Last time I tried to use one, I got lost. I had scroll around on its map to figure out where I was and a SANE way of getting where I wanted to be. I think I'll stick to paper maps, which actually help me get acquainted with where I'm going so I can concentrate on traffic more, thanks.
Sounds like a fun story. Care to share?
Personally, my navigation skills allows me to get lost in a phone booth, so I depend on my GPS to navigate anywhere I am not intensely familiar with. And while paper maps are fine, they are just not safe to use while driving. Besides, they take up too much space or are not detailed enough.
Actually, while ugly as sin, git gui is quite useful and intuitive. Works a charm for staging and managing branches, which are the 2 areas where I feel a GUI is most welcome.
I never do. I set up DHCP with static addresses for the known computers, and dynamic for the guests. So much easier to ssh between machines with proper ip addresses and names.
And... what's stopping you from doing that with ipv6?
Nothing. The minute the ISP's in this country supports ipv6, I'll be migrating. I am SO tired of being constrained to one freaking IP address.
Not to mention that IPv6 has no security whatsoever in its design. Any form of encryption is either a bolt on, or goes on a higher layer, such as how SSL and SSH ride on top of TCP. On the IP layer, there isn't any standard form of encryption.
If that is the case, could you please fix wikipedia? It has some mention of integrated ipsec in ipv6;)
He doesn't need to. He may want to. He has that option today.
You can assign IPv6 addresses manually to your heart's content as long as you have a block assigned to you, but for client machines there is rarely a reason to do this (just like how you normally don't go about handing out static IPs to every workstation, you set up a DHCP server (or many depending on the size of your organisation) and hand out dynamic addresses to most machines).
/Mikael
I never do. I set up DHCP with static addresses for the known computers, and dynamic for the guests. So much easier to ssh between machines with proper ip addresses and names.
I haven't counted, but I think I have *at least* 6 ISPs to choose from --- Not counting wireless, of course. North of Copenhagen, nothing special. And none of them seems to be able to deliver an unblocked port 25 (just inbound would be cool, I can relayed outgoing no problem). Sad, right?
That merely shows you are a bad Perl hacker. Trust me, Java programmers can make impenetrable code, too:) Just a 2 examples from real life: Why not have getStuff() update the object too? (Java has no const, otherwise this one might have been avoided.) Why use a small class holding 6 named doubles when you can just use an array of doubles and remember what is in each of them?
Are you really so totally uninformed about the world outside your own country?
Why the heck should any American care about Europe? You seem to think that the world should be interested in Europe like it is self evident, and honestly I see no reason to care at all. The Japanese make better stuff than you, the Chinese are cheaper than you, Israelis and Indians are smarter and the Mexicans make better food.
There's just no point to the old continent these days.
I'd say we should withdraw from NATO.
PS. Europe has a punishment system too. It's your bad TV and lousy music.
Why buy a PC if your console can do it? Last time I checked a Wii cost $250 US dollars. Xbox 360 cost $200. Never mind buying used. Hell if the requirements are so low why not a PS2?. What's stop the Big 3 from creating a newer, low power box for little of nothing?
Nothing, but exclusivity will go out of the window: every console of that type will be able to play everything the other cans. Thus, it will simply be the cheapest one that will win, and the service will get all the real money.
I have had thoughts myself along doing something along these lines, but rejected it as impossible due to latency. Will be interesting to watch, at least:) My secret dream is, of course, to see stuff like Monkey Island games again.
Because I want the garantee that the garbage is collected the moment the last reference is gone. I hope you understand why that garantee is critical
In that case, you want reference counting with cycle detection, rather than mark and sweep, but you do need garbage collection.
No. That would only be the case if I did not have a clear owner for my object... a rare case, and one that would be better addressed by mark and sweep.
The static analysis required to do this in any program involving passing pointers to shared libraries is not possible in the general case.
Really? Depends on what you *mean* with a pointer. There is no requirement that a pointer is a memory address for pointer arithmetics to be possible.
