Even better is the Pixel Qi displays, which have a backlight switch. When off, the display is reflective and at least as easy to read as an e-ink display. With backlight on, it displays color video with usable refresh rates. I'm super-excited about ARM-based color multi-touch net-tablets with multi-day battery life when in E-book mode. Pop it into it's charging stand, and you have a netbook with wireless keyboard. Our family may need one for each of us.
Truth is, neither platform has enough revenue potential to make much sense for developers. If your $2 game is very successful, and sells 10,000, you've made 1/3 of a good senior software engineer's yearly salary. You'd have to make 3 of those a year! I prefer the model where the app is free. Generally, the free app leader has 10X the user base of the paid app leader. Make money on being a community leader, but give the software away for free (as in beer). What counts in this case is the user base. iPhone has the lead for now, but at the moment, I think Android has a higher growth rate.
I think long-term Google will be proven right: a universal operating system running on phones from all but a couple of the major cell phone manufacturers will create a far larger user base than proprietary offerings from any single vendor. It's Windows vs Mac all over again.
Sorry for this lame thought: One day, long after it as it becomes self-aware, the Internet will collapse under it's own weight into a black hole, becoming the creator of a whole new universe.
Re:Mod parent up!
on
Less Than Free
·
· Score: 2, Insightful
Unfortunately, a lot of this comes down to hardware support from cellphone manufacturers. Maemo 5 is open-source, and looks very very cool... However, the only phone that runs it is the N900, which is more of an old-fashion brick than a real cell phone.
I prefer Maemo, because I like to program in Qt4, rather than Java, and there are real advantages to a cell phone that runs Xorg, However, Android has momentum that I think is now unstoppable. If you want to hack for fun, go with the N900. If you want a customer base for your apps, go with iPhone first, and Android second. Long term, I suspect Android will even surpass the iPhone in user base. If you want to own an application niche, now is a good time to hop onto Android.
Bush Jr also canceled all the funding for fusion experiments. It's the only reason we ever even heard of the Bussard Polywell, since the scientists were free to talk about it after their contract with the Navy ended. Of course, now that the Navy funding is back, we're not allowed to hear how development is going.
The obvious conspiracy theory is Big Oil doesn't like the threat of an alternative energy source, and they have a lot of clout at the White House when Republicans are in power. Other Bush Jr decisions included halting nearly all new permits for solar array power stations. So, the conspiracy theory has legs.
I've written such an algorithm in the past, to solve a specific problem (placement of components in an IC). It's a ton of fun, though my program had poor results.
So, you're pondering interesting philosophical questions... If we could build a computer as creative and smart as a person, would it feel alive like we do? When we build the next version, should we grieve for the previous one when we turn it off? These are decades old questions now.
Anyway, lot's of geeks are fans of space exploration. I think that living here on Earth is literally coded into our genes. I think humans will never live anywhere else. However we currently send our machines to the stars. At the current rate of progress, I might live to see the day that we send evolving, self-replicating machines into space. I see those as the children who may have a chance of inheriting the stars. Hopefully, and I am an optimist here, those children will be alive. Hopefully they'll be kind enough not to kill us all.
So, if I were an intelligent entity (you can say God if that sounds close enough), and I had a finite source of quantum computation (but really really big), I'd want to do several things with that computation to get the biggest bang for my limited buck:
- I would never compute the position of every particle in the universe at every quantized point of time. Instead, I'd use an event-driven simulation, and only compute interactions between particles. Kind of like we see in quantum mechanics. - No point letting the universe seem grainy. I'd hide my integer based math with the Heisenberg Uncertainty Principle.
Of course, both of those are highly flawed ideas, however I see no proof that the universe is truly "infinite" in any way. God can use integers.
The hypothesis that no deity of any kind exists solves the problem in an unbeatably elegant fashion.
Early humans could feel their "aliveness" or "conscience" even before they could imagine math or science. They naturally assumed the simplest possible solution: some God who looks like them gave them a soul, which gives them this feeling of being alive. All these years later, we know so much about science and technology, but nothing about that feeling of being alive. It's there, and unexplained in any way so far. Without it, our lives would be simply meaningless computation. There's still some magic in the universe we need to explain. "no deity" as a refutation of the literal truth of the Bible is very logical. However, don't throw out the baby with the bath water - there is something magical about being alive, and cause to be "spiritual".
Thanks for the ABA example. It made very interesting reading. I agree that GC is a good solution to this difficult problem. Do you feel there are any more common problems programmers run across where GC is also helpful vs the C++ mechanisms? I can't think of any.
