On 02/02/2012 10:43, Alan Bateman wrote:
On 01/02/2012 18:26, Volker Simonis wrote:
I think the prerequisite for a port can not be that there exists
access to it on "x86 hardware".
With this constraint there will never be an OpenJDK port to other
architectures like PPC/Itanium/ARM to name just a few.
(By the way, there once upon a time was a mips -porting project
http://openjdk.java.net/projects/mips-port/ which was hosted at
http://icedtea.classpath.org/hg/openjdk6-mips and which seems to be
dead now although it probably didn't run on x86 hardware:)
On the other hand, I think it is valid to request build/test access to
a platform if there is a port for that platform. However I think the
modalities for such an access should be handled by the group which
leads the port.
The real problem with such an access is that the current Oracle
build/test infrastructure is NOT open and does not easily allow the
integration of external machines. This is a pain point since long time
and we constantly here that Sun and now Oracle is working on it (don't
want to blame anybody here!) - but it's just the way how it currently
works. After all it seems that we get a new bug tracking system REAL
SOON now. The next steps would be a new review system AND of course an
open, extensible(from outside Oracle), automatic build and test
system.
But until we get there, I think an AIX port project would be the
appropriate way to handle such changes. Of course IBM would be free to
make some build/test hardware/infrastructure available and if that
works pretty well it would probably accelerate the integration of the
port into the mainline.
This thread highlights that we need to figure out how OpenJDK can
accommodate additional ports going forward. I would be interested to
hear how other projects do this. I've heard of projects that have a
core set of platforms and then additional teams of volunteers that
keep ports that aren't as main stream up to date. If OpenJDK takes on
too many platforms then there is the risk that we spend a lot of time
just trying to keep things stable and it also means needing ready
access to every platform (I suspect this is what Phil is concerned
about). There are probably other things we can learn that would help
with the code layout too (the jdk repo has to change).
Yes, agree - we do need to figure this out. There is a balance to be
met between mainstream and non-mainstream.
I will point out though that everyone having ready access to every
platform is not a prerequisite to being mainstream. Even if hardware
is "readily" available it would not be a panacea. A developer would
still need some level of platform knowledge. If you look at this from
another angle - as developers we all have particular areas of expertise
and areas we no less well. I certainly wouldn't want to claim
expertise in the intricacies of Window Managers and I would therefore
defer to the experts. It the same for platform specific issues.
The idea is that as a community we work as a group to ensure that it
all comes together. What makes a platform mainstream is driven most by
the number of people actively working to develop and maintain it.
The changes for AIX are small and there are a number of people who are
willing to provide the necessary long term support. So I'm struggling
with why we can't just get these changes into OpenJDK directly.
