Category: Technology

There is a nice article on Wikipedia that discusses the multiple document interface, a horrible hack that took hold in the Windows world to deal with the fact that Microsoft’s default window manager was inadequate to handle multiple windows existing under the same application.

I think Mac OS/OS X handles the MDI plague best by simply grouping all application windows under a single application class, with a single menubar. But usability experts have debated whether that makes the most sense, since the menubar can change, for example, depending on what window is in focus. I think users get used to that, and it also allows the menubar to be as long as necessary while the window can remain as small as necessary. That’s a nice win.

However, given our current model on *nix/Windows of menubars for every reasonably complex window, and given the lack of the MDI hack in GTK+, we do have a mess for applications that need more than one window to operate properly, i.e. Glade and The Gimp.

What I’ve been doing is giving these programs their own workspace as a workaround. That seems quite greedy of them, and indeed it is. What’s more, however, is that it’s unusable. Even when I switch to my Glade workspace, I see 4 windows in my taskbar, each with the same icon and with the following names: “Glade: h2h”, “Properties: image95”, “Widget Tree”, “h2h”. The first three are actually part of the Glade window class, and if I enable Metacity’s taskbar grouping, I see them as part of the same group. The last one, however, is just my actual window, and so is separate.

This is good–but I only get some form of usability when I actually enable window grouping. The thing is, in Metacity you can’t enable window grouping on a per-application basis. It’s all, sometimes, or nothing. Sometimes means metacity only groups windows when I’m running out of space. Otherwise it’s either on or off.

The thing is, grouping isn’t just about space saving. It’s about being able to perform window manager operations on a group of windows, i.e minimize all and maximize all.

This doesn’t even solve all problems: alt+tab still shows me all 4 glade windows, which can be quite confusing since only one comes into focus at a time. But that’s a separate issue, separate debate.

Wouldn’t it be nice if instead of Metacity just “figuring out” when to group my windows together, it let me just press a hotkey “Group all X windows on this workspace”, where X is the application I’m currently in?

I’ve decided this feature is so valuable, I may just hack metacity to add it. It will at least provide a path for solving the MDI nonsense.

Update: check out these screenies of a “different” approach to MDI written in GTK. It’s called GTK ADI.

I met with Runar (he’ll have a blog soon, I swear) today, and we discussed open source, Python, and all related goodness over coffee and vegetarian lunch free-riding on the ‘sNice wireless network.

We spent about 3 hours there, just talking about Runar’s project, “sqlstring”, my ideas about inferred typing and static source code analysis in Python, Python’s niceness in general, user interface toolkits, AJAX being a big, nasty hack, and web application frameworks in Java and Python. Our discussion really degrenerated into praise of vim once we discovered that we were both happy users. Text editors really bring people together.

Runar kind of convinced me that trying to infer all the types of objects is very “unpythonic,” which I guess is true since it discourages the crazy stuff you can do with Python. Maybe the best thing to do is judiciously eval code, as was my original impulse for getting nice completion out of Python? Not sure.

Or maybe I should just give up the idea and accept the fact that vim plus ipython is just about as good as it gets. That seems like a cop-out, though.

Regardless, Runar seemed somewhat willing (only half-willing) perhaps to give a small talk for Free Coders on Python, I’ll see if I can convince him that it’ll be fun. I suppose I could give the talk myself, but I already do all the talkin’.

I’ve been hacking up a user interface for my motion capture/computer vision project called “Hand2Hand,” found here.

At first I was gonna do the user interface in Python and have the image processing done in C, but then I decided that the user interface was simple enough that I should just give GTK+ in “pure C” form a try. Of course, I used Glade, which drastically reduces the amount of annoying code for things like Vboxes and Hboxes and Containers you have to write. In fact, using Glade, interface design becomes somewhat straightforward in C. Which is weird, because C seems like it was never built for user interface design, but the g_signal system makes it easy to catch events that occur in your program, and GTK+ is high enough abstracted that you can do pretty well. I don’t know how well GTK+ scales for large programs (i.e. many dialogs, many lists, etc.)–in that case, I think I’d definitely pick a higher level language.

