From: Charles Oliver Nutter <charles.nutter@...>
Date: 2007-12-07T02:07:20+09:00
Subject: Re: Ruby App Distribution

Michal Suchanek wrote:
> However, the one-click-jar is far from portable. It uses Sun
> proprietary solution which runs "everywhere" (read: on the platforms
> for which they bothered to make a binary). It runs on Windows, at
> least the recent NT based. On OS X but it's usually a bit behind when
> new version is out, especially for people that do not have the newest
> OS X version as well. On Linux it runs on x86 and x86_64. It also runs
> on Solaris, and probably HP-UX. There is some ancient and incomplete
> port to ppc-linux by IBM, probably as a result of support for their
> something360.

OS X is one version behind, and will probably have a full Java 6 release 
soon. Java 6 versions exist for Linux 32/64, Solaris, and Windows. And 
there are at least two alternative Java 5 VMs for Linux, Solaris, and 
Windows...often many more. There's dozens of Java implementations, and 
Java 7 source is GPL and rapidly being ported to platforms that don't 
have up-to-date JVMs (the BSDs, OS X, and more).

> If you did not require Swing there are alternative VMs that are free
> (and free to compile on anything) but then you would be back to
> distributing binary extensions because you would need something like
> AWT which is not part of the Java standard library. Also these free
> alternative VMs do not have complete standard library. As far as I
> understand the issue the Java documentation license does not allow
> implementing a library complying with the documentation which is the
> primary reason why efforts at implementing complete alternative VMs
> did not get very far.

AWT is part of the standard library. And Sun's JVM has been free for 
years, and now Free in Java 7. You are wrong about any documentation 
licenses...I'm not sure what you even mean by that. And as I mentioned 
before, there's dozens of JVM implementations. I'm not sure what more 
you're looking for here.

> Then there is that "write once run everywhere" thing with which Sun
> tries to push Java. If I have never heard that I would probably not
> get so much irritated whenever I hear about Java portability.
> 
> Java software is *NO MORE* portable than C software. Given the
> proprietary nature of the VM it is actually *LESS* portable. It has
> all the common portability issues plus it won't run on platforms for
> which there is no JVM, ever.

C code would not run on platforms where there's no C compiler...but even 
worse, a given app won't run unless all libraries it uses have versions 
for that platform as well. There's certainly a "debug everywhere" 
component to Java apps, but once you compile them they're basically 
runnable on any Java implementation. C apps, not so much.

But you're still thinking of JRuby in terms of Java here. JRuby is not 
Java. JRuby is Ruby on the JVM. The JVM exists in dozens of forms, free, 
proprietary, embedded, and on big iron. With the exception of some of 
the smaller embedded profiles and some of the incomplete free impls, 
JRuby allows Ruby to run on any of them...faster than Ruby 1.8.6, and 
with all the "good bits" of the Java platform available. Don't think of 
JRuby in terms of Java.

> At least the Java software you meet in practice is swamped with
> compatibility problems. Most Java software requires specific minor
> version of the JVM, and with this one click approach you can hardly
> install several versions of the JVM. Yes, the JDK tools have the
> option to compile for older runtimes. I have yet to see this working.
> When java 1.5 came out, and our CMS software developer upgraded,
> everybody had to upgrade their runtime. He said he is compiling with
> the compatibility switch but it just did not work. Maybe it's just the
> Java crowd being mostly incompetent. I don't know.

I think you're just misinformed, or getting your information from bad 
sources. It's not anywhere near this difficult...no more difficult than 
apps that target a specific C library version. Except in this case, you 
know that one version numbers guarantees you a specific version of every 
core API you're going to use, from database access to GUIs.


> What's worse, I have seen a banking application that requires a
> specific patchlevel version (yes, specific -pn suffix). This same
> application also works only in IE, how portable. Even for applications
> that try, I have seen subtle input event handling differences between
> Windows and Linux, enough to break stuff. And an application that
> would not have sound on Linux - not sure if it simply did not work or
> it caused so much trouble they decided to turn it off.

The patchlevel thing sounds like they probably hit a bug in that impl of 
the JVM and needed an updated version. That happens to any software.

The IE thing is a web/UI design issue completely unrelated to Java.

Sound on Linux has been a traditionally hard thing to get working for 
any application, so Java's certainly not alone here either. And the 
Linux situation has improved for all apps. You're going on really old 
information.

> With these version and platform constraints you may need a quite large
> vmware stack if you want to run a number of Java applications :->

Ask anyone who's actually run a lot of Java applications recently, and 
you'll get a completely different story. You're either getting bad 
information or only looking for bad information.

And again, please don't think of JRuby in terms of Java.

> hmm, anyway, thanks for bringing ruby even to such a hostile
> environment for the benefit of those stuck in there ;-)

It's not just for those stuck on Java, it's for those who want a fast 
Ruby 1.8 implementation plus a whole lot of other bells and whistles.

- Charlie