Runtime analysis of instrumented real-world programs (see the LtU archives for an interesting paper on this subject) has demonstrated that there is often a significant delay between the last time the contents of an object is accessed or modified and the time the last reference is destroyed in programs employing manual memory management. This is due to the fact that a developer has to be very conservative about deallocations, destroying the object after the last time it might be needed, rather than after the last time it is needed.
I think that just illustrates that releasing memory instantly is not a priority for most; a fact which makes GC a useful tool for memory management at all (unlike most other resources)
Surely, we can do better than Djikstra for GC? I haven't actually implemented as memory graph, but surely you can get a better heuristic than h(v)=0?
As a place to start, there's nothing better than Djikstra.
Of course there is. E.g, the a mark+sweep looks several times through the same graph... surely there is information that can be exploited in that, possibly using a derivate of A* or similar. Just guessing here, as I said I have never dabbled in that application of graph theory. I agree though that Djikstra is an excellent place to start: robust, easy to understand and performs reasonable.
For modern techniques I'd recommend looking at the work done at IBM's T J Watson lab in the last few years, in particular their paper on a unified theory of garbage collection. The same group has published a number of interesting papers since then, in particular a very efficient cycle detection algorithm and a concurrent version of the same.
Yeah, but how many cycles does a well-designed program have? Mine tends to "0". Thus I can design a very efficient cycle detection algorithm without those papers at all:P
That garantee is too expensive with a garbage collector, since every deallocation (or perhaps smarter, every allocation) would have to do a graph-walk of some kind.
If you read the paper I mentioned above, you'll see that you only need a graph walk when you decrement a reference counter and the don't free the object.
Eh, that was what I meant, except the bit about "not free the object..." which I just overlooked. Not what I wrote, I see. My apologies
You can defer this and in a later paper the team runs the cycle detector in a separate thread and in many cases this results in the object's reference count reaching zero before the cycle detector needs to run.
But that is just not good enough! I need that object deleted *now*, not sometimes later... something I get for *free* with stack based allocation, single ownership and shared ownership. And never have I needed anything besides those three (most common first). And considering a well-designed program needs to consider ownership *anyway* to avoid clashes and raceconditions among other things, GC gains me nothing but pain. Which is why I have moved away from GC-languages over to languages which support a richer set of allocation methods.
I'm still trying to understand how they're going to put Sony, Nintendo and Microsoft out of business. That was the line in the summary that confused me.
That, and the fact that the summary doesn't mention what OnLive or Gaikai actually do.
By making consoles obsolete since any browser will do. Yes, I, too, are skeptical. You'd need the data centers pretty close to get below 20-30 ms ping, which I think is the highest tolerable slack for most games (20ms ping ~ 50fps). That works out to 3000 km (back and forth) at the speed of light, which would be the ultimate barrier.
It might have, if it was a Linux application - which it wasn't.
We actually rolled our own memory debugging system, and it generally worked fine, but this particular thing slipped past (it was a nasty piece of code, having to do with lifetime of global statics and DLL load/unload order).
Yuck, global statics. More harm has been done with those than any amount of pointer gymnastics.
As for valgrind: I have a few times actually *ported* a windows app just to be able to run valgrind on it:) That tool seriously prods buttocks;)
Still, corruption bugs are always fun to catch. But as to my question, it seems a matter of the poster being imprecise; he meant that the ability to store pointer values outside pointers, so to be speak, prevented garbage collection. *shrug*. I can live without garbage collection, but not without RAII (though a full closure can fill some of that gap).
Ah, you are just imprecise. Pointer arithmetic does not mean that pointers can be cast to int; that is a whole other shebang
It's semantically equivalent. If you have pointer arithmetic, then your pointers are just another integer type. You can use them to store arbitrary integer values simply by assigning them to NULL and then adding an integer. Similar tricks let you go in the other direction. You can create a doubly-linked list by XORing pointers together without needing any conversion through int and you get a situation where accurate GC won't work (actually, neither will conservative GC, but there are simper arithmetic uses that will break accurate GC).
In C (e.g.), it would be a perfectly valid implementation if the the result of (NULL)+5 was aborting the program. Similar for the other tricks.
