Not seeing the point - all the examples show putting a folder in the quicklaunch, which can be done without HTML in the first place. Either way, it's more of a pure explorer thing than an IE thing.
By it being HTML, it can be pretty much anything. I've made a text field that pops up the contact info for a partial name match directly for my address book withou having to switch into it. Its very handy when I need to quickly call someone. I also made a photo album where I keep pictures of my family visible and up to date rather than frames on my actual desk. And I also made a nice dictionary toolbar. I imagine others might be interested in stock quotes, etc. All in all, its a very quick way to prototype additional functionality for your desktop.
You just can't beat the real Gooogle Toolbar
Hmm. Since Opera 5 or 6 I've just been typing "g searchtext" in my address bar - no google toolbar wasting valuable screen space, and Opera itself blocks popups for me. I think that's beating it pretty substantially
Can it also hgihlight your search words in the page? Can you click on the buttons on teh toolbar to advance between matches of the search terms within the page? How about performing the same search you did on the web in Google Groups? Maybe you can, but I'm just asking.
Fair enough, I suppose, but Opera provides superior bookmark management AND access (ie: open whole folder) regardless of what shell you might be running.
I'm not sure what you meant by this. In IE, I can open a hole folder of favorites too, using the "favorites panel" as opposed to the pulldown menu or toolbar.
I have to admit, I don't see the big deal on that site
The big deal boils down that instead of having a text field, you can have a "CalendarField" that correctly limits input to valid dates and attaches to a date picker -- and when you build thepage, you just have the one tag. Its basically like a way of extending a tag (i.e. object-oriented ineheritence) and extend/alter its behavior without having to repeat that modified behavior everywhere you use it in the code. I think its anice idea, and doesn't have to be MS_specific -- non one else has implemented it though (and MS did propose it as a standard, wary though I am of their tactics).
In fact, given the geek fondness for tinkering, I'd be damned surprised if you can't do it in Firefox and Mozilla too. Although given the geek attitude for user interfaces, you'd likely have to edit some obscure config file to do it 100% the way you wanted it...
(I onced shared this story with a popular computer magazine, and it got published)
Years ago, I went with a friend of mine to his boss's house to fix his computer. This guys' house wasn't a house, but a log cabin in a rustic town along a river with a winding one-lane road to get there. There was a general store and everything!
So we get there, and I sit down at this computer on a desk and start plugging away. Meanwhile, my friend gets a call from his wife. He starts pacing around, handling the various antiques and oddities one rarely sees except in a rustic environment like this.
When he picks up the revolver, his boss starts yelling "Put that down! Put that down!", but my friend was too distracted with his conversation to pay attention to the outside world. Sure enough, a loud *BANG* rang out. My friend dropped the phone and everyone checked themselves for holes.
After we were confidet we were all still alive, we noticed the HP LaserJet III printer sitting inches away from me on the desk had a hole in it. Wiping the sweat form my brow, I started laughing because the printer was still printing at the time! We later took some of the external housings off the printer and found some fragments, but the printers guts weren't damaged and it was printing fine.
(I wish HP's products were still that good)
The gun, of course, was real. My friend's boss sais he kept it on the end table to shoot the bats that inevitably found their way into his cabin. I think now he might stowe it when guests come over.
Your definition of hacker is just that -- "your definition". My definiton of a hacker is someone who tinkers with things outside of the mainstream way, figuring out how they work, why they broke, or ways to get things to do things they don't want to do.
I know sysadmins who fall into both categories. Some just plug along doing the usual tasks. Others are thinking of ways to imrpove the systems and the lives of those who live in them. I also know Java programmers who code exactly what the requirements say. But there are also Java hackers. Compared to the former, I am one. I decompile code, I disassemble class files, I integrate Java with the world outside of Java, I debug the VM dumps in the rare event the VM itself crashes -- all to make the lives of myself, and the others who live in the Java world with me, better.
I seriously doubt you know much about Java, just as my sysadmin experience is feeble. But, I am ready to admit that. And, in true hacker form, my lack of knowledge is not a road block but a challenge for me to dig deeper, experiment, understand and learn more, not to spout off myopic opinions as some sort of deep insight.
