I'd be very surprised by any university that doesn't have a LaTeX style available. Here, it's the de facto standard for maths and computer science. It may be useful to have a look at the Otago thesis style (scroll down) which is what I use. Of course, if you tried to use it seriously, you'd need to adjust it for your university standards, but that's probably easier done when you have a starting point, rather than learning everything from scratch.
It's also nice in that you can use emacs/vim/kile/whatever else you like to write in. A laptop away from the internet, with only your editor of choice running in a VT is a great distraction-free way of getting things written.
I'm also given to understand (and I haven't used word for a very long time) that doing stable crossreferences and contents pages in word can be an exercise in frustration. With LaTeX, all those sorts of things just magically work (and, with the inclusion of hyperref) you get a hyperlinked PDF file.
Take a day or two and give LaTeX a go. It's worth it, it's just not great to get started with. It's worth checking to see if your university has a thesis style you can use, that will take care of how everything looks for you.
After you've learnt it, you just sit down and write. Maybe when you finish a section or something you'll run 'make' (or whatever) and see how it looks, but that's not usually in the front of your mind. Most of the time you just write away, and when you need to show someone a draft you then turn it into a PDF.
As for differences between the original and the LaTeX version done by the OP, the most obvious one is that the word 'column' doesn't need to be hyphenated. There are a few other slightly odd kerning things going on there too. However, the original definitely looks like it has been done in LaTeX, or at least styled after it, so I do wonder if the publisher is trying to micromanage a few too many things, to the point of breaking a lot.
Hmm, the way I did my Free software project was to get a basic but functioning app done and released as an alpha to people who were willing to test it. This got me valuable feedback, and people to help out with things like art work (that I'm not so good with), that I wouldn't have gotten otherwise. I'm still not up to 1.0, and I don't know if that's so bad. Maybe the 100th release will become 1.00. The version number no longer really means anything.
But my point is, it's come a long way since release, it's gone from 'just barely able to do what it says on the can' to 'stable, robust, and companies give me money to do custom branded versions for their own services', while trying to keep the 'release early, release often' philosophy going.
That would help immensely. I have not found such an option, though. (And we've been looking for it!) How do you enable full symbols?
It's not full symbols like you'd get from a regular stacktrace, but it does give you more meaningful names. I think it's passing '-style' to the compiler. I can't recall the options it takes, but I know one of them is 'PRETTY', which IIRC gives you reasonably formatted JS.
My point was not that it can't be done, only that it is clunky. Invariably I have to rework my use of the widgets several times before I find a layout that I can style appropriately. This is less of an issue when you're building a desktop-replacement app, but it can be a bit figidty when you're trying to integrate GWT into an existing webpage layout.
Yes, that is a different case. In my situation, we're building an app that is completely GWT-based, no old code to fit it in with, and the only HTML we have is in the loader index.html.
Except they don't work. It is not feasible in many cases to configure the server not to cache the *.nocache.js.
Ours get served up by glassfish, so I wrote a filter that gets the requests and adds appropriate caching headers to the files on the way out. It doesn't seem to be 100% perfect, but I suspect that's just because I haven't taken the time to ensure it works in all cases yet.
Your solution is workable, but easy to break. Path issues cause enough headaches to where it has yet to be economical to switch over to GWT compiles for builds. Of course, our current situation is slightly complicated by a predecessor who configured quite a few projects without ANT. This does make converting more difficult than starting from scratch.:-)
Again, all of our stuff is ANT driven, which makes it pretty easy. I have build scripts set up that do all the various things, like set up the paths, compile the parts of the application, build war files, upload them to the application server, etc. So our build process (staging or production) is just run the appropriate ANT script, and sit back and wait for an hour (on the fast machine).
Not really. If you're using modular Javascript, it will be cached once and then reused after that. If I have 3 GWT components on my site, the AJAX, JSON, and GUI libraries are getting loaded three times over. If I have 3 Javascript-based components, the AJAX, JSON, and GUI libraries are loaded once, then reused by each component.
Fair point, I probably wouldn't be too inclined to use GWT to give fancy addons to a page, unless it was really worth it. I'm coming more from the direction of a large application that happens to run in the browser.
Speaking of JSON, I distinctly dislike the way it's handled in GWT. One of the reasons why JSON is so great is that I have a simple Javascript Object once the JSON is parsed.
