Solving the Alt-Tab Problem | Hacker News Reader

Solving the Alt-Tab Problem

azarask.in

204 pointsomergertel98 comments

This is a classic example of the design tradeoff that directly pits new users against power users. MRU is a simple, solid solution. Like many powerful tools, it takes some learning, and is then dependable. The alternative suggested here is a very complicated solution that power users will hate because it's unpredictable.

    You’ve been using alt-tab to bounce back-and-forth
    ...
    tab. What happens? Because of your habit, you expect it
    to go to your web browser, but because the last used
    application was your Twitter client, that’s where it
    switches.

Experienced users will handle this by repopulating the cycle when they switch. In this case alt+tab, release, alt+tab, alt+tab, release, alt+tab.

I guess I'd consider myself an "experienced alt-tab user", and I have this frustration on a regular basis. Yes, as you say, I can manually reorder the apps, but it annoys me to have to do so.

Same here. I've solved the problem by not using any of this alt-tab behaviour. I have a suboptimal usage pattern that's still more intuitive than alt-tab.

I've bound alt-tab and ctrl-tab to raise/lower window. While I do use four virtual desktops to group windows of similar interest, I generally just hide the frontmost window when I want to switch. This keeps the cycle static and I can do most things by using keyboard only. I rarely have more than 3-4 windows on a single desktop so I've kind of manually grouped the desired applications together already. If I want to break the cycle and switch directly to a specific window, I just point my mouse to it and use the keyboard to raise it front.

I agree with this. While I don't find the current tab system particularly wonderful - back when I used arch linux I think I just used ratpoison /w windows mapped to numbers, this solution means that the user would need to either understand the current mode the task switching program is in at any given time or actually interpret the order of the programs. Yes, it minimizes key-strokes between the most used windows, but it also means that each window might have a connection and behavior that could change at any time. It might work, but I'm sure there are better solutions.

W7 provides a rough equivalent of a static ordering by virtue of its taskbar. My browser and IDE are pinned at positions 1 and 2 and I could typically reach them "always" with Win+1 and Win+2.I'm yet to take full advantage of that though. I use alt+tab, release, alt+tab, alt+tab, release, alt+tab most of the times as well.

That's a great tip, that i'm going to start using.

Thanks.

As far as I can tell, W7 doesn't have a learnable internal logic.

I've started a new job recently where I got W7 (having never used Vista) and have played with it quite a bit. Just when I'd think I had it worked out, I'd open up a larger volume of windows and it would change behaviour.

On top of that there are prominent applications which have never played nice - outlook (and anything that minimises to the status bar - behaves inconsistently), and excel (which has never played nice).

I disabled minimise to status bar in outlook as soon as I moved to W7. Trust me you won't miss it. Win+<num> should work with Vista too (in quick access bar) if I'm not wrong.

From what I learn from people at MS, Office team follow their own UI conventions and they dictate Windows not the other way round. So it would always be office does something and windows follows.

    I disabled minimise to status bar in outlook as soon
    as I moved to W7

I've had a google in search of this and not turned up anything. Playing with the view of things in the systems tray doesn't fix the problem of it staying near the top of your MRU list, against windows convention. How do you do it?

Win+num swaps by application, but not by windows.

Win+tab has a horrible transition effect that I don't seem to be able to turn off without destroying the feature altogether.

Right click on Outlook status bar icon and uncheck "Hide When Minimized".

It's just that it provides a mix of MRU and static ordering. The goal for me typically is to move to a frequently used app e.g., my IDE/Browser asap. Win+<num> gets hardwired and I don't have to remember if my fav app is third/fifth from my current window/app.

    Right click on Outlook status bar icon and uncheck "Hide When Minimized".

OK, now I see what you meant when you said, 'Trust me you won't miss it.'. Yeah, I can see your point but I do miss it, but it's still an improvement :)

* What that in original state (i.e. Show when minimized) and I minimise, the app stays at the top of the MRU stack.

* When that is on, and I minimise, the app disappears out of my alt+tab order altogether.

