I get similar performance after letting FF8 run too long, but my usage pattern is a poster child for the problems with memory and CPU bloat. I open lots of tabs, including Facebook and Gmail, and some heavy Javascript sites required by my job. I leave them open sometimes for days. My only plug-ins are noScript and the JRE. Under my usage pattern, FF often reaches a 400MB footprint.

Now add to that the fact that my home computer is a little MSI Wind U100. 1GB RAM, 1.6GHz Atom processor (2-stage pipeline), and the thrashing is unbearable. Eventually I have to give in and close / restart FF.

I credit most of my troubles to the heavy javascript sites, not FF. I suspect that Facebook in particular is caching massive amounts of data and actually claiming that it uses that memory. At some point I should compare to Chrome to find out, but that's a project from some day when I have the time.

That's what happened here in the USA. Remember, prohibition led to the rise of organized crime in the USA. When we repealed it, organized crime didn't go away - it turned to other illegal activities to survive. Rooting it out was a multi-decade process, still on-going, and a long way from complete. The situation isn't just analogous, it's almost identical: the Cartels have power, and they won't simply let it go. They'll try to hold onto it.

Regarding the assign operation inside the if condition: I believe Java flags it if you do that.

Close. Java doesn't allow implicit casts from int to boolean, even though they use the same data type in the JVM. The assignment "x=5" is also an expression returning the int 5, so that things like "x=y=5" will work. C/C++ treats all non-zero ints as the boolean value true, so "x=5" ==> "5" ==> true, so "if (x=5)" is "if (5)" is "if (true)".

In Java, that can't happen because 5 is an int, and the conditional expression must be a boolean. However, you can make this mistake in Java if you really want:

boolean a = true;
if (a = true) { ...; }

Mining in space will be important, but it's unlikely to be important or cost effective to get the materials back to Earth. Whatever is ultimately mined in space will likely stay in space to build things there.

Not so. It's easy to get things back to Earth once you get out there. Sure, it's very expensive to get there, but so long as you're prepared to stay for a long time you can amortize over the mission.

The delta-V you'd need to hit a 3-5 years delivery window from the asteroid belt to Earth would be pretty small. There's a small matter of the delivery being, shall we say, postage due. Bit of an, er, impact on the wallet. Not to mention the neighborhood. But it would get there!

Ah, I see it. You're quite right, and I see the point. Using pure set cardinality leaves the probability undefined, but this gives a workaround. I'd have some difficult reading to prove whether the original assertion is correct under measure theory, but I suspect it is.

And here's the poison pill to demonstrate how far I missed the boat: the same arguments I used apply to something as simple as picking a number x from [0,2] and computing the probability that x < 1. Obviously that's 0.5, but the technique I outlined would claim it's undefined.

I'm so glad I don't claim to make my living as a mathematician!

I get the benefit that I can incorporate other GPL'd software without violating their licenses. I'm releasing my source code, which technically satisfies the GPL. But it's useless to everyone else. I assumed that the GPL dealt with this one, but maybe not....?

Two arbitrary lines in a 2D plane will meet with probability 1.0.
Two arbitrary lines in 3D space will meet with probability 0.0.
(In each case, the exceptions are vanishingly few relative to the norm.)

From a mathematical point of view, this isn't actually true (although it is intuitive). The probability that two arbitrary lines in 2D space will meet is undefined. Going back to basics, there are an unbounded number of possible lines. Without loss of generality, select y=0. (We don't lose generality because we can translate, rotate, and scale the plane to make any other selection equivalent to y=0.) Then the probability that the second line selected meets the first is defined as the number of lines which meet it divided by the universe of lines. Both sets are unbounded, so the probability is undefined (infinity / infinity).

If you'd like to resort to an argument based on transfinite numbers, I don't think it's valid. But it still undercuts the statement. It turns out that the two sets are of equivalent (transfinite) cardinality[1], so I suppose you could argue that the probability is still 1.0. But the same argument applies to arbitrary lines in 3D, 4D, or N-D space[2], so those are also 1.0.

Anybody see any problems there? I'm not a serious math geek, just a professor of computer science.

[1] The set of lines y=mx+b which don't meet y=0 are all pairs m=0, b!=0. The cardinality of that set is equal to the cardinality of the reals. The lines which do meet y=0 are everything else, and has cardinality equal to the set of all pairs of reals. It's fairly straightforward to show by, e.g., interleaving digits of our pair values, that there's an injective function from the pairs of reals to the reals. By inclusion, there's also an injective function from the singleton to the pairs. By the Cantor-Berstein-Schroeder Theorem, this means that a bijective function exists and the sets are of equivalent cardinality.

