers the concepts and ropes of SDLC and Rational Unified Process, with emphasis on adding actor definitions and use-cases to docs and project wikis so that these elements are continually refined and re-thought eventually becoming the centerpiece of requirements. Create use-case instances (scenarios) in close association with unit/app testing scripts. Anything to keep developer minds on the kinds of users and situations the software is meant to satisfy. Encourage budding Business Analysts to do 6-12 month stints with FOSS projects.

6. Create settings persistence (configuration) APIs for crucial system services like X11, Samba, Apache, sound, etc. Get these projects to set and manage their own config files, as no one else seems capable for doing this consistently or well. Maybe when they have to write AND parse their own config data, they will stop creating needlessly bizarre & open-ended formats that umpteen distro tools only understand halfway.

7. Next-generation, object oriented shell based on something like Ruby, Python or even Groovy.

Lastly, all of the above must be in the spirit of fulfilling primary personal computing scenarios like app and driver installation, and configuration of essential services (change screen res, use a network share, etc) in a predictable manner. Unlike MS and Apple, Linux does not yet grok PC land because

Focus on education and after this, documentation

[cdn-programmer]

IMHO the toughest job facing the OSS community is education: teaching, learning, and how to document.

This is compounded by the issue that most developers do not find documentation fun. If the common perception that geeks tend to be nerdy and poor at communication is true, then we have a triple whammy. This is one reason I say documentation and communication and education is our collective biggest failing.

The learning curves for _any_ of our packages are steep. SysAdmins rejoice in the job security they perceive they gain as their expertize for apache, mysql, postfix, postgreSQL and so forth increases. The thing is each package has so many options that it takes forever to learn how to set them up. At last count Debian boasted over 30,000 packages available. How is one suppose to even know what a small percentage of these packages do? That is much less than to learn how to install, support and maintain them?

But this is just the systems administration arena. The API's and programming is an order of magnitude more difficult to keep up with.

Then the documentation. To use WxWidgets for instance I am faced with over 3,000 pages of main manuals, I need to decide if I use DialogBlocks or CodeBlocks or neither. I need to figure out what each does and what each doesn't, and after I buy Julian Smart's book - its another over 500 pages to read. In spite of the fact he's written DialogBlocks there is no useful information on same in his book. Thanx.

This is only one (1) package. I have not addressed version differences and library dependencies and so forth. I have not considered the issues of limitations and bugs.

To keep up is typically information overload to the gawd-zillion'th degree.

---

M$ recognized this and attempted a solution. From what I can tell in around w95 they pulled all the error messages out of the system. I experienced the great joy of accidently turning off the external SCSI hard drive on a W95 computer while the system was accessing the disk... reading it actually. No error message was reported. We got what looked like "END OF FILE". This was M$ code reading the disk.

Then on another occasion I noted a networking message from NT4.0 had the exact same text as from OS/2. The error number in NT4.0 was missing. Everything else was the same. On a hunch I looked up the message in OS/2 and lo and behold the error number lead to the issue at hand. Of course NT4.0 was no help at all because this information had been removed.

Either it was removed or never put in. I dunno. What I do know is that the systems ability to correctly diagnose was hamstrung.

So what do we have in the OSS world?

1) volumes of crappy documentation layered on more volumes of poorly organized documentation.

2) When problems are found and corrected - no good method exists to upgrade the docs.

Here is an example. Many years ago I ran into a sound configuration issue in Debian Woody. This had to do with esoteric issues of generic SCSI drivers and bad permissions and so forth. I ended up posting in SourceForge a complete description of the problem and how to walk through it and fix it.

Two (2) years later none of this information had been disseminated through the documentation of the package at hand where I had discovered it. Debian was still misconfigured. People were still coming into IRC pleading for assistance on how to get the software running (It was GRIP as I recall).

---

This is just terrible performance and we are not getting much better at it.

There are several websites of documentation. SourceForge does this. IMHO they do it poorly. There are many wiki's dedicated to various packages. Nothing is coordinated. The man and info pages I have in my latest system are still the first place I would like to look for information and they are basically just as bad now as they were in 1997. Probably these documentation sources have not been updated much since 1997. Why not? If there is new

Performance Squad the desktop

[schwaang]

I'd like to see a Performance Squad attack all the desktop apps and their underlying components.

Not the kernel, but kernel hackers do know a ton about how to get good performance, so if they all took time out from the kernel to make the rest of the desktop snappy that would be just fine with me.

Of course I've seen some efforts at this over the years. Dave Jones' perennial "why userspace sucks" talk, some work by Robert Love, some other GNOME folks looking at memory usage, the recent Intel tool looking at CPU-wakeups eating battery life on laptops, and lots of other pieces of the puzzle.

It would be great if the basics of performance "best practices" would become widely known by desktop app programmers again. Instead we're falling into Microsoft's habit of being lazy about performance and expecting Moore's Law (increasing CPU speed and cheaper RAM) to bail us out.

Now my girlfriend's answer would be different: OpenOffice still sucks too much (feature-wise), and it's keeping her from switching from Windows.