When I code, I use only two kinds of objects. "Value" objects typically live on the stack or as members of objects. These are things like strings and vectors. C++ deals with value objects more efficiently than GC. The other kind of object I call "database" objects. They live on the heap, and typically are involved in several relationships with other objects, like lists or hash tables. They are created with new statements, and freed with delete. GC does not help me here, since to remove an object from the database, I still need to call a method to clean up it's relationships. This method is nearly identical to a C++ destructor. In C++, when I have bugs in destructors, I get segvs due to dangling pointers, and usually I the debugger helps me find the problem quickly. In Java, memory just starts filling up. I'm not convinced one or the other is better.
Is there any major hole in this analysis?
Mostly I care because of efficiency, which of course is less of a problem for most applications over time. However, no GC I've seen comes close to the cache efficiency I get for most of my code. IMO, new languages should try to improve cache hit rates with clever memory organization. This is not possible in C++ because the memory layout is visible to the programmer. However, Java is so close...
Here's what I would do in Java to make it the fastest language in the world for memory intensive applications:
Each inner loop would be analyzed to determine what members where being accessed of what classes. I'd create a separate GC heap for each such set of members, and the other members of an class would live in a different heap. Rather than using 64-bit pointers, I'd use 32-bit indexes into these heaps to access data. That way, the same index can be used regardless of which heap contains a member - the organization of an object's heaps would all be identical. This would save a ton of memory, due to 32-bit indexes. It dramatically speeds up code due to cache performance gains. It's compatible with GC, though why the heck I want GC built-in to a language, I can't figure out. However, it does break casting a derived class to a base class. To access data, code has to know exactly the class type. This can also be dealt with, in various ways.
Regardless of whether a language has GC, I'd also automatically create destructors for the user. This feature has been in relational databases for years, and I also use it in DataDraw-based C code. It rocks. It nearly eliminates the memory leaks and dangling pointers that are so hard to debug. It's way better than just adding GC, IMO.
My in-memory data-base built in C using DataDraw does this already. The speed rocks. Before writing the first version of DataDraw back in 1991, I read about relation databases, and was somewhat inspired. Why we have these innovations in databases, but not computer languages, I can't guess.
GC is good for simple objects like strings and vectors that typically live on the stack. You never use new and delete with such objects, and instead treat them like integer values that go away when you leave scope. However, C++ handles these "value" objects without GC just fine.
The other type of object is where you call new and delete because the objects represent things in your database, like employees or graphs. For these types of objects, GC buys you nothing at all.
There's tons of room for improvement that's real, like automatically generating all my destructors for me, or analyzing my program and optimizing the memory layout for best cache performance. All GC does is turn an easy to find dangling pointer (segv in a debugger) into a hard to find memory leak.
In fact, good accurate garbage collections could be considered the most important development in programming languages since OO (and some even think more important)
Really? Who thinks GC is more important? GC is helpful only for simple classes with no relationships, like strings and vectors. For real objects in real databases, where all the real work is done, GC is useless. Here's the argument for why Java has GC. From the link:
Garbage collection relieves you from the burden of freeing allocated memory
Brilliant... instead of calling delete, I have to call a method that does the exact same thing as a destructor.
Knowing when to explicitly free allocated memory can be very tricky.
Seriously? You mean the line of code where I decide I want the object gone from memory isn't the right place to call the destructor? Somehow I'm better off not knowing whether or not an object in my database is still in use?
I mostly agree with you. I have in fact found a niche that is getting squeezed, but not as fast as it should be. These techniques are simple to automate compared to the optimizations that are already done, but for some reason, innovation in runtime efficiency seems to have evaporated in the last 10 years or so. It's probably because everyone sees the niche getting squeezed and moves on to some other problem.
A native Java compiler should be able to analyze my loops, and break up objects into well chosen chunks, and allocate those chunks together in memory. This alone would make it instantly the fastest computer language in the world for memory intensive applications.
I think language developers are also missing out on innovations that could greatly reduce bugs while simplifying coding. Garbage collection is good for what I call "value" objects like strings, and matrices. However, for normal objects representing things you'd typically see in a relational database, like "employee", you still have to write a destructor manually. The destructor will remove the employee from it's department and any other relationships with other objects before deleting it. GC doesn't help me at all in managing simple databases with objects like employees. If I forget to clean up a relationship, with GC I get a memory leak, while in C++ I get a dangling pointer. I wouldn't say one is better than the other.