[2] Lines in N-D space are defined by a set of N real coefficients. We've already shown that the cardinality of the set of reals is equal to the cardinality of the set of pairs of reals. By induction on the number of coefficients, the cardinality of lines in N-D space is also the same as the cardinality of the reals.

This brings up a subtle point of the GPL which I've never researched. Does anybody know how the GPL handles the following hypothetical scenario?

Suppose I write software in a proprietary language. Using a non-free compiler, I can build binaries -- whether they're executables or some other source language, the GPL considers them binaries. If I release the source and the binaries under the GPL, but not the compiler, I presume that this doesn't suffice for the GPL. (Ignoring the fact that there's no copyright infringement if I'm the sole author.) People can't modify the source and build new versions because they don't have the compiler.

At the same time, we don't require GPL distributions to include everything you need to compile, or almost every package would need to include gcc.

Common sense implies that it's OK so long as the compiler is also freely available somewhere. Is there a clause to that effect in the GPL? I've never found it.

RMS is not infallible. Mind you, I think he's right in this case, but simple appeal to authority doesn't really prove anything.

Since I'm pretty sure RMS is right this time, I think the use of the discussion here is to provide education to those with flawed interpretations.

Ultimately, this sounds a lot more like people that are mad about no longer being able to endanger other people's lives by running red lights without having to worry about getting caught. I keep see the increased incidents of rear end collisions referenced as to why these cameras are bad, but ultimately, in much of the country we have tailgating laws for a reason, people are supposed to keep plenty of space so that they don't have an accident if the driver ahead suddenly applies the breaks. And rear end collisions are significantly less likely to result in death or serious injury than t-bone collisions, the ones that they're trying to prevent.

If I'm to be completely honest with myself, my gripe is that I believe that the cameras enforce the lights too precisely, with zero leeway. This forces me to be less safe as a driver. If I'm coming to a light and it turns yellow, sometimes I can see it's safer to scoot through it than to screech to a halt. With a camera, I'm a lot less willing to take the chance of a ticket. I should usually have enough time to make it on yellow, but we don't exactly publish the yellow light durations. A number of studies have shown that the yellow light durations are often much shorter than they should be.

It bothers me that I'm more likely to get a ticket by doing what I believe is safer.

On that topic, increased yellow durations are one of the few things which has been shown to reduce accidents with no real trade-offs. But that would require governance by people who actually look at real data and don't need the public to see them "doing something about XYZ problem."

A lot of people seem to want to scream about censorship, but they're missing the point. This is one of the best case scenarios I've seen in relations between companies and security researchers.

For those who can't be bothered to RTFA, here's a summary.

Researchers found a serious flaw. The company developed a fix. It turned out that the fix was flawed. The company told the researchers about the potential impact of giving the talk before the flaw was fixed, and the researchers voluntarily postponed the talk while a better fix is built.

That's it, and it looks like everybody did the best thing they could. Isn't this what we'd want Siemens to do? "You've got a right to give your talk, but we'd like you to postpone it. Here's why. Your call."

My bet is this fuck up will cost her the real case.

Unlikely. I doubt any of this screw-up will be admissible in the real case, as it is irrelevant to the matters under trial there.

What interests me is the results of the "oops" case. A lot of that depends on whether service was properly carried out. If so, then the $11m judgment will probably all go away, but it'll be difficult to get much in the way of sanctions or a successful countersuit. When served for a trial, you need to answer it! Absent a defendant, the judge acted properly. There are no disputed facts, so the plaintiff's assertions are more or less taken as gospel.

If service was improperly carried out, then sanctions are probable, and the judgment goes away instantly.

I remember what Prof. Alan Meltzer (Physics, RPI) said on the matter of using his own textbook. He made a personal point of never using a book he'd written. Either the book should cover everything he wanted the class to cover, and then his lecture is superfluous; or the lecture is necessary to supplement the book, and he wrote a poor book.

I'm not really sure I agree with him, since there's plenty of value in classroom examples which don't belong in a textbook. Nevertheless, the thought bears consideration.

1 April, 2040:

Three-digit Slashdot ID sells at auction for $240,000!