I don't see why you would do that...I mean, I'm sure there's a use-case, but I don't know what it is. Oh, I suppose if you're getting JSON from another source...perhaps there's a library out there that makes it easier? 'rocket' seems to have stuff that can do that, from a glance.
That's not really true. The MVC model of Swing is highly customizable, and SPIs make Java APIs extremely pluggable. Even if it was true that Java code is not highly pluggable, pluggability would still be the Javascript way.
We've just moved (as in, the day before yesterday) to GWT 1.5 for testing on our project (a surprisingly easy migration, BTW, provided you're prepared to patch.jars that are no longer being maintained), which supports annotations (and other Java 1.5 stuff). I have yet to see if they can be used in such a fashion.
For example, let's say I've got a feature-rich component that displays records returned from the server. I originally write it to handle JSON data. Now let's say my company has a legacy XML stream they want to use the
I use GWT for a fairly large project. It does have it's problems, but I think they're far outweighed by the gains.
Looking at your cons:
Debugging outside of hosted mode is a bit of a pain, but there are things that make it easier. There is a module you can add that gives you a debug console, allowing 'printf' style debugging. Combine that with setting the compiler to output full names, rather than compressed ones helps. That said, I find it pretty rare to need to do that. 99% of issues are taken care of in hosted mode where you have proper debugging with eclipse.
Styling of widgets appropriately comes with practice. You learn where you need to call.setStyleName/.addStyleName pretty quickly, and using things like firebug to experiment with CSS options makes it no more difficult than anything else.
Hashed filenames are used to prevent browser caching issues. The browser will cache the large hashed filename, but not the small 'loader' file (because that's how you configure your server). So when you update the application, the client will always get the latest version. On the other hand, if you haven't updated, then the client can cache the large wad of javascript, thereby reducing load times. I don't put the derived files into version control anyway...they're derived files, you shouldn't need to do that, typically.
You don't need an ANT plugin. I use: [java classpathref="gwt-classpath" classname="com.google.gwt.dev.GWTCompiler" fork="true" maxmemory="1512m" failonerror="true"]
[arg value="-logLevel"/]
arg value="ERROR"/]
[!--arg value="-style"/]
[arg value="PRETTY"/]--]
[arg value="-out"/]
[arg value="dist/gwtbuild"/]
[arg value="${entrypoint}"/] [/java] (de-XMLed for slashdot) and it works fine.
While the compiled javascript can be quite large, it's not accurate to say it includes all its libraries every time. It optimises any unused code out in order to reduce the size (I think this is why the compile process takes so damn long on a large project). If, instead, you mean that it doesn't load on demand, or separate them into files, or whatever...you (as a browser user) don't want that, it'll increase load time by causing more HTTP requests.
I do agree that there is no runtime introspection such as reflection, in order to customise behaviour, but I haven't found that to be an issue. I just wrote the application to be flexible on its own. Load a different application model to have it do different things within one runtime. But, that is the way most regular java apps work anyway, until you get in to dependency injection and similar techniques.
However, if you want to use all the handy tools that java coding gives you, and (almost) seamlessly pass complex objects between the server and the browser, there is nothing better that I'm aware of. Basically, our application manages a lot of data and presents it in various ways while allowing the user to interact with it in the browser, and GWT has been ideal for this. It's also nice that you can manipulate the output code if you really need to (e.g. part of my build.xml modifies the compiled javascript to prevent a bad interaction between firebug and the way the GWT JS does something).
There are tradeoffs involved in taking a language from one environment (JVM) to another (JS), but I really don't think it's as bad as you make out.
Point of info: I donated to Fedora, Ubuntu, DSL, Puppy, OOo, Gimp, ClamAV, and will probably donate to others this year if I find I'm using their code regularly.
As someone who gets the occasional donation for an open source project of my own (not one of the ones you listed), I'd like to say thanks for doing this, and keep it up!
Yes, apparently 32-bit Ubuntu works just fine on x86-64. Personally, I use 64-bit and nspluginwrapper to make 32-bit flash work in the 64-bit browser. It's a bit of a hack, and not totally stable, but definitely works well enough. The one thing that is quite tricky is getting the w32codecs working in 64-bit, you need to install a 32-bit mplayer to work with them.
Just for another data point (heh), I did this with =A1+1 in the first column, and =A1^2 in the second and filled down (with a 1 in the first cell so it all worked) to 2400 or so. Drew a scatter plot, and it came up in about half a second. This is OOo on Ubuntu x64.