Instead, the compiler should generate the destructor automatically for me, by analyzing the relationships and through some user declarations. This capability already exists in relational databases. Unlike GC, compiler-generated destructors save the coder time and nearly eliminates very painful memory leaks/dangling pointers.
I'm not just hypothesizing. DataDraw writes 100% of all my destructors when I'm programming in C. My little company has so few memory leaks and dangling pointers most programmers don't even know how to run valgrind. Mostly we use it to measure cache statistics.
As for the cache hit thing, it sounds quite suspicious to me, more details please.
BTW, I agree that > 90% of code doesn't need to run fast or memory efficiently. I am a fan of Python, but when I'm slinging around gigabytes of data, I use C. Very few programmers ever need this kind of efficiency. Anyway, check out this table half-way down. The "DataDraw" benchmark is still in C, but it uses carefully designed memory layout. The speed improvements are 100% due to cache hit rate improvements - 15X better than when using malloc for each object. Most slow memory-intensive programs waste most of their CPU time in L2 cache misses.
You want all the fields that are accessed in inner loops to be together in memory in dense arrays, and the other stuff should be somewhere else. So, for binary tree traversal, you want left and right and maybe some key value in a small structure, and all the other bloat needs to be elsewhere. Ideally, you allocate these structures which are nearby in the tree also nearby in memory. That way, when you get an L2 cache miss, the extra data you load into the cache likely contains data that will soon be accessed in the inner loop.
Garbage collectors typically don't bother allocating objects of the same type together. Careful management of memory can be a PITA, which is one reason I use the DataDraw tool for memory-intensive applications. It allows me to "cache together" fields, and puts wrappers around data access to hide the ugliness.
A larger group than "blind" will be visually-impaired... people who can probably see the book, but not the words. I'm facing such a fate myself soon. I came up with a similar idea, but as usual someone beat me to it. Just use your camera phone, and port some decent open-source OCR software to it.
In general, I'm not a fan of dedicated e-book readers. I think next year we'll see some killer multi-touch arm-based net-tablets with e-paper-like displays and battery life. I think the killer app will be the e-book reader. Twenty years from now, I'll be able to tell kids about the days when we had special devices for reading books, and that's what an e-book reader was, rather than a program. Remember when a "word processor" meant a dedicated and very expensive machine for editing text documents? Yes, I'm that old.
I think the niche it's aiming at is the, "Look! We made a programming language, too!" niche.
Agreed. I don't see a case where Go would be useful to me, given the Java, C++, Python, Perl, PHP, and C that I currently use.
My biggest rant about most of these new languages: garbage collection is useless! I still have to write destructors that clean up all the pointers to an object, and all garbage collection does is force me to call the destructor as a function, rather than a more clear 'delete' statement. Worse, it takes away my most powerful speed optimization tool: careful memory layout for best cache hit rates. I can write simple graph traversal code that is 10X faster in carefully designed C/C++ than in any garbage collected language for large graphs.
"Scantegrity uses a process called “zero knowledge” that allows skilled, independent auditors to verify that the codes result in votes going to the right candidates, without actually revealing an individual voter’s selections."
It's super-cool stuff every slashdot geek needs to know. So, this allows us to insure our vote was counted without enabling us to sell our votes. Very cool! However, it still not fool-proof. A friend of a friend of mine has gotten so worked up over an election that she went to the polls early, and often, and voted for her whole extended family. Without requiring photo-IDs, it's really easy to do. Every show up to a poll and see your name has already been crossed off?
E-book readers will be the killer app for next year's net-tablets, IMO. The good ones will likely use Pixel Qi screens. As for OLPC, they're doing great pioneering work, which launched the netbook phenomenon. Pixel Qi will provide OLPC the paper-like screens at cost. In e-book mode, the battery should last days, not hours, and with the overall reductions in cost for the multi-touch display, processor (can you say ARM?), and power system, net tablets for under $100 may just be possible. Frankly, I'll hold out for a $250 10" tablet that looks like Apple designed it, but runs Ubuntu Netbook Remix. It'll read like e-paper, but also can play HD videos.
Asterisk supports Skype [digium.com]. As does FreeSWITCH [freeswitch.org].
From your link: "The SFA product will be the only solution that integrates Asterisk directly with Skype." It costs something like $80, and is closed-source. With an open-source Skype client, I doubt Digium's closed-source solution will remain the only solution for long.
GP does get a key point: The software is key. Carriers (and freaking Verizon in particular) in the US simply refused to understand this. They build the network, while all the innovation is in the handsets and the software, but for some reason, US carriers seem to think they are the true innovators and handset providers are fluff. Now that Verizon has screwed up on handsets for three years straight, they finally realize that their strength is simply the network they build. Maybe they've been listening to their own adds. They're finally going to ship a modern phone, without screwing up the software first. Stupidity at Verizon may be going out of style.
