[kaffe] Re: good day for kaffe 1.1.4 and ofbiz 2.1

[Dalibor Topic]

Hi Adam,

Adam Heath wrote:
> kaffe 1.1.4 can now run ofbiz 2.1.  I haven't tried a later version of ofbiz.

Thanks for the great news! Could you add the information on how to get 
it to run to the Moving Java to Main wiki on http://java.debian.net ?

> There are some issues, however.
> 
> Startup time under sun 1.4 is 11s.  kaffe, 15.2.

You could try running kaffe with -prof to see where it spends its time 
in the class library. There may be some hot spots that could be 
implemented better.

> Kaffe complains about an invalid bytecode at startup, while sun doesn't.
> 
> WARNING Bad bytecode! Illegal exception table entry: start_pc=140851621 is not
> lower than end_pc=140851621 in method
> org/mortbay/util/Resource.newResource((Ljava/net/URL;)Lorg/mortbay/util/Resource;)
> See Java Virtual Machine Specification 2nd Edition $4.7.3 for details.
> Please report this bug to the developers of the application you're running on
> kaffe.
> A simple fix might be to use another java compiler to build the application.

That's a bug in Sun's VM and the compiler used to generate the JAR file. 
The virtual machine spec says that the start PC of an exception table 
entry must be lower than the end PC. Sun's VM seems to ignore such bad 
bytecode. In order to stop people from filinf bug reports against Kaffe 
for not being able to run broken bytecode, we ignore it too, but 
nevertheless print a warning message so that the bug can be fixed by the 
upstream developers by picking a better compiler.

> Also, shutdown hooks in kaffe are not run.

OK, that's bad. Do you have a small (5-10 lines) test case?

> Attached you will find the entity engine performance test, which is available
> under webtools.

Seems that kaffe is slower than JDK at the tests. That's something that 
can be fixed though, given enough interested developers ;) For example, 
kaffe's jitter is rather simplistic, and could generate better machine 
code given some effort.

cheers,
dalibor topic