I don't know, because I haven't really tried Python yet. I once worked on a project where users were expected to script bsuiness logic in Python (well, Jython) in this big fat Java application. Well, the users couldn't handle the whole indentation thing, so it was an utter failure and the IT department did the scripting (which at that point gained us nothing over on-the-fly Java compilation). Of course, I don't know how they ever expected the users to be able to write a script in ANY language.
Anyways, I came to Java from C++. I've certainly visited and revisit other lanugauges all the time. But the things I like about Java for large projects, or in general, are below. Please reply with how Python stacks up. Maybe I should give it a try...
No more separate header files!
Garbage collection
Implicit bounds checking
Core libraries not afraid to use Exceptions
Standard core libraries provide an enormously useful standard toolset.
Strings are used everywhere freely making the code easier to debug
Stack traces!
Javadoc
Many choices for great IDEs
Much much much more cross-platform
...
I could probably think of more, but my brain is fried. Again, when I started using Java, remember that my previous big projects were in Microsoft ASP (w/ JScript) and (non-Microsoft) C++.
I have *never* met a true hacker who programmed in java
Pleased to meet you.
A true hacker would know that there is no "one language". I've "hacked" together scripts in shell script, batch, EMCAScript, VBScript, and yes.. Perl! I've also written applications in BASIC, C, C++, Java. Many of these applications started life as quick 'n' dirty scripts. As I needed better maintainability, I've rebuilt them in the latter languages because they become easier to extend and maintain.
I recall the day when Perl was the "Practical Extraction and Report Language". Now you can build entire geek community websites out of it. What happened? Perl grew up. It gained namespaces, OO-capabilities, became more consistent, etc. Not all scripting languages have this (i.e. batch files, yuck!). Some develop them overtime (look at the great strides that PHP has made since itself was a bunch of scripts written in another language). So even the best hacker tools mature. Does that make them less of a hacker tool, or more of a hacker and non-hacker tool?
Oh, and as for why anyone would program in Java? Well, personally, I have a wife and kids to feed. It just so happens that my bosses pay me to write applications that deploy to WebSphere servers. While I could use Jython, or probably some strange Java-Perl bindings that no doubt exist, I would be quickly fired leaving my family in an unreliable state of upkeep.
So, I "hack" in the evenings and on weekends. I "hack" at work to help get my job done. And you know what, sometimes I even do it in Java.
BTW, Paul Graham just went down several notches on my respectibility meter. What an idiot.
Maybe they should have sort of moderation system (like Slashdot's) where visitors are allowed to rate an article's content as being accurate or not. With the fluid nature of a Wiki's content, you may have to tie those moderations to specific revisions, but still...
Or, maybe they do. Personally, I'm running Apache HTTPD. When I had a need for Perl, I ran ActiveState's Perl. I don't use MySQL or PHP, so I don't run those but could. I also run Thunderbird, Apache James mail server, APache Tomcat, JBOss and eclipse.
Why would you want only yourbrowser to support mouse gestures when any program can?
Yes, you will get snickers from your geek friends when they see the name of the utlilty you're running, but theywill quickly shut up once they've become addicted to it like you will.
How about some slashdotter form Tennessee goes and visits him? E-mail likely won't work. A letter might, but if you show up in person and talk to him, maybe you can educate him, or he you.
there just seems to be something about it that drives all the Oo fanboys up the wall
Yeah, I know what you mean. These kids can't wrap their tiny minds around the following concepts:
A table is not a class
A row is not an object
A column is not a property
Those points are indeed the problems. All too often those OO-types (wait, I'm an OO-type!) try to treat tables as classes, rows as objects, etc.
But then again, they think everything is a nail when all they have is a hammer. I find that 90% of the times I encounter RDBMS persistence in code, there is no reason for the data to be in an RDBMS.
There are a lot disadvantages to using an RDBMS as there are advantages. You have to weigh the cost of the impedance mismatch with the gains of the robustness of an RDBMS.
Pournelle also gives examples of his works being pirated in Asia and then resold in the US via major retail outlets
I see how watermarking might help this, but all-out DRM I don't think would. Obviously, those people have no fear of copyright law in the U.S. so they would just physically copy the CD thousands of time, DRM included!
