by Bruce Eckel
January 30, 2007
The Java backlash has been building up steam, and we’re starting to see some fundamental shifts because of it.
Several of Bruce Tate’s books focused on the flaws in Java and the need to let go of some of the ideas that haven’t worked out. And blogs like this and this have been appearing more frequently. And of course there’s Steve Jobs now-famous quote (referring to the iPhone) : “Java’s not worth building in. Nobody uses Java anymore. It’s this big heavyweight ball and chain.”
This backlash has only been necessary because of Sun’s death grip on the idea of ubiquitous, omniscient Java. It was admirable once, but a language only evolves if its designers and advocates can acknowledge problems. Pretending that a language is successful in places where it’s not is just denial.
Some adaptations have occurred. Finally admitting that EJBs have cost the world enormously, the EJB3 team took lessons from Hibernate and Spring, but not enough to really fix the problem. Most people seem to find that Hibernate and Spring are still simpler and more straightforward than EJB3, so the lack of a rush back to a technology that had such a heavy cost in the past shouldn’t be a surprise.
Java 5 was a tacit admission that Microsoft was doing some very interesting things with C#, and proposed features in Java 7 support the idea that Java is now playing catch-up with C# 3.0. Competition is good, and Java is not dead. It continues to evolve, and the appearance of new languages built on the JVM, like JRuby, Scala and Groovy, is a sign of vitality in the world of Java.
The Web is a Mess
If you use Firefox, how many sites do you visit that are at least partially unreadable because they’ve been created only for Internet Explorer (IE)é It seems to me that things are getting worse ; I’m seeing more, not less sites that don’t work right with Firefox…to the point that I’m seriously considering going back to IE.
The noble promise of CSS has not panned out. After years, it’s still implemented inconsistently across browsers. As long as you’re using HTML and CSS, you’ll always wonder if what you’ve created is going to produce an unpleasant effect on different browsers. Browsers other than IE or Firefox can often get it even worse.
How much effort must we expend to compensate for the long sequence of misguided decisions that has produced today’s webé
Rich Internet Applications
At some point we must ask why Java applets haven’t become ubiquitous on the internet as the client-side standard for RIAs (Rich Internet Applications).
This is an especially poignant question because Gosling and team justified rushing Java out the door (thus casting in stone many poorly-considered decisions) so that it could enable the internet revolution. That’s why the AWT and Applets were thrown in at the last second, reportedly taking a month from conception to completion. Bill Venners quotes Patrick Naughton as saying : “It was a timing issue, there was only about a three-month window in which the whole Java phenomenon could have happened. We barely made it.” I’ve heard that argument before,1 and this attitude always seems to be a mistake when building programming languages. You’re creating the fundamental architecture that you hope people will adopt and use for many years. This is the place that requires careful thought, not rushing.
I can see why the Green Team took this attitude : it’s the Microsoft Way. Throw a product out there in order to get your foot in the window. The product doesn’t have to be perfect ; it just has to hold the market space. Over time, you can fix up any flaws that occurred from rushing the thing out. It’s the marketing version of agile methods.
This approach is workable for dynamic languages. One of the most popular languages ever, Visual Basic, has evolved over the years. Python has fixed some things that would even break old code, in order to clean up the language. Reportedly Ruby is planning to do the same thing.
But for static languages with a large code base (verbose languages in particular), fixing things up doesn’t seem to work so well. All the code must be recompiled and possibly changed—although I would argue that Java could have taken the Python approach : just don’t upgrade if you don’t want to change. There are lots of companies that don’t upgrade to newer Java versions, anyway.
The Installation Problem
So Java has been around for 10 years and applets are not the primary way that we interact with the web. I think the main reason for this is the installation problem, another area of Java that wasn’t well thought-out. In fact, why do we like Ajaxé
Various fixes have been attempted for Java over the years, but I think the basic issue is that whoever is trying to solve the installation problem is someone who has a more inward technical focus rather than what’s really needed : an outward user-experience focus.
For example, I was stymied by early Linux distributions because of the challenge of installation. About once a year, I would try to install Linux, and it would immediately start asking me questions. Only a hard-core Linux geek would know the answers to these questions. I couldn’t even fake my way through, so I would give up and try again the next year. Then Red Hat came along (at least, I think they were the first to focus on the installation experience) and set it up to install Linux without asking questions, or at least to give me reasonable defaults. That’s when Linux started becoming popular (more recently, Ubuntu seems to have taken the lead in solving the Linux friendliness problem).
Installing the JRE requires the end user to answer questions. To a techie, the answer to these questions is easy or obvious, but to the rest of the web users, questions frighten them during installation. Here’s an article, and comments, that illustrates the problem with Java installations. Although this article is primarily complaining about updates, there is also a lot of confusion about keeping old versions of Java around. Here’s another “installation matters” blog.