I'd prefer *at least* stack based allocation in addition
Why? Stack-based allocation is a trivial case of generational GC with the constraints that all references to the object allocated in the youngest generation must be weak (and your weak pointers are not automatically zeroed). Allocating on the stack is no cheaper than allocating in the young generation on most GCs (both are a single register-stored pointer increment) and is a lot less flexible than allocating in the young generation with an accurate GC (because the object can not be relocated if it needs to persist beyond the lifetime of the function).
Because I want the garantee that the garbage is collected the moment the last reference is gone. I hope you understand why that garantee is critical. That garantee is too expensive with a garbage collector, since every deallocation (or perhaps smarter, every allocation) would have to do a graph-walk of some kind.
It sounds like your understanding of garbage collection is limited to a poor comprehension of conservative GC. Please read something about accurate garbage collection, ideally something published in the last three decades, although maybe starting with Djikstra wouldn't be a bad idea.
lol. Surely, we can do better than Djikstra for GC? I haven't actually implemented as memory graph, but surely you can get a better heuristic than h(v)=0?
Why should they deliver a standards compliant browser? Might as well use Gecko, KHTML or one of its forks.
This is what I've thought too. Specifically, with all the buzz around WebKit, and its highly successful implementations, why don't they just add their Outlook extensions or whatever to a WebKit rendering engine? Because it's too close to open-source for their taste?
I'd bet on Gecko, not KHTML (or its fork, Webkit). And not yet, because Microsoft hasn't lost on the IE front yet. Moving now would be premature.
Perhaps I would even bet on a fork of Gecko. Why not?
If you have pointer arithmetic in a language, you can't have accurate GC. You can have conservative GC, but that generally has very poor performance (none of the tricks that accurate GC can use, like generations and rapid allocation work with conservative GC) so people rarely use it. This means that the presence of pointer arithmetic in a language is the indirect cause of all dangling pointer bugs and most memory leaks.
Ah, you are just imprecise. Pointer arithmetic does not mean that pointers can be cast to int; that is a whole other shebang. Anyway, *that* problem could be solved simply by forcing the int->pointer conversion to be a conversion to a weak pointer. Problem solved:) For my money, pointer->int should be indefined without a implementation-specific cast (reinterpret_cast in C++ )
GC as the only memory allocation method sucks in any case, since it makes it impossible to control resource allocation through object allocation. I'd prefer *at least* stack based allocation in addition, which is much safer than GC in any case, plus preferable single-ownership and shared, reference-counted ownership (both of which can garantee that the memory is reclaimed at the point where the owner is reclaimed). Bt I digress:)
You can either have software quality or you can have pointer arithmetic, but you cannot have both at the same time
You know, it is funny how pointer arithmetic always gets the blame. I have debugged tons of C++ code, and the pointer errors are nearly always one of null pointer derefence or access through invalid (now deleted or otherwise outdated) pointers/iterators. I have seen errors with pointer arithmetic, but they have been rare. Which is probably why checked pointer arithmetic (possible within the stand ard) has never caught on. I wonder why that bit is blamed? Because it is poorly understood, perhaps?
I have only read it at friends, so this is by memory, but the magazine is fairly large.. I'd say 40 pages is probably not too far off. It has adverts, but not overly many. Not many of the articles are american interest per se, but it is sometimes obvious that it is Americans writing the article (the silly units, e.g.).
or $5 you could buy an issue in the digital edition. Probably the cheapest way to evaluate it!
Slashdot Mirror
Last time I tried to use one, I got lost. I had scroll around on its map to figure out where I was and a SANE way of getting where I wanted to be. I think I'll stick to paper maps, which actually help me get acquainted with where I'm going so I can concentrate on traffic more, thanks.
Sounds like a fun story. Care to share?
Personally, my navigation skills allows me to get lost in a phone booth, so I depend on my GPS to navigate anywhere I am not intensely familiar with. And while paper maps are fine, they are just not safe to use while driving. Besides, they take up too much space or are not detailed enough.
Actually, while ugly as sin, git gui is quite useful and intuitive. Works a charm for staging and managing branches, which are the 2 areas where I feel a GUI is most welcome.