Anyway, as said before on slashdot, Android vs iPhone is just like Windows vs Mac all over again. With Verizon on board, Motorola building 20 new Android phones next year, and 50 Android sets in the works around the world, Android is set to finally deliver on it's promise of unifying the software across a broad spectrum of handsets. There wont be any single iPhone killer, just as no single PC was ever a Mac killer. However, I see nothing that can stop Android from becoming the world's dominant smart phone OS.
The Motorola Droid isn't quite as exciting of a device as the Sony Xperia X10. I suspect we'll keep seeing Android based "iPhone killers" plunk away until Adroid wins the race.
We both must be about the same age - I'm 45. It just kills me that we have to have "play dates" for my kids to play with other kids, and my kids don't venture into the woods the way I liked so much as a kid. We agree that today's environment of fear is just that - pointless fear, driven by the media.
Anyway, some things are different today. My introduction to porn was sneaking peaks at my Dad's Playboy magazines, which he would read while Mom cleaned and cooked and held down a job. Dad's back then had it all - no poopy diapers, wives who did all the housework and had paying jobs, and who felt guilty if you didn't get enough sex...
Today, kids don't get that sneak-peek into porn when they finally become curious about sex. And, let's face it... Playboy had a sense of class and beauty missing from redtube.com. Instead, eight-year girls type "hot guy" into Google, and get hard-core video. Their intro into the idea of sex is likely going to be a foot-long dong butt-f*cking a teenager.
I took advice I got here on slashdot, and use the free opendns.com DNS filter. I also use addblock plus in firefox on all our computers. OpenDNS gives me some control over the content filiter - I use the low settings, only blocking phishing and hard-core porn. These tools are waaaaay better than anything the FCC might dream up. Instead of more government censorship, how about a program for training/educating parents, so we can all learn how to take advantage of the excellent, and free tools that already exist out there? Something as simple as requiring ISPs to send information packets about Internet filtering might do the trick. Perhaps requiring the installers who do house visits to train how to filter, not just how to use the DVR. All parents know how to record Pokemon. How many know how to protect their kids from googling "hot guys"?
There is really no excuse for piracy...despite what the Slashdot groupthink wants to force you to believe.
I'm going blind, and I already own all her books. I want to read them again, but J.K. Rowling refuses to offer screen-reader compatible e-books. That's not a good excuse for a visit to The Pirate Bay?
Yes, I've been practising for months, ever since I got off my butt and started looking into how to survive as a programmer who can't read or drive (I still can). I started off at about half that speed, easily understandable by pretty much anyone.
As a side benefit, I have to say I have never enjoyed books so much. I just finished listening to book 5 in the Harry Potter series (I read it years ago). It only took 8 hours. My reading speed has always been frustratingly slow.
For three years, in the late 90's, I couldn't type and had to program using speech. Now I'm learning to program while listening instead of looking. I think I'd love to write the conversational interface used in Star Trek... "Computer, read me the latest headlines on Slashdot!"
While your reply sounds a bit like flame-bait, I basically have to agree. The format isn't a universal binary that gets translated to each machine architecture when installed. Instead, it's basically an archive of pre-compiled binaries for each platform you support. So, for example, my stupid Qt application has to be compiled separately for Fedora and Ubuntu. This technology would in theory allow me to merge the binaries into a single FatELF binary. Personally, I'd rather just provide separate.deb and.rpm files.
However, the idea of a universal binary is cool. We could do something like the old p-Code, where we compile to a virtual architecture, and then translate it to the machine during installation. I liked the idea when they had it way back in the early 80's (late 70's?), and I was sad to see we didn't have the compute power back then to make it fly. I bet we do now.
If you double the speed, you double the frequency, which is a problem. There are programs that will fix that for you, but even so, about 50% speed increase is about what's practical, from what I've read. I've tried the wsola program on linux, and found that speeding up audio substantially distorts it. Are there better programs that allow for > 2X speed up?
In general, to get really high speed TTS (text to speech), you need a really good TTS engine. Alex on Mac isn't bad, but SFAIK, Eloquence (the old IBM Via Voice) is still king here. I've found no other TTS engine that comes close. A friend of mine listens to Eloquence at 850wpm, which is truly amazing.