JNLP, a.k.a. Java WebStart, was supposed to solve some of these problems to create easy-install desktop applications. I think the reason that JNLP hasn’t become commonly used can be summed up by looking at https://aerith.dev.java.net/, a page for one of the flagship “Cool JavaOne Demos.” If you click on the JNLP version link on that page, it will appear to start up, downloading a bunch of stuff and asking you questions. And then it does nothing. No error messages or any information to tell you what happened. Repeated attempts yield the same results, only faster because the requisite files have already been downloaded. At least, that was my experience. If it worked for you, I’d say that’s even worse—it randomly works on some platforms and not others. How do you debug such a thingé
It’s not impossible to build GUI applications with Java, but it’s been 10 years and there are still installation hiccups with applets, Java WebStart, and regular applications. After 10 years, people don’t trust it anymore. If it’s not there after 10 years, then I’m going to go out on a limb and say that someone doesn’t consider this problem important enough to fix. And even if they did, there have been so many bad experiences among consumers that it would take years to get the trust back.
Experience with Java Applets and Applications
The initial user experience with AWT did a lot to dampen the enthusiasm about Sun’s Java fanfare, and I don’t think applets ever recovered. As a result, Java never became mainstream for RIAs.
Even now, I’ve had bad experiences trying to run Java applets from a web site. They just fail, with no clue as to what’s gone wrong. Worse, they can sometimes leave cruft behind that has even prevented Firefox from opening new windows, until I reboot the machine.
A common response to the “applets are dead” statement is “No they’re not. I still use them.” Applets aren’t useless ; people still create very impressive things with them. The Java Posse highlights one or more applets each week. The statement should instead read, “Applets are dead for Web RIAs.” The installation process for the JRE and for any particular applet is not reliable enough for anyone to depend on them for a general-purpose website.
Desktop applications have also suffered, and left bad marks on Java in general, further sullying applets by association. I had an experience with a Java product called Memorex exPressit. The UI was terrible and buggy. I had a contrasting experience with the Logitech IO pen support software (written in .NET), which worked cleanly and looked nice. You could argue that the Memorex programmer(s) were just inexperienced, but the Logitech software was just a small application that worked well without, it seems, any Herculean efforts on the part of the programmer(s), whereas it’s hard to think of any application that I use on a regular basis that’s written in Java. Eclipse is a brilliant piece of software, but I think “Herculean effort” applies.
Consider Corel’s attempt to create a word processor using Java (I don’t remember whether they were trying to port WordPerfect or write something from scratch). It was obviously too early to try it, since all they had was the AWT. But only obvious in hindsight—if you were listening to Sun’s marketing hype at the time, it was the right choice. Shrugging it off as no big deal doesn’t work, because this kind of failure scares people away from doing things in Java.
OpenOffice is not written in Java, but in C. It was speed, and perhaps the need to more directly control the underlying platform. Despite the party line that has always come out of Sun (Many years ago, I attended a press conference where Gosling stated that “Java has always been as fast or faster than C++,” something that puzzles me to this day), Java is not the right solution for all problems.
Not Cross-Platform Enough
I struggled for years to deal with cross-platform problems for products like the Hands-On Java CD. It’s effectively the same as the RIA problem, because I want installation to be as easy as possible, I want everything to work seamlessly the first time, and I don’t want to be forced to deal with platform issues. Often my solutions would work, but sometimes customers would email saying that it wasn’t working on their machine, and I had no idea what the problem was. The best I could do is say “try it on a different machine,” and often that solved the problem…whatever the problem was. I never want to hear things like this ; I just want everything to work.
My primary goal has been to create a slide-and-audio content delivery system, like you see in the Hands-On Java CD ROM or Thinking in C.
Java, with its promise of write-once, run everywhere, was an attractive contender. Unfortunately there were, at least at the time, significant delays in Linux support (and to a lesser degree, Mac support). Linux and Mac users may be in the minority, but they are vocal.
The show-stopper was Java’s lack of support for MP3s and multimedia in general. As Dick Wall of the Java Posse has pointed out numerous times, the Java Media Framework (JMF) has been virtually ignored for many years. At the time I made my initial decisions, there was no support for any kind of compressed sound format (I would have been willing to use something other than MP3). Even now the only support for MP3s is through open-source software, which is great in concept, but I don’t want to test it and discover platform issues—again, I just want it to work ; the only thing I want to hear from customers is “great !”
It seemed like the only thing that might work was the RealPlayer, so I used that for the HOJ CD, 2nd edition. But the RealPlayer installation process is always trying to trick you into buying the paid version ; I must tell people how to find the free version. And it’s intrusive—it takes over MP3s even though you tell it not to.