What you ahveto do hear is press China to stop them internally and press retail outlets to make a reasonable effort to ensure the legitimacy of their inventory acquisitions.
That can all be done with current laws and DRM won't help. Watermarking might help law enforcement track down the source, but I consider watermarking to be very tame and not part of the unbrella of "truly evil DRM".
What if we made "gift givers" and "donators" anonymous. Think about it. You'd still be helpingthe candidate or party you think fits your views, but they'd have no idea who exactly helped them so they could only follow their views rather than trying to follow the views of a particular contributor.
Pick up a book? I was programming in x86 assembler when I was in diapers! Well, not diapers, but I coudln't drive yet.
As I've said before, you *can* do assembler for the JVM -- you just don't use a Java compiler, you use a compiler made for compiling Java assembler instructions into Java bytecodes.
And as for the HotSpot optimizing for every known input, it doesn't do that all at once. It attempts to optimize on the fly. And it's not just for invariants. It can do so as values change. It may decide to unroll a loop when the counter is less than 3 but keep it in a loop when its more than three. And it might decide to change that critical point on the fly.
I'll give you that it *is* general optimization. I think it has less to do with optimizing for the hardware but optimizing for how your program is expected to operate. As others have pointed out, CPUs already mangle the instructions so much internally, that you're compiler itself doesn't even write the actual instructions that will be exeucted anymore.
I propose a "Rich Browser" in honor of the rich client metaphor I'd like to see resurrected where the client contains meaningful state and has a complex (in a useful way) interface. You can do this with HTML and friends, but it is much more difficult because they weren't built for that.
Don't you see how those two things are not equivalent?
Yes.
However, a general purpose run-time profiler/optimizer is not going to outperform a proficient human.
This is one point we'll continue to disagree on. The theory behind runtime optimizations is that there are some things you can' know ahead of time or some conditions that simply aren't static. Again, a "human" could hand-code his application to do runtime proofiling and choose diferrent predefined code-paths based on that... is that what you meant -- essentially building in a runtime optimizer?
With c++, you can examine the native code, and if the compiler does something stupid, you can drop down into assembly and get your hands dirty removing pipeline stalls, making sure your inner loop fits in the cache, and other stuff that you just have to hope the JVM will do for you if you are a java programmer
You do have an intermediate option. You can disassemble the classfile (standard JDKs come with the tool named 'javap' for this), look at the JVM Instructions and see if they are doing something stupid. Some crazy minded folk have hand-coded Java machien instructions and built *.class files themselves. There are also a handful of other languages you can compile ot the JVM which may give you more control over what you are doing.
But, that's not controlling it in Java. In that case, you stil have an option. There are times when I've decompield a Java class file, looked at the JVM instructions it used, and realized if I coded my Java program diferently, the compiler would generate different instructions that would be fewer and faster.
Fastest performance isn't it.
I still have a faith that Java will be faster than C++ in some cases. The fact that this benchmark was flawed doesn't mean it isn't true. I certainly don't believe it to be true ubiquitously for all cases. BUt blanket statements like "Fastest performance it isn't" are just too absolute for be to pay attention to.
While it can be a little clunky in places, when properly implemented it can be every bit as rich as.NET/GTK/Qt based applications
I can also trim my lawn with hand-clippers, but a lawn mower would be much more efficient.
It is not easy, stable, etc. to make a complex but usable web interface. They all end up cutting corners, relyig on browser-specific functionality, plugins, etc.
What I want to see is something like XUL or XAML turned into a standard and then having a XAML or XUL browser instead of a web browser, with a simple HTML-rendering component built into it.
Reading a news article is a very HTML thing to do. Perusing and replying to threaded discussions is an exmaple of something that would probably be experienced better through an actual UI instead of an HTML-ized version. How about your Yahoo Mail or GMail accounts? They woudl also benefit from a UI-markup-language producing a more traditional client-side UI.
The question is, how long owuld it take a UI-browser to become as ubiquitous as the web-browser?
What bothered me more was his resort to pedantry and his attitude.
Agreed. Sometimes, though, I'm just as guilty. I love Java. And I'ma guy who went BASIC->C->ASM->C++->Java (with some unimportant tangents along the way). I'm not saying it's 100% better than C++ either. The part I get emotional about is the absolutists who can't believe Java could ever be better than C++ in any way. Unfortunately, the benchmark which started this thread -- while done with good intentions -- does nothing to further Java's cause.