The problem with using a learning model in an interface is making it smart enough to do its job, but dumb enough that the user can easily have a mental job of how it will act. (See the excellent article "Wolfram Alpha and Hubristic Interfaces" and HN discussion: http://news.ycombinator.com/item?id=695582)

The more knobs and heuristics used in the interface, the harder its going to be for me to both guess what's going to happen, and know how to "fix" it when it guesses wrong. I think this is an interesting UI problem, because I encounter it frequently. But I also know how to remedy it easily.

The problem with using a learning model in an interface is making it smart enough to do its job, but dumb enough that the user can easily have a mental job of how it will act.

Yes, exactly. At the same time as the switcher program is trying to learn the user's common patterns, the user is also trying to learn how the switcher works. I can see that not ending very well!

It would be particularly frustrating during the learning phase, when the switcher doesn't know the user's habits and so switches poorly, and at the same time the user is struggling to figure out what order the switcher will put things in.

I use Spaces (OSX's virtual desktops) and have them on hotkeys. Command+1 gets my editor, Command+2 is browser, Command+3 is terminal, Command+4 is Photoshop.

This setup eliminates the need for alt-tabbing. I just go directly to what I want.

With Quicksilver, you can assign hotkeys to the applications themselves. Once I got this set up, I stopped using Alt+Tab.

Though it seems few people here are Windows users, Windows 7 lets you directly switch to/launch apps by using WinKey + [Num], where [Num] corresponds to the app's position in the dock.

Chrome lets you do the same with tabs using Ctrl + Num.

This behavior is also true on Chrome/Firefox for Mac (s/ctrl/cmd/). Safari differs in that they correspond to bookmarks on the bookmark bar; I actually find this to be my preference.

Once I learned that it works for the bookmarks, I prefer it that way too. Especially for the bookmarklets for delicious, yc (thread-collapser), and instapaper: just mash the key, and I never need to touch the track pad.

On Safari+OSX, cmd-{ / cmd-} switch tabs. I'd still like a to-beginning / to-end key pair (maybe cmd-[], and reversed?), but that's pretty minor compared to not needing to navigate with a cursor.

But typically I have more than 10 tabs opened.

Nice to know though. I'm also pleased that it is available for Windows 7, might make my whole Windows 7 experience much more smooth.

Is there anything similar for Ubuntu? With cairo dock perhaps?

WinKey + Num also works in Vista.

Any ideas for comfortable hotkeys that are unlikely to conflict with hotkeys in applications?

This could be a good idea but there is also something to be said about not having to have an animation delaying the switch, etc.

This solution is basically static order with manual (but pita) rearrangement.

Rearrangement is pretty easy, just click+hold the titlebar of the app you want to move, and hotkey to the location where you want to drop it.

I agree about the slide animation, though. I wish there was a way to disable that.

I use KDE3, where each desktop is indeed a different desktop, not (as with Spaces), a sectioned view into a single desktop.

What frustrates me on OS X is that even when I'm on Space 2, alt-tab still shows me apps from Space 1. I much prefer the KDE3 approach where I can tell it to only show me apps and windows from the current desktop.

And if I really want to see all apps I can use Kompose, a version of Expose for KDE3, though that's rare because, like you, I use a different desktop for specific tasks, grouping the apps I need.

That gives you a good tabbing mechanism between apps, but you lose the ability to use Command+n to switch to tab #n in whichever app your are using (Chrome, iterm, etc.) I switch between tabs in the terminal and web browser more than I switch between apps so I prefer to leave the Command+number combinations for that purpose.

I just use Ctrl+number to change spaces, and Cmd+number to change browser tabs/terminal windows.

Works pretty well actually

The solution that (mostly) works for me is a combination of multiple virtual desktops (I use 9), and a tiling window manager (dwm). I usually don't exceed 2 or 3 windows per virtual desktop and the number of desktops is fixed. I have it wired into my fingers that Super+<n> takes me to desktop 'n', and then, typically I'm on the window I want. If not, it is a constant number of "Super+n" to get me to the window (because I don't exceed 3 windows per desktop).