And for all that, the RealPlayer is unreliable. The installation occasionally causes problems, and I get emails from people. I generally have no idea what’s causing the problem—which is frustrating—and the customer is usually convinced that there’s something wrong with the CD. The Daily Show, for years, used RealPlayer and it was not only painful because it was always starting and stopping (you couldn’t download everything beforehand ; it forced you to stream), but it always had these streaks on the left side of the picture. Now Comedy Central has changed to a new system but that only works intermittently. So I end up just waiting for them to be posted on YouTube.
The Flash Solution
In my own search, I’d really like such a system to solve not just some, but all of my UI problems. If I’m going to go to the trouble of learning it, I don’t want to run into a wall just when I start to develop some speed. That’s happened plenty of times already.
The only obvious solution is Flash. Flash has always been all about cross-platform multimedia experiences and user interfaces. People are very familiar and comfortable with Flash, and it is installed on almost all machines in the world. It’s trusted, stable and reliable.
Installation is a no-brainer for everyone. You don’t have to answer questions or do anything special ; it just works. That’s why it’s already ubiquitous. Current and future versions of Flash are now released concurrently for the three main platforms (yes, except for 64-bit Linux, but they’re working on it and that user base typically has more than one computer, and so has alternatives in the meantime). The standard Flash installation plays MP3s and various types of video, so you’re not stuck with “write-once, run everywhere…except for multimedia.”
And you can’t deny that Flash produces very pleasing user interfaces. What did Flickr and Picasa both useé Not Java, not Ajax, but Flash. Google videos, written with Ajax, must not have been going anywhere because they bought YouTube, which uses Flash. Even the most die-hard Swing fan secretly wishes their UI could look this good, especially without all the extra work Swing would require.
Solving the UI problem
One of the difficulties of GUI programming is choosing a GUI library. Sometimes there’s a standard library, but that can change. In Java, we started with AWT, which turned out to be a mistake so we had to be patient as Swing was developed, until IBM and Eclipse came along and offered us the additional choice of SWT. In Python, there are many GUI libraries, including the built-in Tkinter (which reduces the installation problem), WxPython, Qt, and more. Windows-specific libraries have similar choices, but if you want to create cross-platform applications, those libraries aren’t available.
If you explore these GUI libraries, as I have, you run the risk of getting broad knowledge that is not particularly deep. Each library requires significant effort to learn, and each seems to have its own philosophy regarding the activities that should be easy, and those that should only be possible. Each one looks at the world of GUI programming in a different way.
I’d like to be able to learn one solution and use it for all my applications. That way I could stop learning about GUI solutions and start acquiring some depth. Ideally, this will be a real programming language with consistent results across all platforms.
I believe that to solve the user interface problem, we need the equivalent of a domain-specific language dedicated to the user experience. For me, Flash-based technologies like Flex are the best solution to this problem. (Full disclosure : I’m in the process of working out a consulting contract with Adobe, to help them teach people about Flex. But long before this, I became convinced that Flash, and Flex in particular, was the best solution for the user-interface problem, and I began writing this article long before Adobe expressed interest in my assistance).
What is Flexé
Traditionally, Flash content and applications have been created with the Flash authoring tool, currently called Flash 8 Professional (it’s confusing, but the decision to give the name of the tool the same name as the runtime was made ten years ago). There was also a tool called Macromedia Director, a multimedia production tool for CD-ROMs, which predates Flash and outputs a format called Shockwave that runs in a Plug-In/ActiveX control much like Flash content, but is an entirely different control. Shockwave had its heyday, and it is still widely used, especially for games, but Flash is much lighter weight and achieved mainstream adoption far beyond Director.
Cost was one of the main things that originally held me back from using Flex, primarily because of readers who were unwilling or unable to pay it. In the initial versions of Flex, you had to buy the server edition just to create static SWFs. The server edition is designed for dynamic content, and is certainly worth the money for larger organizations that are creating dynamic SWFs from databases and the like. But there was no reasonable entry point for people who just wanted to experiment with it. It was hard for me to recommend it if the average person didn’t have a reasonable experimentation path, including creating static SWFs to deliver from their own server.
Now, however, you can download the free command-line Flex compiler to create static SWFs, and you can deliver these from your web site without paying any fees. The compiler, framework, and debugger are all free, so there’s no reason to avoid using Flex.
You can buy the Flex Builder IDE to help you create Flex apps. This is built on top of the Eclipse platform (instead of creating a new GUI development system from scratch—a wise approach). It has the usual sorts of things we’ve come to expect, like auto completion, context help, debugging, and even a GUI layout tool. The layout tool can give you a quick start when you’re beginning a design, although I found it more valuable to hand-tune the code once the rough draft of the design was in place.