I never do. I set up DHCP with static addresses for the known computers, and dynamic for the guests. So much easier to ssh between machines with proper ip addresses and names.
And... what's stopping you from doing that with ipv6?
Nothing. The minute the ISP's in this country supports ipv6, I'll be migrating. I am SO tired of being constrained to one freaking IP address.
Not to mention that IPv6 has no security whatsoever in its design. Any form of encryption is either a bolt on, or goes on a higher layer, such as how SSL and SSH ride on top of TCP. On the IP layer, there isn't any standard form of encryption.
If that is the case, could you please fix wikipedia? It has some mention of integrated ipsec in ipv6 ;)
He doesn't need to. He may want to. He has that option today.
You can assign IPv6 addresses manually to your heart's content as long as you have a block assigned to you, but for client machines there is rarely a reason to do this (just like how you normally don't go about handing out static IPs to every workstation, you set up a DHCP server (or many depending on the size of your organisation) and hand out dynamic addresses to most machines).
/Mikael
I never do. I set up DHCP with static addresses for the known computers, and dynamic for the guests. So much easier to ssh between machines with proper ip addresses and names.
I haven't counted, but I think I have *at least* 6 ISPs to choose from --- Not counting wireless, of course. North of Copenhagen, nothing special. And none of them seems to be able to deliver an unblocked port 25 (just inbound would be cool, I can relayed outgoing no problem). Sad, right?
That merely shows you are a bad Perl hacker. Trust me, Java programmers can make impenetrable code, too :) Just a 2 examples from real life: Why not have getStuff() update the object too? (Java has no const, otherwise this one might have been avoided.) Why use a small class holding 6 named doubles when you can just use an array of doubles and remember what is in each of them?
Four years ago, the "standard" model of our car came with a casette player only. I was pretty amazed myself at that :) No air condition either.
Are you really so totally uninformed about the world outside your own country?
Why the heck should any American care about Europe? You seem to think that the world should be interested in Europe like it is self evident, and honestly I see no reason to care at all. The Japanese make better stuff than you, the Chinese are cheaper than you, Israelis and Indians are smarter and the Mexicans make better food.
There's just no point to the old continent these days.
I'd say we should withdraw from NATO.
PS. Europe has a punishment system too. It's your bad TV and lousy music.
I disagree. Mexican food is not that great.
Why is it that you think class action lawsuits are something found all over the world?
At least they're found in Denmark (Retsplejeloven chapter 23 a), so in this case, they may be entirely appropriate. Unlike your comment.
as of 1st of January, 2008.
Do note that the amount sought for is dkk 1000 (about 130 euro)+ the costs of justice, so Lenova is probably not going bankrupt over this.
Well, as long as you like living in 3rd world conditions anyways.
Technically, China is 2nd world --- but is still (barely) a developing country, according to wikipedia.
Why buy a PC if your console can do it? Last time I checked a Wii cost $250 US dollars. Xbox 360 cost $200. Never mind buying used. Hell if the requirements are so low why not a PS2?. What's stop the Big 3 from creating a newer, low power box for little of nothing?
Nothing, but exclusivity will go out of the window: every console of that type will be able to play everything the other cans. Thus, it will simply be the cheapest one that will win, and the service will get all the real money.
I have had thoughts myself along doing something along these lines, but rejected it as impossible due to latency. Will be interesting to watch, at least :) My secret dream is, of course, to see stuff like Monkey Island games again.
Because I want the garantee that the garbage is collected the moment the last reference is gone. I hope you understand why that garantee is critical
In that case, you want reference counting with cycle detection, rather than mark and sweep, but you do need garbage collection.
No. That would only be the case if I did not have a clear owner for my object... a rare case, and one that would be better addressed by mark and sweep.
The static analysis required to do this in any program involving passing pointers to shared libraries is not possible in the general case.
Really? Depends on what you *mean* with a pointer. There is no requirement that a pointer is a memory address for pointer arithmetics to be possible.