This works for me because I can remember the "place" where windows are, and it is usually one or two chords away.

I do that too, it's pretty great. I'm on Windows so I use Dexpot, which is pretty okay.

I hadn't heard of Dexpot - thanks for mentioning it. I've been looking for some decent tools for Windows and have been unsatisfied with others so far. I'll try Despot for a while and see how I like it.

First impressions are good, though (at least out of the box) the fullscreen preview doesn't play nicely with multiple monitors :-(

I'm a fan of VirtuaWin myself: http://virtuawin.sourceforge.net/

Clickable link: http://www.dexpot.de/index.php?lang=en

I do the same, only instead of dwm, I use musca. And I have 11 virtual desktops assigned to Super+<1 through 0, and the key left of 1>. I also always use the same virtual desktops for certain tasks, so finding windows is easy: firefox? its in the web browsing desktops: 7 and 8! Pidgin? Thats desktop 9, etc etc.

Finally, switching between windows within a single desktop is a simple matter of changing focus in a direction by using Super+<cursor keys> (I can resize and swap by using Super+Shift+<cursor keys> and Super+Alt+<cursor keys>).

I can also cycle through windows in one frame (Super+c), but its not often I have more than one window in a frame (since I normally have either one fullscreen window in a desktop or i have two to four windows tiled), but when I do, its normally only one, so cycling is more of a "toggle between windows" for me.

Super, being the "windows key", is a great key to use for window management. Windows key is for windows management is easy to explain to people ;-)

I was debating on whether or not "Super+n" was too much of a Linux/Emacs-ism. But, at least on my keyboard, it is a diamond: http://comedialabs.com/images-matti/lite2.jpg

"I have it wired into my fingers that Super+<n> takes me to desktop 'n', and then, typically I'm on the window I want. If not, it is a constant number of "Super+n" to get me to the window"

So you have turned your workspace into a hash table.

It saddens me to see Aza proud of dropping out of "3/4 of the schools he attended" as his title bar proclaims. It's one thing to note that a lot of people drop out of schools to seek their passions, but I never think of the dropping out as being the moment of pride.

Your loyalty should be to yourself, and your sadness comes from within. "Finishing what you start" is not virtuous. Mining what you need and then redirecting your attention is a smarter approach.

Questions of loyalty aside, you're trying too hard to be profound. I'm using sympathy, an interesting human emotion, to make commentary, not dredging up personal strife or attempting directly to change Aza's life and purpose.

Also, I still have to disagree with marketing dropping out 3 times so blithely. Opinions may differ, but as it stands I could interpret that to mean he's had 3 semesters at different colleges and the a degree from a community college. Without even critiquing that course of action, it certainly doesn't seem to convey the same message Aza undoubtedly wants to share.

"I've attended four colleges to learn what I use today" is closer to useful, though still weak when you're talking about such easily misunderstood life choices. It invites questioning, discusses the positive mentality that might lead to those decisions, and doesn't sound like a Business Week fluff piece.

98% of people who've had a successful life have dropped out of college; isn't that surprising and almost certainly causal?

Pretty cool to see a design lead with knowledge of Markov models. Granted, someone on his team may have filled him in.

Aza Raskin is also a developer: http://code.google.com/p/pythonicjavascript/

And he majored in math and physics at University of Chicago. lol

Knowledge of, but as a statistics geek, I see a lot of people (like this guy) throw around the term "hidden Markov model" when they don't know what "hidden" means in context. In this context, an ordinary Markov model would be more appropriate.

A hidden Markov model denotes the existence of some dynamic parameter that the computer cannot directly observe, such as the user's current mood. Such a model would take into account transitions between "happy" and "angry" in addition to transitions between "Firefox" and "TextMate". These additional hidden variables are very useful in a wide variety of circumstances, but in this situation I suspect it would cause a combinatorial explosion of state space with little measurable benefit to the user.