In short, blind/VI users need access to the text. There are web sites that offer the blind free e-books, which is cool, but only authors who want to participate are available (not Rowling). As my blind friend says, "The Pirate Bay is your friend." That's just reality for the blind.
piracy helps to support a product- it enables a huge number of people to get involved with a product they otherwise would have no part of
Insightful. I often felt that Microsoft in the 80's and 90's encouraged students to pirate their software with lax copy protection, so that when we got into the workforce and could afford it, we'd get sucked into buying all their stuff. In my case, it worked.
Slashdot Mirror
Even better is the Pixel Qi displays, which have a backlight switch. When off, the display is reflective and at least as easy to read as an e-ink display. With backlight on, it displays color video with usable refresh rates. I'm super-excited about ARM-based color multi-touch net-tablets with multi-day battery life when in E-book mode. Pop it into it's charging stand, and you have a netbook with wireless keyboard. Our family may need one for each of us.
Truth is, neither platform has enough revenue potential to make much sense for developers. If your $2 game is very successful, and sells 10,000, you've made 1/3 of a good senior software engineer's yearly salary. You'd have to make 3 of those a year! I prefer the model where the app is free. Generally, the free app leader has 10X the user base of the paid app leader. Make money on being a community leader, but give the software away for free (as in beer). What counts in this case is the user base. iPhone has the lead for now, but at the moment, I think Android has a higher growth rate.
I think long-term Google will be proven right: a universal operating system running on phones from all but a couple of the major cell phone manufacturers will create a far larger user base than proprietary offerings from any single vendor. It's Windows vs Mac all over again.
Sorry for this lame thought: One day, long after it as it becomes self-aware, the Internet will collapse under it's own weight into a black hole, becoming the creator of a whole new universe.
Unfortunately, a lot of this comes down to hardware support from cellphone manufacturers. Maemo 5 is open-source, and looks very very cool... However, the only phone that runs it is the N900, which is more of an old-fashion brick than a real cell phone.
I prefer Maemo, because I like to program in Qt4, rather than Java, and there are real advantages to a cell phone that runs Xorg, However, Android has momentum that I think is now unstoppable. If you want to hack for fun, go with the N900. If you want a customer base for your apps, go with iPhone first, and Android second. Long term, I suspect Android will even surpass the iPhone in user base. If you want to own an application niche, now is a good time to hop onto Android.
Bush Jr also canceled all the funding for fusion experiments. It's the only reason we ever even heard of the Bussard Polywell, since the scientists were free to talk about it after their contract with the Navy ended. Of course, now that the Navy funding is back, we're not allowed to hear how development is going.
The obvious conspiracy theory is Big Oil doesn't like the threat of an alternative energy source, and they have a lot of clout at the White House when Republicans are in power. Other Bush Jr decisions included halting nearly all new permits for solar array power stations. So, the conspiracy theory has legs.
I've written such an algorithm in the past, to solve a specific problem (placement of components in an IC). It's a ton of fun, though my program had poor results.
So, you're pondering interesting philosophical questions... If we could build a computer as creative and smart as a person, would it feel alive like we do? When we build the next version, should we grieve for the previous one when we turn it off? These are decades old questions now.
Anyway, lot's of geeks are fans of space exploration. I think that living here on Earth is literally coded into our genes. I think humans will never live anywhere else. However we currently send our machines to the stars. At the current rate of progress, I might live to see the day that we send evolving, self-replicating machines into space. I see those as the children who may have a chance of inheriting the stars. Hopefully, and I am an optimist here, those children will be alive. Hopefully they'll be kind enough not to kill us all.
So, if I were an intelligent entity (you can say God if that sounds close enough), and I had a finite source of quantum computation (but really really big), I'd want to do several things with that computation to get the biggest bang for my limited buck:
- I would never compute the position of every particle in the universe at every quantized point of time. Instead, I'd use an event-driven simulation, and only compute interactions between particles. Kind of like we see in quantum mechanics.
- No point letting the universe seem grainy. I'd hide my integer based math with the Heisenberg Uncertainty Principle.
Of course, both of those are highly flawed ideas, however I see no proof that the universe is truly "infinite" in any way. God can use integers.
Early humans could feel their "aliveness" or "conscience" even before they could imagine math or science. They naturally assumed the simplest possible solution: some God who looks like them gave them a soul, which gives them this feeling of being alive. All these years later, we know so much about science and technology, but nothing about that feeling of being alive. It's there, and unexplained in any way so far. Without it, our lives would be simply meaningless computation. There's still some magic in the universe we need to explain. "no deity" as a refutation of the literal truth of the Bible is very logical. However, don't throw out the baby with the bath water - there is something magical about being alive, and cause to be "spiritual".