Have you never heard of self modifying code? Look it up. You can do it in c++. You cannot do it in Java.
Could you be more specific? I mean, having a code path that chooses replaces part of its functionality could be termed "self-modifying code". You can infact do that in Java with the Functor pattern. If you mean actually modifying discrete portions of logic without the aid of virtual functions that can be replaced, then wouldn't that be exactly what the HotSpot is doing? Now, perhaps you can't do that in the Java language. But, can you do it in JVM bytecodes (other languages can be compiled to JVM bytecode)?
With c++, you can take advantage of architecture specific optimizations
You have that in Java as well. The JVM, being part of Java, is certainly allowed to be written differently for different architectures -- and the HotSpot VMs by extension can re-arrange the machine code it JIT'ed to be more optimal for the hardware at hand.
Furthermore, the specific optimizations you name could be implemented in c++. The compiler won't do it for you, but a clever enough developer could do them
This is teh same argument ASM programmers used against higher-level languages, and the same argument C programmers made against C++. If you're such a hard-core hand-optimizing code freak, by not write your code in your favorite hex editor as the actual machine instructions. This is what the authors of Comanche Maximum Overkill did years ago (sorry, it's so oldMy point is that proper design-time optimization is going to be better than compile-time optimization no matter what. No algorithm can yet approach the intelligence of a human when it comes to optimizing.
I disagree. Runtime optimizations are basedon the actual state of a running system, nto a developer's pre-conveived notion of what it will encounter. Profiling can get you part of that, but only if you profile for every possible input. Have you never sat down and watched a user use your program and realize, "hey, that user is using it in a way it wasn't designed for!". Runtime optimizations could handle that, but for all of the intelligence of a human, we're biased, short-sighted and emotional (just look at some of the posts in this thread!)
The whole point of java is to remove you, the developer, from the architecture.
Yes, you're right! Well, no you're not becuase that's not the whole point of Java. But the underlying point is still true.
With Java, you let the friggin machine do the work of optmizing it for where you're deploying. As we've seen elsewhere in this thread, you can use special nitty-gritty flags on C++ compilers to target certain architectures. But how often do you do to a page to download binaries and see one compiled for your exact stepping model of the Pentium III? You just don't. Sure, open source has the advantage that you can get teh source and compile it for your hardware, but that does not help the rest of the non-geek world. Maybe if machines saw a *.c file and when you double-clicked it, it automatically compiled it specifically for your machine and exeucted it users would adopt that.
But even then, the statically-compiled binary will not have runtime optimizations unless it is hand-coded in. That takes me back to my earlier point that if you want to sit in the dark corners writing tight device drivers in machine code, go ahead -- but the rest of us need to get paid at jhobs where performance is measuredn ot just in how fast your application runs, but in how long it takes to get it done. And while you're may be 3% faster, mine will be developede in 3% of the time that yours was.
Did you see a definition for Java somewhere in the benchmark writeup? Certianly Java is not in its entirety defined by the JVM, but the casual term "Java" is an unbrella term that certainly includes a JVM.
This lack of an explicit definition is one of the problems with the banchmark. Did he mean the language? The standard libraries? The VM? The version or vendor-specific implementation so of the standard libraries or VM?
Once again, static compile-time optimizations cannot match run-time optimizations unless the compile-time optimization is to write out every possible code-path that is most optimal for the entire input space. Now, you could write your program such that it analyzes its own runtime state and switches code paths while running -- but that is no the compiler making the optimization. And, at that point, you may say it is human-optimized, so it's better...
But then you might as well also point out that human-optimized C code can beat out C-compiler-optimized C code.
FWIW, I've heard arguments that even that isn't true because humans are extremly biased and a computer can just see things for what they are in terms of bits and bytes without getting distracted by our human abstractions.
Slashdot Mirror
You can embed HTML in the TaskBar
Not seeing the point - all the examples show putting a folder in the quicklaunch, which can be done without HTML in the first place. Either way, it's more of a pure explorer thing than an IE thing.