Runtime analysis of instrumented real-world programs (see the LtU archives for an interesting paper on this subject) has demonstrated that there is often a significant delay between the last time the contents of an object is accessed or modified and the time the last reference is destroyed in programs employing manual memory management. This is due to the fact that a developer has to be very conservative about deallocations, destroying the object after the last time it might be needed, rather than after the last time it is needed.
I think that just illustrates that releasing memory instantly is not a priority for most; a fact which makes GC a useful tool for memory management at all (unlike most other resources)
Surely, we can do better than Djikstra for GC? I haven't actually implemented as memory graph, but surely you can get a better heuristic than h(v)=0?
As a place to start, there's nothing better than Djikstra.
Of course there is. E.g, the a mark+sweep looks several times through the same graph... surely there is information that can be exploited in that, possibly using a derivate of A* or similar. Just guessing here, as I said I have never dabbled in that application of graph theory. I agree though that Djikstra is an excellent place to start: robust, easy to understand and performs reasonable.
For modern techniques I'd recommend looking at the work done at IBM's T J Watson lab in the last few years, in particular their paper on a unified theory of garbage collection. The same group has published a number of interesting papers since then, in particular a very efficient cycle detection algorithm and a concurrent version of the same.
Yeah, but how many cycles does a well-designed program have? Mine tends to "0". Thus I can design a very efficient cycle detection algorithm without those papers at all :P
That garantee is too expensive with a garbage collector, since every deallocation (or perhaps smarter, every allocation) would have to do a graph-walk of some kind.
If you read the paper I mentioned above, you'll see that you only need a graph walk when you decrement a reference counter and the don't free the object.
Eh, that was what I meant, except the bit about "not free the object..." which I just overlooked. Not what I wrote, I see. My apologies
You can defer this and in a later paper the team runs the cycle detector in a separate thread and in many cases this results in the object's reference count reaching zero before the cycle detector needs to run.
But that is just not good enough! I need that object deleted *now*, not sometimes later... something I get for *free* with stack based allocation, single ownership and shared ownership. And never have I needed anything besides those three (most common first). And considering a well-designed program needs to consider ownership *anyway* to avoid clashes and raceconditions among other things, GC gains me nothing but pain. Which is why I have moved away from GC-languages over to languages which support a richer set of allocation methods.
Sure, but why buy a console if you PC can do it? At least, why buy an expensive console with tons of CPU power of a meager one will do?
What makes you think the code is "mostly KHTML" at this point?
It's open source, eh? You can look at both repositories. They are still quite alike, except the ecmascript engines.
Also, are you still running XFree86, then?
Actually, I just call it X --- imprecise, I know, but I haven't dabbled much in that part of the software stack.
I'm still trying to understand how they're going to put Sony, Nintendo and Microsoft out of business. That was the line in the summary that confused me.
That, and the fact that the summary doesn't mention what OnLive or Gaikai actually do.
By making consoles obsolete since any browser will do. Yes, I, too, are skeptical. You'd need the data centers pretty close to get below 20-30 ms ping, which I think is the highest tolerable slack for most games (20ms ping ~ 50fps). That works out to 3000 km (back and forth) at the speed of light, which would be the ultimate barrier.
Surely valgrind could have spotted that for you?
It might have, if it was a Linux application - which it wasn't.
We actually rolled our own memory debugging system, and it generally worked fine, but this particular thing slipped past (it was a nasty piece of code, having to do with lifetime of global statics and DLL load/unload order).
Yuck, global statics. More harm has been done with those than any amount of pointer gymnastics.
As for valgrind: I have a few times actually *ported* a windows app just to be able to run valgrind on it :) That tool seriously prods buttocks ;)
Surely valgrind could have spotted that for you?
Still, corruption bugs are always fun to catch. But as to my question, it seems a matter of the poster being imprecise; he meant that the ability to store pointer values outside pointers, so to be speak, prevented garbage collection. *shrug*. I can live without garbage collection, but not without RAII (though a full closure can fill some of that gap).
As long as the code is mostly KHTML, I prefer calling it that, forks notwithstanding. That is why...