Thanks for the ABA example. It made very interesting reading. I agree that GC is a good solution to this difficult problem. Do you feel there are any more common problems programmers run across where GC is also helpful vs the C++ mechanisms? I can't think of any.
When I code, I use only two kinds of objects. "Value" objects typically live on the stack or as members of objects. These are things like strings and vectors. C++ deals with value objects more efficiently than GC. The other kind of object I call "database" objects. They live on the heap, and typically are involved in several relationships with other objects, like lists or hash tables. They are created with new statements, and freed with delete. GC does not help me here, since to remove an object from the database, I still need to call a method to clean up it's relationships. This method is nearly identical to a C++ destructor. In C++, when I have bugs in destructors, I get segvs due to dangling pointers, and usually I the debugger helps me find the problem quickly. In Java, memory just starts filling up. I'm not convinced one or the other is better.
Is there any major hole in this analysis?
Mostly I care because of efficiency, which of course is less of a problem for most applications over time. However, no GC I've seen comes close to the cache efficiency I get for most of my code. IMO, new languages should try to improve cache hit rates with clever memory organization. This is not possible in C++ because the memory layout is visible to the programmer. However, Java is so close...
Here's what I would do in Java to make it the fastest language in the world for memory intensive applications:
Each inner loop would be analyzed to determine what members where being accessed of what classes. I'd create a separate GC heap for each such set of members, and the other members of an class would live in a different heap. Rather than using 64-bit pointers, I'd use 32-bit indexes into these heaps to access data. That way, the same index can be used regardless of which heap contains a member - the organization of an object's heaps would all be identical. This would save a ton of memory, due to 32-bit indexes. It dramatically speeds up code due to cache performance gains. It's compatible with GC, though why the heck I want GC built-in to a language, I can't figure out. However, it does break casting a derived class to a base class. To access data, code has to know exactly the class type. This can also be dealt with, in various ways.
Regardless of whether a language has GC, I'd also automatically create destructors for the user. This feature has been in relational databases for years, and I also use it in DataDraw-based C code. It rocks. It nearly eliminates the memory leaks and dangling pointers that are so hard to debug. It's way better than just adding GC, IMO.
My in-memory data-base built in C using DataDraw does this already. The speed rocks. Before writing the first version of DataDraw back in 1991, I read about relation databases, and was somewhat inspired. Why we have these innovations in databases, but not computer languages, I can't guess.
GC is good for simple objects like strings and vectors that typically live on the stack. You never use new and delete with such objects, and instead treat them like integer values that go away when you leave scope. However, C++ handles these "value" objects without GC just fine.
The other type of object is where you call new and delete because the objects represent things in your database, like employees or graphs. For these types of objects, GC buys you nothing at all.
There's tons of room for improvement that's real, like automatically generating all my destructors for me, or analyzing my program and optimizing the memory layout for best cache performance. All GC does is turn an easy to find dangling pointer (segv in a debugger) into a hard to find memory leak.
Really? Who thinks GC is more important? GC is helpful only for simple classes with no relationships, like strings and vectors. For real objects in real databases, where all the real work is done, GC is useless. Here's the argument for why Java has GC. From the link:
Brilliant... instead of calling delete, I have to call a method that does the exact same thing as a destructor.
Seriously? You mean the line of code where I decide I want the object gone from memory isn't the right place to call the destructor? Somehow I'm better off not knowing whether or not an object in my database is still in use?
I mostly agree with you. I have in fact found a niche that is getting squeezed, but not as fast as it should be. These techniques are simple to automate compared to the optimizations that are already done, but for some reason, innovation in runtime efficiency seems to have evaporated in the last 10 years or so. It's probably because everyone sees the niche getting squeezed and moves on to some other problem.
A native Java compiler should be able to analyze my loops, and break up objects into well chosen chunks, and allocate those chunks together in memory. This alone would make it instantly the fastest computer language in the world for memory intensive applications.
I think language developers are also missing out on innovations that could greatly reduce bugs while simplifying coding. Garbage collection is good for what I call "value" objects like strings, and matrices. However, for normal objects representing things you'd typically see in a relational database, like "employee", you still have to write a destructor manually. The destructor will remove the employee from it's department and any other relationships with other objects before deleting it. GC doesn't help me at all in managing simple databases with objects like employees. If I forget to clean up a relationship, with GC I get a memory leak, while in C++ I get a dangling pointer. I wouldn't say one is better than the other.