By it being HTML, it can be pretty much anything. I've made a text field that pops up the contact info for a partial name match directly for my address book withou having to switch into it. Its very handy when I need to quickly call someone. I also made a photo album where I keep pictures of my family visible and up to date rather than frames on my actual desk. And I also made a nice dictionary toolbar. I imagine others might be interested in stock quotes, etc. All in all, its a very quick way to prototype additional functionality for your desktop.
You just can't beat the real Gooogle Toolbar
Hmm. Since Opera 5 or 6 I've just been typing "g searchtext" in my address bar - no google toolbar wasting valuable screen space, and Opera itself blocks popups for me. I think that's beating it pretty substantially
Can it also hgihlight your search words in the page? Can you click on the buttons on teh toolbar to advance between matches of the search terms within the page? How about performing the same search you did on the web in Google Groups? Maybe you can, but I'm just asking.
Fair enough, I suppose, but Opera provides superior bookmark management AND access (ie: open whole folder) regardless of what shell you might be running.
I'm not sure what you meant by this. In IE, I can open a hole folder of favorites too, using the "favorites panel" as opposed to the pulldown menu or toolbar.
I have to admit, I don't see the big deal on that site
The big deal boils down that instead of having a text field, you can have a "CalendarField" that correctly limits input to valid dates and attaches to a date picker -- and when you build thepage, you just have the one tag. Its basically like a way of extending a tag (i.e. object-oriented ineheritence) and extend/alter its behavior without having to repeat that modified behavior everywhere you use it in the code. I think its anice idea, and doesn't have to be MS_specific -- non one else has implemented it though (and MS did propose it as a standard, wary though I am of their tactics).
In fact, given the geek fondness for tinkering, I'd be damned surprised if you can't do it in Firefox and Mozilla too. Although given the geek attitude for user interfaces, you'd likely have to edit some obscure config file to do it 100% the way you wanted it...
Amen to that!
I've posted this before, but I like to keep bringing it up for the sake of being the devil's advocate:
I'm starting to feel sorry for IE. Everyone's picking on it. It does have some nice features:
(I onced shared this story with a popular computer magazine, and it got published)
Years ago, I went with a friend of mine to his boss's house to fix his computer. This guys' house wasn't a house, but a log cabin in a rustic town along a river with a winding one-lane road to get there. There was a general store and everything!
So we get there, and I sit down at this computer on a desk and start plugging away. Meanwhile, my friend gets a call from his wife. He starts pacing around, handling the various antiques and oddities one rarely sees except in a rustic environment like this.
When he picks up the revolver, his boss starts yelling "Put that down! Put that down!", but my friend was too distracted with his conversation to pay attention to the outside world. Sure enough, a loud *BANG* rang out. My friend dropped the phone and everyone checked themselves for holes.
After we were confidet we were all still alive, we noticed the HP LaserJet III printer sitting inches away from me on the desk had a hole in it. Wiping the sweat form my brow, I started laughing because the printer was still printing at the time! We later took some of the external housings off the printer and found some fragments, but the printers guts weren't damaged and it was printing fine.
(I wish HP's products were still that good)
The gun, of course, was real. My friend's boss sais he kept it on the end table to shoot the bats that inevitably found their way into his cabin. I think now he might stowe it when guests come over.
The best way to write short Java programs is to first write a robust library, and then call it from a very short program!
import library.RobustHelper;
public static void main(String[] args) {
RobustHelper.getInstance().execute(args);
}
He's right.
Or, you're both wrong.
Your definition of hacker is just that -- "your definition". My definiton of a hacker is someone who tinkers with things outside of the mainstream way, figuring out how they work, why they broke, or ways to get things to do things they don't want to do.
I know sysadmins who fall into both categories. Some just plug along doing the usual tasks. Others are thinking of ways to imrpove the systems and the lives of those who live in them. I also know Java programmers who code exactly what the requirements say. But there are also Java hackers. Compared to the former, I am one. I decompile code, I disassemble class files, I integrate Java with the world outside of Java, I debug the VM dumps in the rare event the VM itself crashes -- all to make the lives of myself, and the others who live in the Java world with me, better.
I seriously doubt you know much about Java, just as my sysadmin experience is feeble. But, I am ready to admit that. And, in true hacker form, my lack of knowledge is not a road block but a challenge for me to dig deeper, experiment, understand and learn more, not to spout off myopic opinions as some sort of deep insight.