On the other hand, they now know a lot about making parabolic antennas, which could serve well in the future. Or, say it was work a few thousand dollars worth of time, and now they can produce them at $10/each. That could pay off not too badly if they wanted it to. But for many people, the learning and fun of implementing could well be it's own reward.
Besides, if you RTFA, he's a volunteer who did it in his spare time.
Yes, the binary firmware thing is a bit annoying. At Linux.conf.au this year, Keith Packard (who works for Intel on video drivers and runs X.org) made the comment that they wish they could open source the firmware "but then we wouldn't be able to sell to Japan". I presume that Japan has some rules about how easy it should be for radio's power levels to be altered or something.
I haven't used Mandrake in a long time, but in Ubuntu, you install network-manager-gnome (and I think there's a KDE equivalent). That gives you an icon that you can click that will let you connect to WPA2 networks without trouble. I wouldn't be too surprised to find that MDK has something similar.
The only real issue with WPA2 with AES is that the firmware in the network card must support it. If it doesn't, the only solution is to by a new network card. Even kernel compiling won't fix that.
I also do genetic algorithms. My take agrees with yours, and I mentioned this in a presentation, getting strange looks. What I said is basically "Mutation is bad. There are a few small situations where you need it, but every where else it does significantly more harm than good". It seemed quite contrary to people's expected views, however they were fairly happy to accept it when I explained why.
(For the interested: crossover takes two generally fit organisms and combines them. This means that it's likely to take a fairly fit chunk of the genotype from one parent, and a fairly fit chunk from the other, and combine them. In a perfect world, this result would be more fit. It's not always the case, but often is. This is the way that new combinations are found. Mutation on the other hand is only useful to introduce the potential for new combinations. So it's necessary if there isn't enough information in any of your population's genotypes to explore new areas in the space of possible genotypes. Almost always the new areas it suggests are bad, but it's the one in a million right guess that you need to keep your population up in the arms race.)
I've heard that Sun considered 64-bit machines servers only, and so they don't have a plugin or Java webstart for it yet. There is a bug report about it, which has been open for ages, and is among the top 25 RFEs in their database. Hopefully they'll do something about it soon. I'm missing webstart.
It's easy enough for a mirror to replace a legitimate package with a trojaned package. And if they offer a valid signature for that trojaned package, you likely won't even know it has been compromised. You'd have to compare the signature from the mirror with that of the main distribution point.
I think you misunderstand how the signatures work. If a mirror replaced a legit package with a trojaned one, they would either have to have it unsigned, or have it signed with a key that isn't one of the ubuntu release keys. In either case, the package manager would spout warnings about it, and it would get found out pretty quickly. Apt tools, in effect, do check the signature against those from the main distribution point. Or, more accurately, against the keys they know about. This means that the mirror couldn't replace the package and have noone the wiser.
The compromise that makes the most sense to me is RFPolicy. Put simply, this provides a 5-day contact period, and requires the vendor to keep the reporter notified of the status of the fix. Time to actual disclosure is then based on how cooperative the vendor is being. This (in theory) ensures a fix in a reasonable time frame, from the point of view of the reporter, while suggesting that the disclosure of the vulnerability should be held back as appropriate in order to do a proper fix, and giving good timelines should the vendor not be responsive or cooperative.
Re:Hate them! Hate them! Hate them!
on
A New Kind of OS
·
· Score: 1
I reckon something like personalised menus could be effective, if done right, and the changes were slow to allow time for people to adjust, and for the computer to get them right.
A simple example: the list of recently used applications that appears on KDE's kicker, and the windows start menu. The KDE one is pretty straightforward, but it starts to fail if you use more applications than fit on the list. How about, if you launch an application the computer notes the time. It looks at a week or two of use, and sees that you launch, say, eclipse when you get to work and then almost never relaunch it until the next day. That could appear on the recent applications list until you've launched it for the day, and perhaps go away after a while if it looks like you won't be using it at all that day. Maybe over lunch you like to play a game. That game will only show up on the recent applications list around lunch time, leaving the list full of things you normally use the rest of the time. If it was a slow learning, so allowing peoples habits to change slowly, and not changing how it works because you decided to play that game at 11 rather than 12 one day, I think it could work.
