The whole reason GNU has been forgotten is because it has such a dumb-ass name, complete with recursive acronymity (giggle-snort) and painfully nerdy pronunciation instructions.
OTOH, Linux is a cool name (because it contains an X), and Linus doesn't care how people pronounce it.
This is only half-joking. I think the name might really have something to do with it.
Hard-to-optimise code will generall not do much worse than unoptimised code (unless you go out and write something specifically to kill the machine).
No, hard-to-optimize code will usually run much faster than unoptimized code. Have you ever seen unoptimized code? It's ridiculous.
Yes, unoptimized code will give you a lower bound on performance, but it's nothing like a useful lower bound that should be used to reach any conclusions.
No, unoptimised code is not exact, but it is a hell of a lot better than looking at optimised code. In fact it gives a pretty good estimation of which cluster to run on.
I must not be understanding your claim here, because to me this seems totally bogus. What exactly are you measuring? Are you running unoptimized code on all three clusters and comparing the results? I'm pretty sure the results will be meaningless. Any resemblance to the actual relative or absolute performance of optimized (even poorly-optimized) code on the same three clusters will be mostly coincidental. You might as well just compare the CPUs' clock rates.
You're going to have to trust me when I say that the performance of unoptimized code is not relevant, because there's just no way to deal with such a complex topic in the space of a Slashdot post. Unoptimized code is designed to be correct and simple to generate, and it bears very little resemblance to hard-to-optimize code.
If you want to know how hard-to-optimize code will perform, you probably need to find some and test it.
...not just gcc, but both with all optimizations on, with all off, both with the best compilers, worst compilers, and middle of the road.
That doesn't make any sense at all. There's no point comparing compilers with optimizations turned off because the performance of unoptimized code is completely divorced from reality. When translating to an Internal Representation (IR), the front-end typically makes no effort whatsoever to produce efficient code because the optimizer will clean it up anyway. That means two front ends, or two versions of the same front-end, could produce substantially different IR for the same code, and that doesn't matter in the least.
In fact, it's conceivable that the front-end could intentionally produce sub-optimal code in order to "normalize" the code to make subsequent passes simpler. For instance, it could emit "a*2" instead of "a+a" because later optimizations might be pattern-matching on multiplication by a constant.
Furthermore, besides testing the most popular compiler, and the fastest compiler, what's the point in testing any other compiler?
Well, that's pretty good, but it's not the whole story. Some companies modify GPL'ed code for internal use, and the GPL allows that. If an employee were to redistribute that code outside the company, they could be in breach of their employment contract, and justifiably so.
Likewise, is there any reason that SCO could not write up a contract with a customer requiring that the customer not redistribute the code? If so, why is this situation different from the employee's situation?
So the question is whether they are still distributing it to customers. If they are not still distributing it, having stopped when they found out about the offending code, then I think they can tell their customers "go ahead and keep using it, but don't distribute it" and their legal position is still consistent with the GPL.
This guys says a number of crazy things, but this one takes the cake:
Both Apple and Dell are guilty of using misleading prices. For example, Apple gives the price of the low-end G5 as "$1999", and the high-end G5 as "$2999". In other words, they have subtracted $1 from a $3000 computer to make it seem cheaper, which is absolutely ridiculous. This demonstrates that both Apple and Dell are willing to mislead people when stating their prices.
I'm speechless. This statement is completely absurd.
There are some simple questions:
Do they know "their" code is in a GPL'ed product?
Are they still distributing this product?
It still doesn't matter. I can modify GPL'ed code, and I can distribute it to a select group of people. So long as I give them the source code along with the binaries, it's all good.
If you're claiming the GPL has some kind of "if you give it to someone, you must give it to everyone" clause, then please post it. I can't find it.
Right. And SCO's claim is that their code was released without their knowledge or permission, by someone who does not hold the copyright, so the GPL doesn't really enter into it.
The interpretation that "the GPL holds that IP SCO allows to be used by a select few must be freely released to any and all" is simplistic at best, and I think it's specifically, explicitly incorrect. The GPL says that whomever you release binaries to should also get the source; I'm pretty sure it does not say that if you give someone the code, you have to give it to everyone.
If someone is going to make a claim like that, they ought to quote the relevant clause from the GPL.
Why would you use the same ratio? That's meaningless. The right formula is that the number of bits is log2(N) where N is the number of encodings, assuming the encodings are equally likely. log2(1536) = 10.6, and log2(128) = 7.
The ascii table doesn't "encode" any information. It's a mapping from various symbols to integers. It would be misleading to say that each symbol "conveys 128 separate units of information".
Maybe it conveys one unit of information, or seven units of information (if your unit is a bit), but conveying 128 units of information in one ascii character would require very redundant input data.
It's just the difference between two different meanings of the word "symbol". Yes, 'a' is a symbol, so there are 26 different letter symbols. But supposing I have some device that can represent 26 distinct states, it would be incorrect to say it can "convey 26 separate units of information". It can only convey the choice of one symbol chosen from 26, so it's only conveying one unit of information. (That unit happens to be 4.7 bits, assuming the symbols are equally likely.)
Not 1500 units of information
on
Incas Used Binary?