Looking forward to how this application may turn out. OpenCV looks like a pretty awesome library.

I think everyone should revisit Outfoxed, if you’re interested in a truly interesting new approach to bookmarking and web browsing that actually takes advantage of all this “Web 2.0” hype and nonsense.

I just mentioned this to Free Coders on the mailing list yesterday.

The application of “trust-based networks” is very wide. I think it’s the “trust” factor that makes eBay successful (perhaps even viable!). New services like Pandora and Last.fm recommend music by trusting that users who rate music are being honest. eBay users buy from trusted sellers by assuming that those who rate the sellers are honest. These assumptions may be fallible, but they’re better than nothing. And it’s only natural that this trend would spread to web browsing.

I think trust networks should be applied to political organizing, both to reduce risk of people showing up just to start trouble, and also to enhance the perceived value of a meeting based on the combined trust of its attendees. I’m going to think about this a bit more in the next few days.

(Meanwhile, I got in touch with Runar from the Google talk with Alex Martelli, who is working on an awesome library for Python called sqlstring. Check it out. I’ve also been thinking about expanding on my earlier ideas on Python inferred types; we’ll see if I find the time.)

I run a tool on my server which creates charts based on basic server vitals, like free disk space and CPU load averages.

It’s called cacti, and it’s great.

Except, apparently this security hole allowed the hacker who originally broke into my server to get in. He was able to execute arbitrary commands via the good old URL string hack. (He did the same thing as is described in there: wget’ed his own script which added a new user for himself and added him to sudoers, and then connected via ssh).

In my latest upgrades, I saw that this cacti bug’s been fixed.

Scary stuff. Computer security, these days. How does a php script have code which can run an arbitrary command? My Java Servlets never have a way to run command line apps by way of specific arguments in the URL string. Sigh. In *nix we may have [basically] all-or-nothing security (that is, if you discount ACL support)–but knowing this, please prefer “nothing” to “all”, for crying out loud!

While I was studying, I noticed that I wasn’t exactly happy with the cooling/cpu frequency scaling on my laptop. I use a nice program called powernowd which scales my CPU speed up and down depending on various factors related to system load. But I didn’t like how my setup was kind of “all or nothing.” When I am plugged into AC, I switch to “performance” mode which just runs me at 100% CPU frequency all the time (making my laptop hot, my fan noisy, but my machine fast), whereas when I’m unplugged I switch to “userspace” mode, which lets powernowd kick in, and he jumps about from 400mhz to the full 1.6Ghz based on load, keeping the machine cool but also making it feel a bit sluggish since if I’m overloading my CPU at 400mhz it’s already “too late” to pump it up, it will have already felt slow for at least an instant.

So I have this conflict: hot and responsive, or cool and sluggish. I thought, well, I must be able to come to a compromise.

I decided to take a look at powernowd’s code, and it turns out it’s written quite straightforwardly. Within 30 minutes of tinkering, I had a patch that did what I wanted. With another 30 minutes, I polished it and made it quite commitable.

Basically, I added a new mode called “COOLING” to powernowd, which runs your CPU a few notches below your full frequency (which I call your “cool_spot”), based on the following approach:

• if you have two frequencies available, you normally run with the lowest.
• if you have three frequencies available, you normally run with the second from highest.
• if you have four frequencies available, you normally run with the third from highest.
• if you have five or more frequencies available, you normally run with the fourth from highest.
• if your load goes above your specified trigger (“highwater” in the code), you jump to highest frequency. When it lowers (“lowwater”), you go down to your cool_spot, but not below it.

On my machine, I have 5 frequencies (1.6Ghz, 1.5Ghz, 1.4Ghz, 1.2Ghz, and 400Mhz), and so I normally am running at 1.2Ghz. This new COOLING mode runs while I’m plugged in, and keeps my machine nice and cool but still lets it immediately respond when I want to do something, like a workspace switch.

I then hacked the init.d script to have a BATTERY and AC mode, and switch between AGGRESSIVE and COOLING modes accordingly. Now, when I’m unplugged, I get the best battery life and pretty good performance, and when I’m plugged in I get a cool notebook with good performance.

I’ll probably post the patch after my midterms…