Slashdot Mirror
I'd be very surprised by any university that doesn't have a LaTeX style available. Here, it's the de facto standard for maths and computer science. It may be useful to have a look at the Otago thesis style (scroll down) which is what I use. Of course, if you tried to use it seriously, you'd need to adjust it for your university standards, but that's probably easier done when you have a starting point, rather than learning everything from scratch.
It's also nice in that you can use emacs/vim/kile/whatever else you like to write in. A laptop away from the internet, with only your editor of choice running in a VT is a great distraction-free way of getting things written.
I'm also given to understand (and I haven't used word for a very long time) that doing stable crossreferences and contents pages in word can be an exercise in frustration. With LaTeX, all those sorts of things just magically work (and, with the inclusion of hyperref) you get a hyperlinked PDF file.
Take a day or two and give LaTeX a go. It's worth it, it's just not great to get started with. It's worth checking to see if your university has a thesis style you can use, that will take care of how everything looks for you.
After you've learnt it, you just sit down and write. Maybe when you finish a section or something you'll run 'make' (or whatever) and see how it looks, but that's not usually in the front of your mind. Most of the time you just write away, and when you need to show someone a draft you then turn it into a PDF.
As for differences between the original and the LaTeX version done by the OP, the most obvious one is that the word 'column' doesn't need to be hyphenated. There are a few other slightly odd kerning things going on there too. However, the original definitely looks like it has been done in LaTeX, or at least styled after it, so I do wonder if the publisher is trying to micromanage a few too many things, to the point of breaking a lot.
Hmm, the way I did my Free software project was to get a basic but functioning app done and released as an alpha to people who were willing to test it. This got me valuable feedback, and people to help out with things like art work (that I'm not so good with), that I wouldn't have gotten otherwise. I'm still not up to 1.0, and I don't know if that's so bad. Maybe the 100th release will become 1.00. The version number no longer really means anything.
But my point is, it's come a long way since release, it's gone from 'just barely able to do what it says on the can' to 'stable, robust, and companies give me money to do custom branded versions for their own services', while trying to keep the 'release early, release often' philosophy going.
That would help immensely. I have not found such an option, though. (And we've been looking for it!) How do you enable full symbols?
It's not full symbols like you'd get from a regular stacktrace, but it does give you more meaningful names. I think it's passing '-style' to the compiler. I can't recall the options it takes, but I know one of them is 'PRETTY', which IIRC gives you reasonably formatted JS.
My point was not that it can't be done, only that it is clunky. Invariably I have to rework my use of the widgets several times before I find a layout that I can style appropriately. This is less of an issue when you're building a desktop-replacement app, but it can be a bit figidty when you're trying to integrate GWT into an existing webpage layout.
Yes, that is a different case. In my situation, we're building an app that is completely GWT-based, no old code to fit it in with, and the only HTML we have is in the loader index.html.
Except they don't work. It is not feasible in many cases to configure the server not to cache the *.nocache.js.
Ours get served up by glassfish, so I wrote a filter that gets the requests and adds appropriate caching headers to the files on the way out. It doesn't seem to be 100% perfect, but I suspect that's just because I haven't taken the time to ensure it works in all cases yet.
Your solution is workable, but easy to break. Path issues cause enough headaches to where it has yet to be economical to switch over to GWT compiles for builds. Of course, our current situation is slightly complicated by a predecessor who configured quite a few projects without ANT. This does make converting more difficult than starting from scratch. :-)
Again, all of our stuff is ANT driven, which makes it pretty easy. I have build scripts set up that do all the various things, like set up the paths, compile the parts of the application, build war files, upload them to the application server, etc. So our build process (staging or production) is just run the appropriate ANT script, and sit back and wait for an hour (on the fast machine).
Not really. If you're using modular Javascript, it will be cached once and then reused after that. If I have 3 GWT components on my site, the AJAX, JSON, and GUI libraries are getting loaded three times over. If I have 3 Javascript-based components, the AJAX, JSON, and GUI libraries are loaded once, then reused by each component.
Fair point, I probably wouldn't be too inclined to use GWT to give fancy addons to a page, unless it was really worth it. I'm coming more from the direction of a large application that happens to run in the browser.
Speaking of JSON, I distinctly dislike the way it's handled in GWT. One of the reasons why JSON is so great is that I have a simple Javascript Object once the JSON is parsed.
I don't see why you would do that...I mean, I'm sure there's a use-case, but I don't know what it is. Oh, I suppose if you're getting JSON from another source...perhaps there's a library out there that makes it easier? 'rocket' seems to have stuff that can do that, from a glance.