I don't know, because I haven't really tried Python yet. I once worked on a project where users were expected to script bsuiness logic in Python (well, Jython) in this big fat Java application. Well, the users couldn't handle the whole indentation thing, so it was an utter failure and the IT department did the scripting (which at that point gained us nothing over on-the-fly Java compilation). Of course, I don't know how they ever expected the users to be able to write a script in ANY language.
Anyways, I came to Java from C++. I've certainly visited and revisit other lanugauges all the time. But the things I like about Java for large projects, or in general, are below. Please reply with how Python stacks up. Maybe I should give it a try...
I could probably think of more, but my brain is fried. Again, when I started using Java, remember that my previous big projects were in Microsoft ASP (w/ JScript) and (non-Microsoft) C++.
I have *never* met a true hacker who programmed in java
Pleased to meet you.
A true hacker would know that there is no "one language". I've "hacked" together scripts in shell script, batch, EMCAScript, VBScript, and yes.. Perl! I've also written applications in BASIC, C, C++, Java. Many of these applications started life as quick 'n' dirty scripts. As I needed better maintainability, I've rebuilt them in the latter languages because they become easier to extend and maintain.
I recall the day when Perl was the "Practical Extraction and Report Language". Now you can build entire geek community websites out of it. What happened? Perl grew up. It gained namespaces, OO-capabilities, became more consistent, etc. Not all scripting languages have this (i.e. batch files, yuck!). Some develop them overtime (look at the great strides that PHP has made since itself was a bunch of scripts written in another language). So even the best hacker tools mature. Does that make them less of a hacker tool, or more of a hacker and non-hacker tool?
Oh, and as for why anyone would program in Java? Well, personally, I have a wife and kids to feed. It just so happens that my bosses pay me to write applications that deploy to WebSphere servers. While I could use Jython, or probably some strange Java-Perl bindings that no doubt exist, I would be quickly fired leaving my family in an unreliable state of upkeep.
So, I "hack" in the evenings and on weekends. I "hack" at work to help get my job done. And you know what, sometimes I even do it in Java.
BTW, Paul Graham just went down several notches on my respectibility meter. What an idiot.
I'm starting to feel sorry for IE. Everyone's picking on it. It does have some nice features:
Maybe they should have sort of moderation system (like Slashdot's) where visitors are allowed to rate an article's content as being accurate or not. With the fluid nature of a Wiki's content, you may have to tie those moderations to specific revisions, but still...
Obviously, there must be some reason they don't
Or, maybe they do. Personally, I'm running Apache HTTPD. When I had a need for Perl, I ran ActiveState's Perl. I don't use MySQL or PHP, so I don't run those but could. I also run Thunderbird, Apache James mail server, APache Tomcat, JBOss and eclipse.
mouse gestures
Why would you want only yourbrowser to support mouse gestures when any program can?
Yes, you will get snickers from your geek friends when they see the name of the utlilty you're running, but theywill quickly shut up once they've become addicted to it like you will.
Actually, guns can be effectively used for self defense.
How about some slashdotter form Tennessee goes and visits him? E-mail likely won't work. A letter might, but if you show up in person and talk to him, maybe you can educate him, or he you.
there just seems to be something about it that drives all the Oo fanboys up the wall
Yeah, I know what you mean. These kids can't wrap their tiny minds around the following concepts:
Those points are indeed the problems. All too often those OO-types (wait, I'm an OO-type!) try to treat tables as classes, rows as objects, etc.
But then again, they think everything is a nail when all they have is a hammer. I find that 90% of the times I encounter RDBMS persistence in code, there is no reason for the data to be in an RDBMS.
There are a lot disadvantages to using an RDBMS as there are advantages. You have to weigh the cost of the impedance mismatch with the gains of the robustness of an RDBMS.
Pournelle also gives examples of his works being pirated in Asia and then resold in the US via major retail outlets
I see how watermarking might help this, but all-out DRM I don't think would. Obviously, those people have no fear of copyright law in the U.S. so they would just physically copy the CD thousands of time, DRM included!