Here’s something else that caused me trouble in the past : although the Flash player for Windows and the Mac have always been on the same release schedule, Flash for Linux has formerly been released quite a bit later. I didn’t know this until I put out the first beta of the Thinking in C eSeminar, and got complaints from Linux and Mac people about it. After some investigation, I decided to back-port the application (which isn’t that fancy, and doesn’t make use of any features that don’t exist in Flash 7). This seemed to be the best solution for me, because I didn’t have to wait for a final release of the newer version of Flash, and in particular I didn’t have to worry about Linux. One of my main goals for using Flash in the first place was to make the use of this application transparent across platforms, and to reduce any installation issues to a minimum.
However, with Flash 9 and beyond, all the players will be released within weeks of each other, and this policy should hold for future versions of Flash. So now you don’t have to worry about complaints from anyone. Build your UI with Flex, and it will “just work.”
Flex as a DSL for Graphics
One of the most appealing things about Flex is that Flash was created with the idea of UI first. In a very real sense, it’s a domain-specific language (DSL) for graphics, multimedia, and UIs, whereas most other solutions have been languages with UI libraries tacked on afterwards.
Because of this design goal, Flex and Flash provide a complete, unlimited, flexible tool to build user experiences. From the standpoint of a programmer’s time investment, you can learn a single language for building UIs without worrying about running into problems or limitations later—issues like :
Constraints on what you can create
Sudden steep climbs in the learning curve
There are plenty of fancy components that you can just drop in and use—the Flex Framework (part of the free download) comes with over 100 components. There’s an active marketplace for component creators, both open-source and payware. One such library comes from Adobe : the Flex Charting Components (in the few hundred dollar range), but there are also competing charting components.
One of the more interesting aspects of Ajax, of course, is the “A” for “asynchronous.” This allows information to flow between client and server without doing a whole page refresh. For Flash, a much more sophisticated version of this is provided by Flex Data Services, a publish/subscribe API for data management. Flex Data Services automatically performs caching and updating between client and server, to produce an optimal experience without forcing you to write extra code. This allows you to work with real-time data, build collaborative applications, and perform enterprise messaging integration. You can use Flex Data Services on a single CPU for free ; if your application requires multiple CPUs you’re considered an enterprise and the licensing fee kicks in. You can learn more about Flex Data Services here.
Earlier I mentioned Gliffy, which was built using OpenLaszlo. OpenLaszlo was especially attractive before the Flex compiler and framework became free. However, the OpenLaszlo group has decided that they will accommodate the lowest common denominator technologies by targeting DHTML combined with Flash, and this incorporates the limitations of DHTML. The reason Flex is compelling to me is that it allows me to take advantage of things I can’t do in the normal browser and those things produce the same results, everywhere. On top of this, Flex is magnitudes faster than OpenLaszlo because it takes advantage of the JIT compiler in Flash 9. Since Flex is now free, there’s no reason not to go to the source.
Flex on the Desktop
Of course, if my dream is to be able to learn a single GUI system in depth, is Flex the right tool, since it was originally designed for web RIAsé
A Flex UI can initiate communication with its server, or any other server it chooses. A server cannot initiate communication with a Flex UI, which makes sense because of security (it would be effectively the same as having an open port on your machine).
However, a Flex UI is not limited to communicating with a server. It can also communicate with a local application. Thus, you can create an application in any language you prefer, even a dynamic language like Python or Ruby, and use Flex to build a beautiful UI.
Adobe is developing a new tool called Apollo, a cross-OS runtime that allows you to use Flex to create desktop RIAs. This means that your Flex skills can be further employed for creating smooth desktop applications, but it also means that you can more easily build applications that will work on both web and desktop (I’ve seen expensive and difficult-to-use tools that allow other languages to do this).
It’s clear that we can’t wait for Sun to fix all of Java’s problems. Open-sourcing Java might, eventually, have a huge impact on repairing Java’s deficiencies. For example, work on the JMF might get resurrected. Maybe installation issues will even be fixed someday. The possibilities might be limitless, but if you need to solve problems now, then the solution is to hybridize parts of the language.
We do this already. You don’t insist on using a Java database for an application ; you use a specialized system like MySQL or Oracle. Sun is directly supporting the development of JRuby for hybrid Java/JRuby programming. We are seeing other special-purpose languages arise to solve specialized problems. Why insist on using a Java library for UI if a specialized system solves the problem betteré
As shown in the TurboGears-Flex demo I created with James Ward, it’s possible to use a language like Python (or Java, Ruby, C#, etc.) as a back-end and build your user interface using Flex. This can even be achieved for desktop applications (even more so with the upcoming Apollo tool).
You can learn all about Flex, and download it, at http://www.flex.org/. This is an incredibly comprehensive site with lots of examples, tutorials and screencasts. They even have an online Flex compiler for instant experimentation. There’s an in-depth presentation that James Ward gave about developing with Flex. And here is a screencast that I did with James Ward, that shows how to use Flex as a front-end for Java server applications. You can now create a low-cost, powerful combination of Java on the server (which can of course be running Linux on commodity hardware) and an interactive Flash interface on the client.