That's not really true. The MVC model of Swing is highly customizable, and SPIs make Java APIs extremely pluggable. Even if it was true that Java code is not highly pluggable, pluggability would still be the Javascript way.
We've just moved (as in, the day before yesterday) to GWT 1.5 for testing on our project (a surprisingly easy migration, BTW, provided you're prepared to patch .jars that are no longer being maintained), which supports annotations (and other Java 1.5 stuff). I have yet to see if they can be used in such a fashion.
For example, let's say I've got a feature-rich component that displays records returned from the server. I originally write it to handle JSON data. Now let's say my company has a legacy XML stream they want to use the
I use GWT for a fairly large project. It does have it's problems, but I think they're far outweighed by the gains.
Looking at your cons:
Debugging outside of hosted mode is a bit of a pain, but there are things that make it easier. There is a module you can add that gives you a debug console, allowing 'printf' style debugging. Combine that with setting the compiler to output full names, rather than compressed ones helps. That said, I find it pretty rare to need to do that. 99% of issues are taken care of in hosted mode where you have proper debugging with eclipse.
Styling of widgets appropriately comes with practice. You learn where you need to call .setStyleName/.addStyleName pretty quickly, and using things like firebug to experiment with CSS options makes it no more difficult than anything else.
Hashed filenames are used to prevent browser caching issues. The browser will cache the large hashed filename, but not the small 'loader' file (because that's how you configure your server). So when you update the application, the client will always get the latest version. On the other hand, if you haven't updated, then the client can cache the large wad of javascript, thereby reducing load times. I don't put the derived files into version control anyway...they're derived files, you shouldn't need to do that, typically.
You don't need an ANT plugin. I use: /] /] /]
[java classpathref="gwt-classpath" classname="com.google.gwt.dev.GWTCompiler" fork="true" maxmemory="1512m" failonerror="true"]
[arg value="-logLevel"/]
arg value="ERROR"/]
[!--arg value="-style"/]
[arg value="PRETTY"/]--]
[arg value="-out"
[arg value="dist/gwtbuild"
[arg value="${entrypoint}"
[/java]
(de-XMLed for slashdot)
and it works fine.
While the compiled javascript can be quite large, it's not accurate to say it includes all its libraries every time. It optimises any unused code out in order to reduce the size (I think this is why the compile process takes so damn long on a large project). If, instead, you mean that it doesn't load on demand, or separate them into files, or whatever...you (as a browser user) don't want that, it'll increase load time by causing more HTTP requests.
I do agree that there is no runtime introspection such as reflection, in order to customise behaviour, but I haven't found that to be an issue. I just wrote the application to be flexible on its own. Load a different application model to have it do different things within one runtime. But, that is the way most regular java apps work anyway, until you get in to dependency injection and similar techniques.
However, if you want to use all the handy tools that java coding gives you, and (almost) seamlessly pass complex objects between the server and the browser, there is nothing better that I'm aware of. Basically, our application manages a lot of data and presents it in various ways while allowing the user to interact with it in the browser, and GWT has been ideal for this. It's also nice that you can manipulate the output code if you really need to (e.g. part of my build.xml modifies the compiled javascript to prevent a bad interaction between firebug and the way the GWT JS does something).
There are tradeoffs involved in taking a language from one environment (JVM) to another (JS), but I really don't think it's as bad as you make out.
As someone who gets the occasional donation for an open source project of my own (not one of the ones you listed), I'd like to say thanks for doing this, and keep it up!
Citation in BiBTeX format, please :)
There's 26 different letters there, plus a blank. That's what's copyrighted :)
Here is a similar example: http://www.fukung.net/images/1221/detention_dumbte acher.gif
I'd expect the (multithreaded or otherwise) rooster came first.
Yes, apparently 32-bit Ubuntu works just fine on x86-64. Personally, I use 64-bit and nspluginwrapper to make 32-bit flash work in the 64-bit browser. It's a bit of a hack, and not totally stable, but definitely works well enough. The one thing that is quite tricky is getting the w32codecs working in 64-bit, you need to install a 32-bit mplayer to work with them.
Just for another data point (heh), I did this with =A1+1 in the first column, and =A1^2 in the second and filled down (with a 1 in the first cell so it all worked) to 2400 or so. Drew a scatter plot, and it came up in about half a second. This is OOo on Ubuntu x64.