Well, in Firefox I just move together the tabs I'm likely to switch between. This way I can use Ctrl + Tab and Ctrl + Shift + Tab to navigate between all tabs with a very specific and simple behavior.

Yes, it takes a little time to move the tabs next to each other, but I think it's still better than breaking the "suspension of disbelief" when things go wrong.

So, I think the solution for the Alt + Tab problem would be use MRU, but with a movable, always visible order: basically, let me drag the apps in the task bar.

I always felt that left hand ctrl+tab is an awkward position for the fingers. I prefer Ctrl + PageUp/PageDown to switch, and like Chromes Shift + Ctrl + PageUp/Down to move tabs.

It's not awkward if you remap caps lock to control.

Or have a keyboard that places them "correctly".

I don't know about you, but my Tab is on one side of the keyboard and the PageUp/Down are on the other. Pressing this keys would require me to have both hands on the keyboard, which, after I uninstalled Vimperator, I don't have anymore.

I can understand the issue the author discusses with making tabbing MRU, but I think preference for this behavior differ between users. I have no trouble internalizing the order of the last few applications (or browser windows), thus MRU is the best for me. I'm likely to have several windows open, and it's fastest for me to be able to quickly bounce between a constantly-updating MRU list.

Let's take the "Order of Application Opening" model: If, say, I had opened iTunes between a messenger program and a browser, and I were flipping between browsing and chatting, I would think that iTunes, being a background sort of application, would require me to remember to double-tab every time I wanted to bounce, requiring an extra button press and an internalization of that data instead. (As well, wouldn't the static list then require me to Alt-shift-tab every time I want to bounce backwards? I certainly could tab forward and wrap around if I had only those two applications open, but that's rarely the case, so it seems that a static ordering would only ever hinder me. Keep the static order to the taskbar/dock where it belongs!)

Somewhat related: Something I've always wanted is the ability to navigate tabs in Firefox with CTRL+Left/Right-arrow, similar to the CTRL+# to get to a certain tab number. I looked around once, perhaps a year ago, to no avail, and haven't searched since. Anybody know of an add-on that adds this functionality?

edit: BoppreH, in another post, mentions CTRL+[Shift]+Tab. I'm basically asking for this functionality, remapped to CTRL+Arrows (It's more comfortable this way, rather than isolating the button presses to one hand)

I would simply use autohotkey:

    ^{Right}::Send ^{Tab}
    ^{Left}::Send ^+{Tab}

But then, that changes behavior for all applications, not just Firefox.

There is a keyconfig extension for Firefox: http://kb.mozillazine.org/Keyconfig_extension you could try that.

In Opera it is easy. Personally, I have this functionality under 'j' and 'k'.

You can do this in OS X's preference pane. All my tabbed applications, even Terminal, now use Command+Option+Arrows.

And how do you break such a "habit", once formed?

Sorry, MRU is still the best. Pisses me off to no end that Firefox and Chrome, by default, do not use it anymore for their tabs...

I prefer browsers' tab switch system, because I always see my tabs. I do not always see my Cmd-Tab prompt, nor do I want to study it when tabbing around.

And the "breaking" is just a question of implementation, not a big theoretical headache.

For me, the most desirable method changes based on the context. In the context of Alt+Tab between applications, I agree that MRU works best. (Even though I get annoyed when I get a third application in the loop and accidentally tab to the wrong one, it seems like the "habit" business would just make the order of switching unpredictable, which would be even more annoying.)

However, in applications like Firefox/Chrome, where the tabs are always visible, I'm the opposite from you: MRU behavior drives me absolutely bonkers. This is because the actual order of tab switching no longer matches the order implied by the visual position of the tabs. To my disgust, Visual Studio changed to this MRU behavior in (I think) 2005, and ever since then the Ctrl+Tab behavior has been useless to me, because even though the tab I want is right next to the active one, it might take six or seven switches to reach it. My memory of the order of the most recently used tabs while I'm coding is poor enough that the ordering might as well be random.