What you ahveto do hear is press China to stop them internally and press retail outlets to make a reasonable effort to ensure the legitimacy of their inventory acquisitions.
That can all be done with current laws and DRM won't help. Watermarking might help law enforcement track down the source, but I consider watermarking to be very tame and not part of the unbrella of "truly evil DRM".
What if we made "gift givers" and "donators" anonymous. Think about it. You'd still be helpingthe candidate or party you think fits your views, but they'd have no idea who exactly helped them so they could only follow their views rather than trying to follow the views of a particular contributor.
Pick up a book? I was programming in x86 assembler when I was in diapers! Well, not diapers, but I coudln't drive yet.
As I've said before, you *can* do assembler for the JVM -- you just don't use a Java compiler, you use a compiler made for compiling Java assembler instructions into Java bytecodes.
And as for the HotSpot optimizing for every known input, it doesn't do that all at once. It attempts to optimize on the fly. And it's not just for invariants. It can do so as values change. It may decide to unroll a loop when the counter is less than 3 but keep it in a loop when its more than three. And it might decide to change that critical point on the fly.
I'll give you that it *is* general optimization. I think it has less to do with optimizing for the hardware but optimizing for how your program is expected to operate. As others have pointed out, CPUs already mangle the instructions so much internally, that you're compiler itself doesn't even write the actual instructions that will be exeucted anymore.
I propose a "Rich Browser" in honor of the rich client metaphor I'd like to see resurrected where the client contains meaningful state and has a complex (in a useful way) interface. You can do this with HTML and friends, but it is much more difficult because they weren't built for that.
Don't you see how those two things are not equivalent?
Yes.
However, a general purpose run-time profiler/optimizer is not going to outperform a proficient human.
This is one point we'll continue to disagree on. The theory behind runtime optimizations is that there are some things you can' know ahead of time or some conditions that simply aren't static. Again, a "human" could hand-code his application to do runtime proofiling and choose diferrent predefined code-paths based on that... is that what you meant -- essentially building in a runtime optimizer?
With c++, you can examine the native code, and if the compiler does something stupid, you can drop down into assembly and get your hands dirty removing pipeline stalls, making sure your inner loop fits in the cache, and other stuff that you just have to hope the JVM will do for you if you are a java programmer
You do have an intermediate option. You can disassemble the classfile (standard JDKs come with the tool named 'javap' for this), look at the JVM Instructions and see if they are doing something stupid. Some crazy minded folk have hand-coded Java machien instructions and built *.class files themselves. There are also a handful of other languages you can compile ot the JVM which may give you more control over what you are doing.
But, that's not controlling it in Java. In that case, you stil have an option. There are times when I've decompield a Java class file, looked at the JVM instructions it used, and realized if I coded my Java program diferently, the compiler would generate different instructions that would be fewer and faster.
Fastest performance isn't it.
I still have a faith that Java will be faster than C++ in some cases. The fact that this benchmark was flawed doesn't mean it isn't true. I certainly don't believe it to be true ubiquitously for all cases. BUt blanket statements like "Fastest performance it isn't" are just too absolute for be to pay attention to.
While it can be a little clunky in places, when properly implemented it can be every bit as rich as .NET/GTK/Qt based applications
I can also trim my lawn with hand-clippers, but a lawn mower would be much more efficient.
It is not easy, stable, etc. to make a complex but usable web interface. They all end up cutting corners, relyig on browser-specific functionality, plugins, etc.
What I want to see is something like XUL or XAML turned into a standard and then having a XAML or XUL browser instead of a web browser, with a simple HTML-rendering component built into it.
Reading a news article is a very HTML thing to do. Perusing and replying to threaded discussions is an exmaple of something that would probably be experienced better through an actual UI instead of an HTML-ized version. How about your Yahoo Mail or GMail accounts? They woudl also benefit from a UI-markup-language producing a more traditional client-side UI.
The question is, how long owuld it take a UI-browser to become as ubiquitous as the web-browser?
What bothered me more was his resort to pedantry and his attitude.
Agreed. Sometimes, though, I'm just as guilty. I love Java. And I'ma guy who went BASIC->C->ASM->C++->Java (with some unimportant tangents along the way). I'm not saying it's 100% better than C++ either. The part I get emotional about is the absolutists who can't believe Java could ever be better than C++ in any way. Unfortunately, the benchmark which started this thread -- while done with good intentions -- does nothing to further Java's cause.