·
· Score: 5, Interesting
After multiplying the different permutations of the knots, they reach the conclusion that there are 1536 possibilities, and then go on to state that "This could mean the code used by the makers allowed them to convey some 1,536 separate units of information".
That is a poor interpretation. 1536 possibilities allows someone to encode 10.6 bits of information. To encode 1536 "separate units" of information, each unit must represent no more than 1/145th of a bit. That is a very, very small amount of information, equivalent to having someone tell you something you were already 99.5% sure was true, such as "wow, this poker hand is not a straight!" or "guess what, my birthday this year does not fall on Friday the 13th".
It may be closer to the truth to say their knot language had 1536 different symbols, as compared with the 50-or-so letters, numbers, and punctuation marks we use in English.
Nonsense. Your average iguana does all these things with a brain the size of a grape. You certainly don't need 10% of the human brain to walk around and smell things.
I don't think you have any idea how much brain power reasoning requires.
Z doesn't have to prove all programs (including the one you describe) correct in order to be useful. It only has to prove my program correct to be useful to me.
I'm going to guess you don't write software for a living, right?
Open standards... I.E. Tell me the frick how your files are saved from your program! It doesn't hurt, hell it don't even tickle. and it does nothing but help everyone.
Except that once you publish your file format, you can't modify it anymore. This is a nightmare.
It's the same principle behind "private" fields in Java.
Slashdot Mirror
I'm pretty sure 1 and 2 are already true, aren't they?
OTOH, Linux is a cool name (because it contains an X), and Linus doesn't care how people pronounce it.
This is only half-joking. I think the name might really have something to do with it.
Yes, unoptimized code will give you a lower bound on performance, but it's nothing like a useful lower bound that should be used to reach any conclusions.
I must not be understanding your claim here, because to me this seems totally bogus. What exactly are you measuring? Are you running unoptimized code on all three clusters and comparing the results? I'm pretty sure the results will be meaningless. Any resemblance to the actual relative or absolute performance of optimized (even poorly-optimized) code on the same three clusters will be mostly coincidental. You might as well just compare the CPUs' clock rates.If you want to know how hard-to-optimize code will perform, you probably need to find some and test it.
In fact, it's conceivable that the front-end could intentionally produce sub-optimal code in order to "normalize" the code to make subsequent passes simpler. For instance, it could emit "a*2" instead of "a+a" because later optimizations might be pattern-matching on multiplication by a constant.
Furthermore, besides testing the most popular compiler, and the fastest compiler, what's the point in testing any other compiler?
Hey, we didn't all fall for it. :-)
You win.
Likewise, is there any reason that SCO could not write up a contract with a customer requiring that the customer not redistribute the code? If so, why is this situation different from the employee's situation?
So the question is whether they are still distributing it to customers. If they are not still distributing it, having stopped when they found out about the offending code, then I think they can tell their customers "go ahead and keep using it, but don't distribute it" and their legal position is still consistent with the GPL.
If you're claiming the GPL has some kind of "if you give it to someone, you must give it to everyone" clause, then please post it. I can't find it.
The interpretation that "the GPL holds that IP SCO allows to be used by a select few must be freely released to any and all" is simplistic at best, and I think it's specifically, explicitly incorrect. The GPL says that whomever you release binaries to should also get the source; I'm pretty sure it does not say that if you give someone the code, you have to give it to everyone.
If someone is going to make a claim like that, they ought to quote the relevant clause from the GPL.
Why would you use the same ratio? That's meaningless. The right formula is that the number of bits is log2(N) where N is the number of encodings, assuming the encodings are equally likely. log2(1536) = 10.6, and log2(128) = 7.
Maybe it conveys one unit of information, or seven units of information (if your unit is a bit), but conveying 128 units of information in one ascii character would require very redundant input data.
It's just the difference between two different meanings of the word "symbol". Yes, 'a' is a symbol, so there are 26 different letter symbols. But supposing I have some device that can represent 26 distinct states, it would be incorrect to say it can "convey 26 separate units of information". It can only convey the choice of one symbol chosen from 26, so it's only conveying one unit of information. (That unit happens to be 4.7 bits, assuming the symbols are equally likely.)
That is a poor interpretation. 1536 possibilities allows someone to encode 10.6 bits of information. To encode 1536 "separate units" of information, each unit must represent no more than 1/145th of a bit. That is a very, very small amount of information, equivalent to having someone tell you something you were already 99.5% sure was true, such as "wow, this poker hand is not a straight!" or "guess what, my birthday this year does not fall on Friday the 13th".
It may be closer to the truth to say their knot language had 1536 different symbols, as compared with the 50-or-so letters, numbers, and punctuation marks we use in English.
So what? So they had two tiny little brains. Your parent post's point still stands.
I don't think you have any idea how much brain power reasoning requires.
Z doesn't have to prove all programs (including the one you describe) correct in order to be useful. It only has to prove my program correct to be useful to me.
You can't really think the people working on formal verification are that stupid.
Sheesh, you old-timers and your premature optimization.
Brilliant. That approach is guaranteed to work so long as this is the only misspelled word in the Linux comments.
It's the same principle behind "private" fields in Java.