If you wanted to keep the MRU method in this type of application, I think a solution is to automatically reorder the tabs. But VS at least doesn't do this...not sure about Firefox (it didn't originally, and I turned off the feature ASAP).

FWIW, your post (even including the VS behavior hate) is an accurate description why we don't do MRU tab switching in Chrome. It's a very common feature request[1], but we think we can do better and don't want to ship a middling stopgap.

[1] Few people that request it continue to request it after trying it in a browser that supports it. Certainly the number is significantly below those that prefer tab-order switching.

And the fact that you don't allow changing this behavior (not even with extensions, as Firefox does) is what keeps me away from Chrome...

In Visual Studio if you hold alt-Tab it'll pop up a quasimodal dialog that you can navigate through with arrow keys to select the tab you want, or hop into one of VS's panes (properties, ReSharper unit tests, etc). It's much less ideal than being able to just switch to the one you want, but if (like me) you often get stuck with too many tabs to cycle through easily it's at least a way out.

True. However, when I get to that point, more frequently my way out is right-clicking a tab and selecting "Close All But This". :)

I don't see how this will apply to browser tabs reliably. Are you doing it based on page url or domain?

Page url has the obvious shortfall of being so rarely used for a long amount of time. It might be used for an hour, or perhaps a day, or in some rare cases where you do interact with a single url for a long period of time, but those cases will be few and far between.

Domain seems too generic. If I'm on HN then I may flip to gmail frequently as I work through the day, but that doesn't mean that when I'm in Gmail I want to always tab to HN, perhaps it's to my blog, or work's admin section.

I guess actually my issue is that he uses a single pair example.

If I'm on HN, then gmail would probably be a likely tab target, but on Gmail, the options of targets greatly expands. HN would be at top because of recent use, but the other sites mentioned would also be likely targets.

So then we suffer from the frequent order changing that was cited as a downfall of the MRU method.

Am I wrong here?

What about simply using two commands? As the article says, there are two use cases which are hard to combine, so why not separating them?

For instance, use Alt+Tab to jump to you latest used application, and use Ctrl+Rightarrow to go to the next app in order. Plus, you get an obvious way to go to the previous app in order: Ctrl+Leftarrow.

I'm on Windows, and I've rolled my own solution that uses incremental search, sort of like Emacs ido-mode buffer switching. http://bitbucket.org/jasulak/switcheroo/wiki/Home

Not for everyone, but I like it.

I'm on a Mac and I also rolled my own solution. I hard-coded my most frequently used applications to function keys. It's a great alternative to having multiple monitors (because I'm a nomad freelance developer and do not have a desk).

Here are the scripts: http://techiferous.com/2009/12/streamlining-your-workflow-wi...

Thanks for the pointer. I was looking for something just like this.

It seems to install and work just fine on Win Server 2008 64bit, for reference :)

Glad you like it. Thanks for the info.

What a misleading title! I was all excited that he'd actually solved the problem, and I could download something.

As Toby says to Will in an episode of the West Wing: "don't come to me with half a thing and not be able to - you know, after you've walked me to the brink..."

Changing from a simple "most recently used" approach to a complex machine learning approach doesn't seem like a a good idea. Perhaps there would be some other small modifications that would keep the system simple and deterministic?

For example, he mentions the problem that one gets into once one switches accidentally to the wrong app: once you've switched, it becomes the MRU when you try to switch back. Perhaps a rule that an application isn't added considered the MRU unless you interact with it in some way? So that if you switch to it but immediately switch away it is no different than hovering?

I'm surprised to see so much skepticism on this thread. When I read the article, I was delighted to see my own experience described so accurately. It's the second failure that's really galling.

It made me wonder what it would take to implement something like this on OS X (in terms of hooking into the OS, rather than in terms of the heuristics). Since the application Witch (http://manytricks.com/witch/) can be used as a replacement for the built-in App Switcher, it must be possible ...