Ah, you are just imprecise. Pointer arithmetic does not mean that pointers can be cast to int; that is a whole other shebang
It's semantically equivalent. If you have pointer arithmetic, then your pointers are just another integer type. You can use them to store arbitrary integer values simply by assigning them to NULL and then adding an integer. Similar tricks let you go in the other direction. You can create a doubly-linked list by XORing pointers together without needing any conversion through int and you get a situation where accurate GC won't work (actually, neither will conservative GC, but there are simper arithmetic uses that will break accurate GC).
In C (e.g.), it would be a perfectly valid implementation if the the result of (NULL)+5 was aborting the program. Similar for the other tricks.
I'd prefer *at least* stack based allocation in addition
Why? Stack-based allocation is a trivial case of generational GC with the constraints that all references to the object allocated in the youngest generation must be weak (and your weak pointers are not automatically zeroed). Allocating on the stack is no cheaper than allocating in the young generation on most GCs (both are a single register-stored pointer increment) and is a lot less flexible than allocating in the young generation with an accurate GC (because the object can not be relocated if it needs to persist beyond the lifetime of the function).
Because I want the garantee that the garbage is collected the moment the last reference is gone. I hope you understand why that garantee is critical. That garantee is too expensive with a garbage collector, since every deallocation (or perhaps smarter, every allocation) would have to do a graph-walk of some kind.
It sounds like your understanding of garbage collection is limited to a poor comprehension of conservative GC. Please read something about accurate garbage collection, ideally something published in the last three decades, although maybe starting with Djikstra wouldn't be a bad idea.
lol. Surely, we can do better than Djikstra for GC? I haven't actually implemented as memory graph, but surely you can get a better heuristic than h(v)=0?
Why should they deliver a standards compliant browser? Might as well use Gecko, KHTML or one of its forks.
This is what I've thought too. Specifically, with all the buzz around WebKit, and its highly successful implementations, why don't they just add their Outlook extensions or whatever to a WebKit rendering engine? Because it's too close to open-source for their taste?
I'd bet on Gecko, not KHTML (or its fork, Webkit). And not yet, because Microsoft hasn't lost on the IE front yet. Moving now would be premature.
Perhaps I would even bet on a fork of Gecko. Why not?
If you have pointer arithmetic in a language, you can't have accurate GC. You can have conservative GC, but that generally has very poor performance (none of the tricks that accurate GC can use, like generations and rapid allocation work with conservative GC) so people rarely use it. This means that the presence of pointer arithmetic in a language is the indirect cause of all dangling pointer bugs and most memory leaks.
Ah, you are just imprecise. Pointer arithmetic does not mean that pointers can be cast to int; that is a whole other shebang. Anyway, *that* problem could be solved simply by forcing the int->pointer conversion to be a conversion to a weak pointer. Problem solved :) For my money, pointer->int should be indefined without a implementation-specific cast (reinterpret_cast in C++ )
GC as the only memory allocation method sucks in any case, since it makes it impossible to control resource allocation through object allocation. I'd prefer *at least* stack based allocation in addition, which is much safer than GC in any case, plus preferable single-ownership and shared, reference-counted ownership (both of which can garantee that the memory is reclaimed at the point where the owner is reclaimed). Bt I digress :)
More likely, they will just drop it. Why should they deliver a standards compliant browser? Might as well use Gecko, KHTML or one of its forks.
You can either have software quality or you can have pointer arithmetic, but you cannot have both at the same time
You know, it is funny how pointer arithmetic always gets the blame. I have debugged tons of C++ code, and the pointer errors are nearly always one of null pointer derefence or access through invalid (now deleted or otherwise outdated) pointers/iterators. I have seen errors with pointer arithmetic, but they have been rare. Which is probably why checked pointer arithmetic (possible within the stand ard) has never caught on. I wonder why that bit is blamed? Because it is poorly understood, perhaps?
I have only read it at friends, so this is by memory, but the magazine is fairly large.. I'd say 40 pages is probably not too far off. It has adverts, but not overly many. Not many of the articles are american interest per se, but it is sometimes obvious that it is Americans writing the article (the silly units, e.g.).
or $5 you could buy an issue in the digital edition. Probably the cheapest way to evaluate it!