« Return to Thread: [rvm-core] Code freeze

Re: [rvm-core] Code freeze

by Ian Rogers (nabble) :: Rate this Message:

Reply to Author | View in Thread

2009/2/3 Steve Blackburn <Steve.Blackburn@...>:
> I think it would be sensible to have a temporary freeze on the trunk,
> avoiding all non-essential commits.

I think this is unnecessary, making changes relating to the thread
model currently would be a bad idea though.

> 1.  There are at least two critical, apparently unrelated performance
> bugs outstanding:
>
>        http://jira.codehaus.org/browse/RVM-758
>        http://jira.codehaus.org/browse/RVM-762
>
> Performance is also significantly (but less dramatically) down on
> jbb2000, luindex, jess, db and javac.  Most of the dips appeared
> around 15301 or so, though it is odd since cattrack says that there
> were two perf regressions against 15301, one of which was fine, and
> the other of which was slow (things have remained slow since 15301).
> As a precaution, Daniel has rebooted habanero (the relevant test
> machine).   Anyway, this is disturbing and I'd really like to see it
> figured out before we move forward.

I'm working on it. We have a situation where we are generating less
and tighter code, but performance has regressed. I profiled JBB'05 and
there are no real clues as to where the performance has gone. The
performance of jack is harder to monitor as it completes in a matter
of seconds. On the plus side performance is up for jython, lusearch
and to a lesser extent chart and javac.

We are getting noise on the AIX PPC tests where we appear to be
failing because another process has stolen a CPU.

> 2. Meanwhile Daniel and Fil are trying to get native threads ready for
> prime time and are now very close, so minimizing churn on the head
> will help a great deal.

I'm really worried again by the work on PNT (that I really want to see
in the head). Reading the JIRA traffic I see that we have a new method
called lockDangerously, when would someone want to lockDangerously
rather than just lock? I think the PNT code is quite a way from being
production ready. In doing some donkey work for the PPC implementation
the documentation wasn't sufficient for me to avoid bugs. Some methods
are lacking documentation and a lot of what should be self documenting
isn't. Recent commits on the PNT branch have backed out assertion
checks and had the infamous commit message of "stuff". This is really
not good enough.

I think a code freeze waiting for PNT is going to lead to a protracted
code freeze. I'm avoiding work on Windows and Harmony because it will
conflict with the PNT work. I don't think we should shut down
development of the rest of the code base as well.

> I would also like to flip production to GenImmix, but to monitor
> performance etc, it will help if things are stable, though this does
> not require a code freeze, and I'll wait till the performance issues
> are ironed out before I do that.

This may be a good use of the quarantine branch.

Regards,
Ian

> Cheers,
>
> --Steve
>
> ------------------------------------------------------------------------------
> Create and Deploy Rich Internet Apps outside the browser with Adobe(R)AIR(TM)
> software. With Adobe AIR, Ajax developers can use existing skills and code to
> build responsive, highly engaging applications that combine the power of local
> resources and data with the reach of the web. Download the Adobe AIR SDK and
> Ajax docs to start building applications today-http://p.sf.net/sfu/adobe-com
> _______________________________________________
> Jikesrvm-core mailing list
> Jikesrvm-core@...
> https://lists.sourceforge.net/lists/listinfo/jikesrvm-core
>

------------------------------------------------------------------------------
Create and Deploy Rich Internet Apps outside the browser with Adobe(R)AIR(TM)
software. With Adobe AIR, Ajax developers can use existing skills and code to
build responsive, highly engaging applications that combine the power of local
resources and data with the reach of the web. Download the Adobe AIR SDK and
Ajax docs to start building applications today-http://p.sf.net/sfu/adobe-com
_______________________________________________
Jikesrvm-core mailing list
Jikesrvm-core@...
https://lists.sourceforge.net/lists/listinfo/jikesrvm-core

 « Return to Thread: [rvm-core] Code freeze