Instead, the compiler should generate the destructor automatically for me, by analyzing the relationships and through some user declarations. This capability already exists in relational databases. Unlike GC, compiler-generated destructors save the coder time and nearly eliminates very painful memory leaks/dangling pointers.
I'm not just hypothesizing. DataDraw writes 100% of all my destructors when I'm programming in C. My little company has so few memory leaks and dangling pointers most programmers don't even know how to run valgrind. Mostly we use it to measure cache statistics.
BTW, I agree that > 90% of code doesn't need to run fast or memory efficiently. I am a fan of Python, but when I'm slinging around gigabytes of data, I use C. Very few programmers ever need this kind of efficiency. Anyway, check out this table half-way down. The "DataDraw" benchmark is still in C, but it uses carefully designed memory layout. The speed improvements are 100% due to cache hit rate improvements - 15X better than when using malloc for each object. Most slow memory-intensive programs waste most of their CPU time in L2 cache misses.
You want all the fields that are accessed in inner loops to be together in memory in dense arrays, and the other stuff should be somewhere else. So, for binary tree traversal, you want left and right and maybe some key value in a small structure, and all the other bloat needs to be elsewhere. Ideally, you allocate these structures which are nearby in the tree also nearby in memory. That way, when you get an L2 cache miss, the extra data you load into the cache likely contains data that will soon be accessed in the inner loop.
Garbage collectors typically don't bother allocating objects of the same type together. Careful management of memory can be a PITA, which is one reason I use the DataDraw tool for memory-intensive applications. It allows me to "cache together" fields, and puts wrappers around data access to hide the ugliness.
A larger group than "blind" will be visually-impaired... people who can probably see the book, but not the words. I'm facing such a fate myself soon. I came up with a similar idea, but as usual someone beat me to it. Just use your camera phone, and port some decent open-source OCR software to it.
In general, I'm not a fan of dedicated e-book readers. I think next year we'll see some killer multi-touch arm-based net-tablets with e-paper-like displays and battery life. I think the killer app will be the e-book reader. Twenty years from now, I'll be able to tell kids about the days when we had special devices for reading books, and that's what an e-book reader was, rather than a program. Remember when a "word processor" meant a dedicated and very expensive machine for editing text documents? Yes, I'm that old.
Agreed. I don't see a case where Go would be useful to me, given the Java, C++, Python, Perl, PHP, and C that I currently use.
My biggest rant about most of these new languages: garbage collection is useless! I still have to write destructors that clean up all the pointers to an object, and all garbage collection does is force me to call the destructor as a function, rather than a more clear 'delete' statement. Worse, it takes away my most powerful speed optimization tool: careful memory layout for best cache hit rates. I can write simple graph traversal code that is 10X faster in carefully designed C/C++ than in any garbage collected language for large graphs.
Good question. They use "zero knowledge" proofs:
"Scantegrity uses a process called “zero knowledge” that allows skilled, independent auditors to verify that the codes result in votes going to the right candidates, without actually revealing an individual voter’s selections."
It's super-cool stuff every slashdot geek needs to know. So, this allows us to insure our vote was counted without enabling us to sell our votes. Very cool! However, it still not fool-proof. A friend of a friend of mine has gotten so worked up over an election that she went to the polls early, and often, and voted for her whole extended family. Without requiring photo-IDs, it's really easy to do. Every show up to a poll and see your name has already been crossed off?
E-book readers will be the killer app for next year's net-tablets, IMO. The good ones will likely use Pixel Qi screens. As for OLPC, they're doing great pioneering work, which launched the netbook phenomenon. Pixel Qi will provide OLPC the paper-like screens at cost. In e-book mode, the battery should last days, not hours, and with the overall reductions in cost for the multi-touch display, processor (can you say ARM?), and power system, net tablets for under $100 may just be possible. Frankly, I'll hold out for a $250 10" tablet that looks like Apple designed it, but runs Ubuntu Netbook Remix. It'll read like e-paper, but also can play HD videos.
From your link: "The SFA product will be the only solution that integrates Asterisk directly with Skype." It costs something like $80, and is closed-source. With an open-source Skype client, I doubt Digium's closed-source solution will remain the only solution for long.
GP does get a key point: The software is key. Carriers (and freaking Verizon in particular) in the US simply refused to understand this. They build the network, while all the innovation is in the handsets and the software, but for some reason, US carriers seem to think they are the true innovators and handset providers are fluff. Now that Verizon has screwed up on handsets for three years straight, they finally realize that their strength is simply the network they build. Maybe they've been listening to their own adds. They're finally going to ship a modern phone, without screwing up the software first. Stupidity at Verizon may be going out of style.