In ion3 I have a fairly efficient set of keyboard shortcuts for tab navigation:

Alt+Left jumps to left tab. Alt+Shift+Left moves the current tab to the left. Same for all other directions. Oh and obviously tabs (windows) never change position on their own.

This works much better than any other system I have used. Muscle memory is a beautiful, magical thing.

It's a pity that all mainstream OS'es insist on deliberately making it impossible to take any advantage of it. MRU in alt-tab is the most idiotic idea since the modal dialog.

Windows 7 got pretty close to solving this problem.

Winkey+Num makes the n-th window on the taskbar active. Since you can modify the order of the taskbar icons, this solves the spatial memory vs. fast switching between two windows.

Winkey+1 for Firefox, Winkey+2 for Twitter, Winkey+3 for something else.

This combined with Alt+tab that cycles in the same taskbar order (as opposed to MRU) seems like the holy grail to me.

The problem is that this would break the existing expectations of Alt+tab, which is why I fear it won't happen.

Stuff like this is why I never learned to use modifier-tab in the first place. I've always been happy to use the slightly more expensive menu/palette selection idiom, since I can reliably communicate intent to the device without situational thinking. There really is a a degree of wasted effort in trying to make computers "read our minds." Trying too hard to make the common case easy can make even the 2nd common case too expensive to justify the savings.

This problem has frustrated me so much, here are things I do currently:

- Ctrl-<number> focus on or start my regular applications (with an autohotkey script). That is 10 applications, more then enough.

- Alt-Tab and Shift-Alt-Tab are replaced with Ctrl-; and Ctrl-h

I've also tried a script that matches Ctrl-<number> with the programs that are in the taskbar. But found that as the order changes from time to time, I have to recheck what number to use.

I'm not convinced by the proposed solution. You're changing an ordering method that's predictable to one that is unpredictable, from a human perspective.

Why not have more than one row of alt-tabs; use the cursor keys to select which one. But in fact I do this already, just without the visual representation, and it's confusing too:

- alt-tab between firefox and vim

- ctrl-6 within vim

- ctrl-tab within firefox.

BTW: you can rearrange tabs in firefox, and tab-mix (plugin) allows you to expt with customizing it in every weird way you can imagine, and some other ways too. A good test-bed for prototyping alt-tab solutions.

My solution is to use mostly non-overlapping windows and add more (and bigger) monitors. Overlapping windows is a poor design forced on us back when we had 640x480 7" monitors. It's time to let it die.

My tab usage in FF is less switching tabs, and much more opening lots of tabs and going through (and closing) them one by one. When I find my self switching tabs in FF too much I drag one tab into a new window.

This is my usage pattern as well, except I am using Chrome, It really highlights (one of) the benefits of having each tab in a separate process: it doesn't need to reload the page saving both time and the keeping the spot I have scrolled to on the page.

I don't understand why you couldn't just let the user optionally specify such pairings manually instead of trying to infer it in an unreliable way. It's not like you regularly use dozens of different apps which would benefit from a pairing like that. If the user frequently wants to switch between the text editor and the browser, just let him specify so directly.

In xmonad (and I'd assume most other tilers), there is not only a fixed tabbing order, it is the visual order.

I am almost unable to use alt-tab on a more standard window manager now, because it doesn't behave sensibly; it requires you to keep an in-memory ordering on your applications to get to them quickly. A static order that isn't obvious is just as bad as a dynamic one.

(tabmix plus for FF(so that it remembers the order of my browsing)) + (alt(pc)/command(mac)-tab for apps) + (alt-~ for windows within apps) = maybe I'm missing something, but i have no problems. The solution proposed is too complicated for me to want.

Missed the point. He’s talking about the order of apps (or tabs) when switching and optimizing the common case. Listing keystrokes adds nothing…

An idea just popped into my head- Tab+1, Tab+2... Allow manual ordering, or perhaps order the same way as the task bar, so a visual guide is always visible and the order does not change often.

Was anyone else hoping for a discussion of how or why browsers almost completely ignore MRU for alt-tab? Only Opera supports cycling through tabs in both MRU and tab order (configurable).

