Magus's Grimoire [Abridged]

Language choice kind of matters not at all to most projects, so the question in my last post is really about what's worth learning to broaden my horizons. It's also a question that's limited to a pretty strict subset of my followers. Here's one explicitly for the bikeshed crowd.

My goal with manakin is to have a totally awesome UI. So, an open poll to help me brainstorm for what I want it to do and get me excited about it:

What features would make for an awesome UI in your opinion? What features have you seen in an existing client and loved. What misfeatures have you hated? What do you really wish your client could do but you've never seen?

Features which don't apply to the terminal still welcome. If you're aware of limitations on scope of a feature (e.g. only makes sense on a SmartPhone), that doesn't make them uninteresting, but mentioning that would be great.

So after my recent experiences with system-wide resource exhaustion thanks to Chromium, I've started paying a bit more attention. This thing gobbles up available system memory like there's no tomorrow. Ok, sure, no point it sitting around idle. Only it's not sitting around idle, because my OS dynamically uses available memory for cache.

Ok, sure, I've actually been somewhat sold on "1. Don't Optimize. 2. Optimize Later.", but seriously, what the hell is this thing doing with all this memory? Yes, the sandboxed multi-threaded model does mean a higher memory footprint, and I really appreciate that I can kill the thread for libflashplayer without bringing the whole browser down, considering that libflashplayer is a piece of junk that crashes on average twice a day (more if I'm actually spending any amount of time using Flash). But still:

• Each extension gets its own thread. Cool. Why the hell is the minimum private memory footprint for an extension (Eye Dropper) over 7 MB? What the hell does it need 7MB for? (Note: actually using it makes it jump over 16MB, then fall down to 15MB.) For a color picker + eyedropper! Yes, yes, process overhead. 7+MB of process overhead?! And Chrome is supposed to work on phones?

• The memory footprint for active tabs is comparably appalling. A process containing only the "new tab" page needs 10 MB. In fact, a process containing only a tab on a 73B file containing minimal HTML for a blank page with the title "Nothing" requires 14 MB! (6 MB for JavaScript?!)

• • A page which adds a link to that page is also 14 MB, but then having both pages open in the same process is almost 18 MB, so not only does each browser process have about 10 MB of overhead, but it takes about 4 MB just to have a blank page open, Changing the window size (and killing the thread and starting a new one) doesn't seem to change this significantly, so it doesn't even seem to be that it's allocating a bunch of memory for the display, unless it's doing so for some arbitrary viewport that's not the actual window size. (Actually, I haven't done the math closely enough, I guess it could be allocating exactly the size of my entire display here so I can resize/maximize quickly.

I guess that's probably what's going on. In order to give a snappier response, it's renders every page and keeps it in memory as a bitmap, but also does so as though it were fullscreened in case you change the window size? Which doesn't explain that extension overhead or what it's doing to keep the processor so busy even when I'm not doing much of anything.

I might even appreciate all that, if it worked to give me a responsive system. Except once I've been using it for a bit I'm not actually sure it does. And as mentioned previously, browsing Amazon gets pretty catastrophic. It should in theory be, say, swapping out memory for pages I haven't viewed recently, except it doesn't really seem to do that (until swap spontaneously decided to go from empty to full).

Not sure this is working...stay tuned...

• Once it's up and running, Chromium seems able to handle my c. 50 simultaneous tabs comparably to Firefox in terms of overall system performance, or at least not noticeably worse.

• Restarting with that many tabs is also comparable. I was hoping for an improvement.

• Actual browsing of the sort of thing I generally browse (which is mostly a lot of text with images) feels noticeably slicker. In particular, in Firefox I get a very noticeable delay when I change tabs or especially windows. In Chromium I can jump around between windows with a delay that is within reason (comparable to typing speed).

• Yesterday in Firefox I found the scribd webapp for the Prop 8 Ruling unusably slow. Here in Chromium today it's a bit pokey, but quite usable. I sincerely doubt it's a difference between yesterday and today.

• In-browser video seems to be inconsistent, perhaps depending on the underlying technology. YouTube videos actually seem slightly improved. Other sources (which I have no idea how different they are under the hood) seem choppier than I'm used to. Not sure what to make of this one.

• I haven't yet had the problem where Flash applications (such as streaming videos) have the audio get completely screwed up and I have to quit the browser to correct the problem. This may just be luck, though.

• I've tried two programs which are supposed to provide a similar experience to vimperator in Chromium: vrome and vimium. Both offer a really inconsistent user experience (not always working, frustratingly), and a strict subset of the features I want. It seems part of this may well be due to limitations of the Chromium extension architecture, which is, well, frustrating. I'd like more of an :ex-style interface, and the ability to hide the address-bar like Vimperator offers. Very annoyingly, both extensions don't seem to use keypresses chorded with Control correctly, and instead pass them through to the browser.