On the other hand, they now know a lot about making parabolic antennas, which could serve well in the future. Or, say it was work a few thousand dollars worth of time, and now they can produce them at $10/each. That could pay off not too badly if they wanted it to. But for many people, the learning and fun of implementing could well be it's own reward.
:)
Besides, if you RTFA, he's a volunteer who did it in his spare time.
de ZL4TRS
Yes, the binary firmware thing is a bit annoying. At Linux.conf.au this year, Keith Packard (who works for Intel on video drivers and runs X.org) made the comment that they wish they could open source the firmware "but then we wouldn't be able to sell to Japan". I presume that Japan has some rules about how easy it should be for radio's power levels to be altered or something.
I haven't used Mandrake in a long time, but in Ubuntu, you install network-manager-gnome (and I think there's a KDE equivalent). That gives you an icon that you can click that will let you connect to WPA2 networks without trouble. I wouldn't be too surprised to find that MDK has something similar. The only real issue with WPA2 with AES is that the firmware in the network card must support it. If it doesn't, the only solution is to by a new network card. Even kernel compiling won't fix that.
Yeah, right. I bet it was the rooster.
I haven't read that, but some of Richard Dawkins' older books cover this kind of thing too. It's a really interesting field.
I also do genetic algorithms. My take agrees with yours, and I mentioned this in a presentation, getting strange looks. What I said is basically "Mutation is bad. There are a few small situations where you need it, but every where else it does significantly more harm than good". It seemed quite contrary to people's expected views, however they were fairly happy to accept it when I explained why.
(For the interested: crossover takes two generally fit organisms and combines them. This means that it's likely to take a fairly fit chunk of the genotype from one parent, and a fairly fit chunk from the other, and combine them. In a perfect world, this result would be more fit. It's not always the case, but often is. This is the way that new combinations are found. Mutation on the other hand is only useful to introduce the potential for new combinations. So it's necessary if there isn't enough information in any of your population's genotypes to explore new areas in the space of possible genotypes. Almost always the new areas it suggests are bad, but it's the one in a million right guess that you need to keep your population up in the arms race.)
I've heard that Sun considered 64-bit machines servers only, and so they don't have a plugin or Java webstart for it yet. There is a bug report about it, which has been open for ages, and is among the top 25 RFEs in their database. Hopefully they'll do something about it soon. I'm missing webstart.
I think you misunderstand how the signatures work. If a mirror replaced a legit package with a trojaned one, they would either have to have it unsigned, or have it signed with a key that isn't one of the ubuntu release keys. In either case, the package manager would spout warnings about it, and it would get found out pretty quickly. Apt tools, in effect, do check the signature against those from the main distribution point. Or, more accurately, against the keys they know about. This means that the mirror couldn't replace the package and have noone the wiser.
Slashdot is behind the times again! According to all my calendars, it's been Saturday the 14th for 13 hours already :)
The compromise that makes the most sense to me is RFPolicy. Put simply, this provides a 5-day contact period, and requires the vendor to keep the reporter notified of the status of the fix. Time to actual disclosure is then based on how cooperative the vendor is being. This (in theory) ensures a fix in a reasonable time frame, from the point of view of the reporter, while suggesting that the disclosure of the vulnerability should be held back as appropriate in order to do a proper fix, and giving good timelines should the vendor not be responsive or cooperative.
Can be scripted in Gnome without too much trouble. I have a bash script to give me random desktop pictures. The magic command is:
/desktop/gnome/background/picture_filename $file
gconftool-2 -t str -s
that immediately changes the desktop background.
I reckon something like personalised menus could be effective, if done right, and the changes were slow to allow time for people to adjust, and for the computer to get them right.
A simple example: the list of recently used applications that appears on KDE's kicker, and the windows start menu. The KDE one is pretty straightforward, but it starts to fail if you use more applications than fit on the list. How about, if you launch an application the computer notes the time. It looks at a week or two of use, and sees that you launch, say, eclipse when you get to work and then almost never relaunch it until the next day. That could appear on the recent applications list until you've launched it for the day, and perhaps go away after a while if it looks like you won't be using it at all that day. Maybe over lunch you like to play a game. That game will only show up on the recent applications list around lunch time, leaving the list full of things you normally use the rest of the time. If it was a slow learning, so allowing peoples habits to change slowly, and not changing how it works because you decided to play that game at 11 rather than 12 one day, I think it could work.