Anyway, as said before on slashdot, Android vs iPhone is just like Windows vs Mac all over again. With Verizon on board, Motorola building 20 new Android phones next year, and 50 Android sets in the works around the world, Android is set to finally deliver on it's promise of unifying the software across a broad spectrum of handsets. There wont be any single iPhone killer, just as no single PC was ever a Mac killer. However, I see nothing that can stop Android from becoming the world's dominant smart phone OS.
The Motorola Droid isn't quite as exciting of a device as the Sony Xperia X10. I suspect we'll keep seeing Android based "iPhone killers" plunk away until Adroid wins the race.
We both must be about the same age - I'm 45. It just kills me that we have to have "play dates" for my kids to play with other kids, and my kids don't venture into the woods the way I liked so much as a kid. We agree that today's environment of fear is just that - pointless fear, driven by the media.
Anyway, some things are different today. My introduction to porn was sneaking peaks at my Dad's Playboy magazines, which he would read while Mom cleaned and cooked and held down a job. Dad's back then had it all - no poopy diapers, wives who did all the housework and had paying jobs, and who felt guilty if you didn't get enough sex...
Today, kids don't get that sneak-peek into porn when they finally become curious about sex. And, let's face it... Playboy had a sense of class and beauty missing from redtube.com. Instead, eight-year girls type "hot guy" into Google, and get hard-core video. Their intro into the idea of sex is likely going to be a foot-long dong butt-f*cking a teenager.
I took advice I got here on slashdot, and use the free opendns.com DNS filter. I also use addblock plus in firefox on all our computers. OpenDNS gives me some control over the content filiter - I use the low settings, only blocking phishing and hard-core porn. These tools are waaaaay better than anything the FCC might dream up. Instead of more government censorship, how about a program for training/educating parents, so we can all learn how to take advantage of the excellent, and free tools that already exist out there? Something as simple as requiring ISPs to send information packets about Internet filtering might do the trick. Perhaps requiring the installers who do house visits to train how to filter, not just how to use the DVR. All parents know how to record Pokemon. How many know how to protect their kids from googling "hot guys"?
I'm going blind, and I already own all her books. I want to read them again, but J.K. Rowling refuses to offer screen-reader compatible e-books. That's not a good excuse for a visit to The Pirate Bay?
Yes, I've been practising for months, ever since I got off my butt and started looking into how to survive as a programmer who can't read or drive (I still can). I started off at about half that speed, easily understandable by pretty much anyone.
As a side benefit, I have to say I have never enjoyed books so much. I just finished listening to book 5 in the Harry Potter series (I read it years ago). It only took 8 hours. My reading speed has always been frustratingly slow.
For three years, in the late 90's, I couldn't type and had to program using speech. Now I'm learning to program while listening instead of looking. I think I'd love to write the conversational interface used in Star Trek... "Computer, read me the latest headlines on Slashdot!"
While your reply sounds a bit like flame-bait, I basically have to agree. The format isn't a universal binary that gets translated to each machine architecture when installed. Instead, it's basically an archive of pre-compiled binaries for each platform you support. So, for example, my stupid Qt application has to be compiled separately for Fedora and Ubuntu. This technology would in theory allow me to merge the binaries into a single FatELF binary. Personally, I'd rather just provide separate .deb and .rpm files.
However, the idea of a universal binary is cool. We could do something like the old p-Code, where we compile to a virtual architecture, and then translate it to the machine during installation. I liked the idea when they had it way back in the early 80's (late 70's?), and I was sad to see we didn't have the compute power back then to make it fly. I bet we do now.
If you double the speed, you double the frequency, which is a problem. There are programs that will fix that for you, but even so, about 50% speed increase is about what's practical, from what I've read. I've tried the wsola program on linux, and found that speeding up audio substantially distorts it. Are there better programs that allow for > 2X speed up?
In general, to get really high speed TTS (text to speech), you need a really good TTS engine. Alex on Mac isn't bad, but SFAIK, Eloquence (the old IBM Via Voice) is still king here. I've found no other TTS engine that comes close. A friend of mine listens to Eloquence at 850wpm, which is truly amazing.
In short, blind/VI users need access to the text. There are web sites that offer the blind free e-books, which is cool, but only authors who want to participate are available (not Rowling). As my blind friend says, "The Pirate Bay is your friend." That's just reality for the blind.
Insightful. I often felt that Microsoft in the 80's and 90's encouraged students to pirate their software with lax copy protection, so that when we got into the workforce and could afford it, we'd get sucked into buying all their stuff. In my case, it worked.