[fedora-java] Re: blueMarine and openJDK -- second attempt

[Mark Wielaard]

Hi Fabrizio,

On Tue, 2008-08-05 at 16:46 +0000, Fabrizio Giudici wrote:
> Thank you very much for following up, Mark. I'm using OpenJDK build 1.6.0-b09 -
> unfortunately I can't tell which is the output of rpm since I'm testing openjdk
> with Ubuntu - I don't have a Fedora box at hand right now.

OK, that explains things. Ubuntu clearly packaged a really old
icedtea/openjdk version. I thought you were using Fedora since you said
"Since OpenJDK passed the Test Compatibility Kit (TCK), this means that
some parts of imaging elaboration are not covered by it - and I'm really
puzzled about that" and only the binary version shipped with Fedora 9
has ever passed the TCK (this is indeed a problem, the TCK is
proprietary and not available to the community at large except under
NDA, which indeed means you cannot in general rely on any openjdk
derivative to having passed the TCK... sigh).

> I was confused too as I saw that one of the bugs has been fixed time ago - I
> suspect it could be a similar bug, but slightly different. I'm working to get a
> stack trace and post to the issue list at launchpad.

I am pretty sure it is that old issue. b09 is just really, really old
(April 2008). And the patch that added the new color profile support was
added after it was released. Current icedtea/openjdk is 1.2+hg/b11,
although fedora actually packages a slightly newer version (we should
release 1.3 for real I guess). The 1.2/b10 version was the version that
formally passed the TCK on x86/x86_64 fedora 9.

> May I ask another thing? I've also run a few performance benchmarks side-by-side
> with Sun's JDK 6 on Ubuntu, and OpenJDK seems to be really really slower, at the
> point that I think I'm doing something wrong. I'll post in a few hours my
> findings on my blog, but in the meantime is there anything I should know, such
> as special command line switches etc? AFAIK OpenJDK runs the same HotSpot
> compiler than Sun's JDK 6, or am I wrong?

I don't know if anybody did any benchmarks to compare the two. They
should not really differ noticeably in terms of speed, since the
underlying code is almost similar. But till now we have been focused on
completeness and correctness. If someone has repeatable benchmarks that
show icedtea/openjdk being really slow that would be interesting.

Cheers,

Mark