*Sigh*, yes, it's true. At least Sun finally seems to be warming up to client-side Java with the partial startup optimizations coming in 1.5 and the recent JDIC project opening.
There is a much more agressive plan, though, that could lead to a Java shell to allow a much faster starting of client-side apps.
Have you never heard of self modifying code? Look it up. You can do it in c++. You cannot do it in Java. Could you be more specific? I mean, having a code path that chooses replaces part of its functionality could be termed "self-modifying code". You can infact do that in Java with the Functor pattern. If you mean actually modifying discrete portions of logic without the aid of virtual functions that can be replaced, then wouldn't that be exactly what the HotSpot is doing? Now, perhaps you can't do that in the Java language. But, can you do it in JVM bytecodes (other languages can be compiled to JVM bytecode)?
With c++, you can take advantage of architecture specific optimizations
You have that in Java as well. The JVM, being part of Java, is certainly allowed to be written differently for different architectures -- and the HotSpot VMs by extension can re-arrange the machine code it JIT'ed to be more optimal for the hardware at hand.
Furthermore, the specific optimizations you name could be implemented in c++. The compiler won't do it for you, but a clever enough developer could do them
This is teh same argument ASM programmers used against higher-level languages, and the same argument C programmers made against C++. If you're such a hard-core hand-optimizing code freak, by not write your code in your favorite hex editor as the actual machine instructions. This is what the authors of Comanche Maximum Overkill did years ago (sorry, it's so old My point is that proper design-time optimization is going to be better than compile-time optimization no matter what. No algorithm can yet approach the intelligence of a human when it comes to optimizing.
I disagree. Runtime optimizations are basedon the actual state of a running system, nto a developer's pre-conveived notion of what it will encounter. Profiling can get you part of that, but only if you profile for every possible input. Have you never sat down and watched a user use your program and realize, "hey, that user is using it in a way it wasn't designed for!". Runtime optimizations could handle that, but for all of the intelligence of a human, we're biased, short-sighted and emotional (just look at some of the posts in this thread!)
The whole point of java is to remove you, the developer, from the architecture.
Yes, you're right! Well, no you're not becuase that's not the whole point of Java. But the underlying point is still true.
With Java, you let the friggin machine do the work of optmizing it for where you're deploying. As we've seen elsewhere in this thread, you can use special nitty-gritty flags on C++ compilers to target certain architectures. But how often do you do to a page to download binaries and see one compiled for your exact stepping model of the Pentium III? You just don't. Sure, open source has the advantage that you can get teh source and compile it for your hardware, but that does not help the rest of the non-geek world. Maybe if machines saw a *.c file and when you double-clicked it, it automatically compiled it specifically for your machine and exeucted it users would adopt that. But even then, the statically-compiled binary will not have runtime optimizations unless it is hand-coded in. That takes me back to my earlier point that if you want to sit in the dark corners writing tight device drivers in machine code, go ahead -- but the rest of us need to get paid at jhobs where performance is measuredn ot just in how fast your application runs, but in how long it takes to get it done. And while you're may be 3% faster, mine will be developede in 3% of the time that yours was.
Uh, no. Java != JVM.
Did you see a definition for Java somewhere in the benchmark writeup? Certianly Java is not in its entirety defined by the JVM, but the casual term "Java" is an unbrella term that certainly includes a JVM.
This lack of an explicit definition is one of the problems with the banchmark. Did he mean the language? The standard libraries? The VM? The version or vendor-specific implementation so of the standard libraries or VM?
Once again, static compile-time optimizations cannot match run-time optimizations unless the compile-time optimization is to write out every possible code-path that is most optimal for the entire input space. Now, you could write your program such that it analyzes its own runtime state and switches code paths while running -- but that is no the compiler making the optimization. And, at that point, you may say it is human-optimized, so it's better...
But then you might as well also point out that human-optimized C code can beat out C-compiler-optimized C code.
FWIW, I've heard arguments that even that isn't true because humans are extremly biased and a computer can just see things for what they are in terms of bits and bytes without getting distracted by our human abstractions.