I'm an unrepentant vim user, eschewing other text editors which might have less terrible internal scripting languages. I love vimperator for Firefox, and I'm about to give Chrome a whirl and the first thing I'm going to do is install Vrome for it. I prefer crawl to nethack and nethack to ADoM (at least in UI terms). I passionately hate almost every "webapp" ever written.

I do almost all my work in the console. It's not actually that I prefer doing everything in a cell matrix containing characters in a monospaced font; in fact, for a lot of things, especially reading, I'm increasingly finding that proportional fonts really are advantageous. [Please don't ask me about antialiasing; I still haven't done a good enough side-by-side comparison to determine whether it results in "smooth" or "blurry".] It's just that console applications as a general rule are much more likely to have an input UI which allows a little more expressiveness than pointing with your finger and grunting. (To be fair, some GUIs do allow you to grunt, whine, or whistle while pointing.)

The goal of the input portion of a tool's UI is to effectively transmit information (what you want to do) from your mind to the tool. I think there are a few basic principles that should be kept in mind when designing or evaluating a UI.

As a designer of a tool, it's your job to think about how people will want to interact with that tool and make it as easy as possible for them to do so and get the most out of it. It's not good enough to do what's easiest for you to design, and it's not good enough to stop at only the features that will be easy to learn. For input, the tool should be able to capture relatively complex ideas in a simple way, and grow with the user. Users shouldn't have to ask, "I need to do this simple task about 20 times a day, why does it take 5 minutes to do each time?" At the worst, the answer should be because the user wasn't aware of the simpler way; it should never be because of an unnecessary limitation of the UI. Designers of database-interactive software, I'm looking at you, here.

I recently purchased a new wireless access point/router (Netgear WNDR3300 for $30 shipped from NewEgg; got $5 off a future purchase due to a minor defect [missing rubber foot], so final cost was $25), which arrived on Friday and I set about to set it up.

Part of the reason for the purchase is that, try as I might, my friends with Macintosh laptops have consistently had trouble connecting to my current Trendnet TEW-432BRP AP/router. In fact, at some point they went from an odd 10 minutes on/10 minutes off cycle to flat-out unable to connect. Weird.

So on Friday I asked kdsorceress to come over while I played with setting up the router. First I confirmed that she was able to connect to the default firmware, so I knew I had this as a fallback. Then I decided to play with installing DD-WRT, which is really cool. Confirmed that the Mac could still connect [though I've played with some settings since and will need to reconfirm].

So last night I got hit by one of those sporadic attacks against my main system that I get because I have ssh enabled, and decided to attempt to configure iptables on the router to rate-limit incoming ssh connections. This is a nice little trick to interrupt brute-force attacks on your password. For some reason, though, nothing I did seemed to work, or at least I couldn't trigger the dropping by attempting to connect from one of my computers to another of my computers.

Thinking the problem might be because all computers in question were inside the LAN, and might thus not be passing through the router even though I was telling them to [for some strange reason], I resolved to borrow a neighbor's wireless for a bit using my netbook so I could test an incoming connection that way. Unluckily, all the visible networks were protected with at least WEP.

Well, WEP is terrible security, I'm told, and there are plenty of packages out there to defeat it rapidly, so I thought maybe it was time to give one of those a try. The one I decide to try depends on another tool to log packets, and recommends kismet, so I go to install that.

Whereupon I discover the driver I'm using for the wireless card in the netbook [the proprietary Broadcom wl driver] doesn't support monitor mode, recommended to switch to the in-kernel driver. After some research, I discover that a very recent version of the b43 driver does support my card in the mode I want, but to make it work I both need to fetch a firmware update [fine], and update my kernel, because the version I'm using doesn't support my particular card even with the firmware update.

So now I'm recompiling the kernel on my netbook. To update the wireless driver. To sniff wireless network traffic. To break WEP encryption. To connect to my neighbor's wireless. To test that my router is properly protecting me from outside attacks.

But I don't know why she swallowed the fly.

[Note: Yes, this is probably technically computer trespass. In much the same way that cutting across the neighbor's lawn to get home is technically trespass. The amount of bandwidth I'll be using is completely negligible, and I won't be snooping around their network while I'm there.]

It turns out there's a bug in jlj, the program I use to compose messages in my favorite editor and submit them to LiveJournal and DreamWidth. I've finally tracked it down and written a quick fix. I feel slightly dirty, but the proper fix requires ripping out the existing band-aid on a band-aid and going back and fixing the underlying problem, which is a bit more effort than I want to go to for someone else's code at this time, as there could well be unexpected consequences.

The problem was that if you had written a post to go to a server other than LiveJournal, and you had it in your queue but it failed to post for some reason, then when you ran jlj -f to flush the queue it would attempt to log in to LiveJournal using the credentials for whatever site you had intended to post to. The symptom looked something like this:

[magus@cabinet ~]$ jlj -f
Logging in to server www.dreamwidth.org
Error Message returned:
Invalid password

( Read more... )

When did Dreamwidth sneak the "Quote" button into the comment editor?

