what's the brief grand-plan for java ?
 

What's the three liner for java openjdk/icedtea/libgcj in the fedora
future ? is it...

Port icedtea to all platforms that libgcj supports, move to icedtea as
default when stable and drop ahead-of-time-gcj-compiling ?

Or is it planned to always aot compile stuff by default and continue to
support using gij as java ?

C.

--
fedora-devel-list mailing list
fedora-devel-list@redhat.com
https://www.redhat.com/mailman/listinfo/fedora-devel-list

what's the brief grand-plan for java ?
 

Caolan McNamara wrote:
> What's the three liner for java openjdk/icedtea/libgcj in the fedora
> future ? is it...
>
> Port icedtea to all platforms that libgcj supports, move to icedtea as
> default when stable and drop ahead-of-time-gcj-compiling ?

Eventually, yes. The most interesting option is to use LLVM to jit-compile;
see http://gbenson.net/ for the current status.

> Or is it planned to always aot compile stuff by default and continue to
> support using gij as java ?

It all depends on how well the LLVM-based jit works.

For example, gcj is currently between 1.6 and 10 times faster than
Sun's C++ interpreter on ARM systems, so it's worth keeping gcj alive
there until we get something better. Only desktop systems fully-
supported by a custom JIT, there is little need for gcj.

Andrew.

--
fedora-devel-list mailing list
fedora-devel-list@redhat.com
https://www.redhat.com/mailman/listinfo/fedora-devel-list

what's the brief grand-plan for java ?
 

On Fri, May 9, 2008 at 2:56 PM, Caolan McNamara <caolanm@redhat.com> wrote:
> What's the three liner for java openjdk/icedtea/libgcj in the fedora
> future ? is it...
>
> Port icedtea to all platforms that libgcj supports, move to icedtea as
> default when stable and drop ahead-of-time-gcj-compiling ?
>
> Or is it planned to always aot compile stuff by default and continue to
> support using gij as java ?
>
> C.
>

Or secret option number three: Getting openjdk to support compiling
aot before dropping support for gij?


--
Mat Booth
www.matbooth.co.uk

--
fedora-devel-list mailing list
fedora-devel-list@redhat.com
https://www.redhat.com/mailman/listinfo/fedora-devel-list

what's the brief grand-plan for java ?
 

Mat Booth wrote:
> On Fri, May 9, 2008 at 2:56 PM, Caolan McNamara <caolanm@redhat.com> wrote:
>> What's the three liner for java openjdk/icedtea/libgcj in the fedora
>> future ? is it...
>>
>> Port icedtea to all platforms that libgcj supports, move to icedtea as
>> default when stable and drop ahead-of-time-gcj-compiling ?
>>
>> Or is it planned to always aot compile stuff by default and continue to
>> support using gij as java ?

>
> Or secret option number three: Getting openjdk to support compiling
> aot before dropping support for gij?

That's right. On some systems the advantage of precompilation could
be considerable. At the moment, though, LLVM looks more promising.

Andrew.

--
fedora-devel-list mailing list
fedora-devel-list@redhat.com
https://www.redhat.com/mailman/listinfo/fedora-devel-list

what's the brief grand-plan for java ?
 

Andrew Haley <aph <at> redhat.com> writes:
> That's right. On some systems the advantage of precompilation could
> be considerable. At the moment, though, LLVM looks more promising.

Couldn't LLVM be used for AOT-compilation too?

Kevin Kofler

--
fedora-devel-list mailing list
fedora-devel-list@redhat.com
https://www.redhat.com/mailman/listinfo/fedora-devel-list

what's the brief grand-plan for java ?
 

Kevin Kofler wrote:
> Andrew Haley <aph <at> redhat.com> writes:
>> That's right. On some systems the advantage of precompilation could
>> be considerable. At the moment, though, LLVM looks more promising.
>
> Couldn't LLVM be used for AOT-compilation too?

Sure could. I don't know if it would work as well, though, and
gcj is a very well-proven stable compiler.

Andrew.

--
fedora-devel-list mailing list
fedora-devel-list@redhat.com
https://www.redhat.com/mailman/listinfo/fedora-devel-list

what's the brief grand-plan for java ?
 

On Fri, May 9, 2008 at 10:05 AM, Andrew Haley <aph@redhat.com> wrote:
>
> That's right. On some systems the advantage of precompilation could
> be considerable. At the moment, though, LLVM looks more promising.

More important than compilation model in my mind is if we want to
continue supporting GCJ, is for it to use the OpenJDK class library;
when I tried GCJ in the past I had problems with incompleteness or
bugs in Classpath, I never had an issue with the way code was
generated.

--
fedora-devel-list mailing list
fedora-devel-list@redhat.com
https://www.redhat.com/mailman/listinfo/fedora-devel-list

what's the brief grand-plan for java ?
 

Colin Walters wrote:
> On Fri, May 9, 2008 at 10:05 AM, Andrew Haley <aph@redhat.com> wrote:
>> That's right. On some systems the advantage of precompilation could
>> be considerable. At the moment, though, LLVM looks more promising.
>
> More important than compilation model in my mind is if we want to
> continue supporting GCJ, is for it to use the OpenJDK class library;

That is a big task. The simplest way to do it, I suspect, is to enable
the OpenJDK runtime to load gcj classes rather than to adapt the OpenJDK
class library to gcj. This sounds rather counter-intuitive, but the
interface between class library and virtual machine is not at all
portable or well-documented, and it is complex.

> when I tried GCJ in the past I had problems with incompleteness or
> bugs in Classpath, I never had an issue with the way code was
> generated.

I agree.

Andrew.

--
fedora-devel-list mailing list
fedora-devel-list@redhat.com
https://www.redhat.com/mailman/listinfo/fedora-devel-list