If you enable browser.ctrlTab.previews in Firefox's about:config panel you can get MRU tab-switching with thumbnail previews.

Anyone solve the "I have to scroll down to reach content" problem?

It's even worse now. I've noticed a trend of starting articles/blog posts with giant, irrelevant/unnecessary pictures and the actual content ends up below the fold. I want to read the article not look at a picture.

Maybe you should try Readability: http://lab.arc90.com/experiments/readability/ . I do notice the trend as well, and I think I have my resolution set pretty high compared to the average user. I wonder what they think.

Yeah, readability is good. The pictures mostly bother me because I do most of my browsing on an 11 inch monitor.

Windows 7 solves this quite elegantly by just showing you small previews which you tab through. It takes away the element of (unwanted) surprise.

If anyone wants something like this in XP, there's Microsoft Powertoys Alt-Tab Replacement.

But it still takes the attention back to the computer.

Actually, as a holdout from my fvwm2 days, alt-tab raises a window, and alt-shift-tab lowers it.

This change is almost impossible to predict and is breaking existing UI. Will not fly.

Keep the order fixed but let me shuffle it manually

For any common pairs of applications that means n/2 tabs on average to switch back and forth.

No, because you would manually shuffle common pairs of applications to be next to each other.

It would need to swap the common pairs on use. Which is basically this guys scheme except you manually define the pairs.

Then again on windows you can alt-shift-tab, so that pretty much takes care of that anyway, if could just have the programs in consistent order.

OMG YES YES YES! That is exactly perfectly described. When you are switching tabs the 3rd application messes things up. And what is worse is that what if the third application was an accident, now you have to re-alt-tab everything to the proper order.

Its kind-of the same problem as when you send out emails. Why can't your email client remember that while there are 5 harrys in your address book, only Harry C is the one you send emails to and thus should be on the top of the auto completion list (FF3 betas missed this feature, FF3 has this now).

The problem is:

1) If we do long-term computer learning, the person might judge that this feature is useless. A second person sitting at the computer screws the whole deal.

2) If we do a manual way with "rearrange" were still screwed. Too much work for temporary windows.

3) If the computer guesses wrong, were also screwed because now were stuck with an incorrect ordering.

Consider this one:

What if instead of trying to get alt-tab to work right, we try to get a consistent (press alt-tab x number of times to get from application foo to application bar) which will remain consistent no matter how many windows are open. So if you go from Eclipse -> Firefox, Eclipse -> Firefox -> Twitter, Firefox -> Eclipse, Firefox -> Eclipse -> Twitter, the user can now form a great habbit: If you are in FF, alt-tab = Eclipse, if you are in FF 2xalt-tab = twitter. Etc.

I think a simple formula of Frequency of Use will help. If a habbit is frequent (person switches from application foo -> bar, even using non-alt-tab ways) we can use frequency of use to determine the ordering. Otherwise fall back to LRU (we can always ignore some random tabbing cases). Since a person will keep the recent "places" they been to in their head, it will feel intuitive for them.

"Why can't your email client remember that" - Gmail can figure this out, its a lab feature. Enable it. (See how I assumed you use gmail :), its probabilistic reasoning in action!)

What's the lab extension called? I can't find it. The closest I could find was "Got the wrong Bob?", but that only warns you after you pick an unusual Bob. Sounds like a dirty hack which wants to parade as a clever feature. They should instead just fix the ordering of the auto-suggest dropdown.

Imagine it is not the tabs of the browser, but rather the knifes of a surgeon or the controls of an passenger airplane. Would adapting the machine to the human still be that desirable?

Humans (still) learn faster and adapt better than machines. Over simplifying stuff for them may actually make things more complex.

I agree with the problem description, but the solution is terrible. It adds unpredictability to the interface, which is worse than simple badness. I'd suggest having new alternative to Alt-Tab that has static positions (reorderable for the power user.)

Reply on news.ycombinator.com