I'm still not all that big on application-in-your-browser type stuff, but this gets points for making a tool that directly addresses a common task one might want to perform and makes it easier for the user. Go Dreamwidth devteam.

I just returned from my polling place. I feel I could have done a better job of being aware of local issues earlier and taking opportunities to discuss the candidates with the small number of people I know who'd be voting on the same tickets and gotten a better sense of the relative merits of the candidates. I did find the relevant debates for the local community access TV station and listen to them, however, and formed an opinion. It helped that for one position there were two candidates, one of whom came across as a single-issue candidate with a stance I was dubious about.

But what I really want to talk about is the voting process. I forget if I mentioned this last year, but it bears repeating: my Ward makes voting really simple and straightforward, in a way that gives me good confidence in the accuracy of the machine-tabulated results, and in a way that leaves a strong paper-trail which would be simple and straightforward to count manually should it be necessary.

You walk into the polling place. A person asks you for your address and name, locates it in the check-in book, and marks you as having entered. You can see all of this. You are handed a ballot in a "privacy sleeve".

You then walk to an empty booth, which is basically a table with walls dividing it into four parts, which you stand at while filling out your ballot. There is a black felt-tip pen there. You slide the ballot out of its sleeve, mark it with the marker, and slide it back into the sleeve.

To elaborate, marking the ballot consists of drawing a dark line from the tail to the head of the arrow pointing at the name of the candidate you wish to vote for. That's it. Using instruments [pen and paper] that most Americans are reasonably proficient in, in a straightforward way. If you want to see more about how that works, the ballot can, at least at present, be seen here. They've made it pretty hard to vote for the wrong person, and it seems to me that ambiguous ballots would be fairly rare. There are even clear instructions with examples provided at the top of the ballot.

Then you walk over to the exit area. You tell the person there your address and name, and they mark you in their book, just like when you came in. Then you walk over to the machine. It has a little green light showing it's ready, and a prominently displayed number which is presumably the count of ballots it has accepted. You feed your ballot out of the privacy sleeve and into the machine using the slot which runs most of the length of the sleeve. This is probably the most difficult part of the process, and I have to take on faith that they have people reasonably trained to assist without violating privacy if somebody needs help. It should be pretty doable. The slot doesn't line up with the arrows. The machine feeds in your ballot, makes a little noise, and another light lights up and you can see the counter increase. You're done.

Some highlights of the system:

• You mark the ballot directly, and can thus see that it is what you intend.
• A machine counts the votes, reducing needed manpower and potential for human error.
• There is a clear paper trail should a recount be called for.
• You don't need multiple computers to allow multiple people to fill out their ballots at the same time.
• Privacy can be readily offered without interfering with the effectiveness of the system [if you wanted more private booths fr marking the ballot, they could certainly be provided; ditto when you're feeding your single ballot into the machine].

I've heard tell of electronic voting machines with touch screens which silently accept input. I'd like to say, for the record, this is terrible design. I personally have a lot of trouble with touchscreen input for some mysterious reason presumably related to my skin chemistry. I also have very steady hands. But I would not trust that the input a touchscreen had accepted would be what I had intended: anything like this must have a feature which makes it clear to the user what input it received, not just that it received valid input. This is true for anything using a touchscreen for input.

The system we use here in Somerville Ward 6 might not be the best one out there, I don't know, but it's the best one I've ever personally used, and it at least passes basic tests of not totally sucking. Whoever was on that search committee, good work.

I need to learn that:

• Manual labor takes me longer than it used to/than I think it should, because I'm likely to work slower due to hand pain.
• A simple task like moving the thermostat through to the other side of the wall will leave me in frustrating amounts of pain, and also leave me feeling to exhausted to do anything but take an immediate nap.

gyges_kant, I'm still planning to come gaming tonight, but if I don't show, know it's because I wiped myself out making it so my room can be heated independently of the rest of the apartment.

I just had an annoying amount of difficulty doing an emerge -avtuND world.

The reason? I'm an idiot.

Note to self, and other Gentoo users: Never add anything to /etc/portage/package.unmask; it will come back to bite you later.

This is a slight lie: it's appropriate to add games-roguelike/nethack for manual unmasking. But in general, if you think this is the correct solution to your problem, you're probably wrong; think two more times.

Also, I'll point out that running ~arch can be viral: trying to run the ~arch version of net-www/mozilla-firefox, for example, results in about a dozen packages having to be added to package.keywords. This is a giant pain in the ass.

Question: is there a tool which will helpfully add a package to package.keywords and helpfully recursively add dependencies as necessary until the package will be compiled? Doing it by hand is really tedious, especially because emerge will only tell you the first package which is going to cause a problem, not all of them, so you can't even get a list. Meh.

Sometimes Gentoo is a right PITA. Also, I'm worried that tuomov may be right about the direction Linux is going, though I can't believe he's right about the solution. I think what he's putting up with by switching to Windows is worse than what I'm putting up with by sticking to Linux, which I guess shows we have different priorities despite generally agreeing about what sucks.