[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
[
List Home]
|
Re: [mdt-ocl.dev] Hudson build became unstable: buckminster-mdt-ocl-core-3.1-nightly #502
|
Folks,
as you may have noted, the impact analyzer comes with a number of
different configurable algorithms that we benchmarked intensely for
various situations. After all evaluations, for the standard cases and in
almost all situations the TracebackStep variant came out best.
NavigationStep performed slightly worse. We still left the
NavigationStep code in for interested parties to review and because we
still think that if someone spends some time optimizing the navigation
step graph layout there may be potential for this variant to perform
better than the TracebackStep approach.
One of the weak parts of the algorithm, as you have discovered now, is
the need to construct subclass trees in an attempt to prune navigation
step graphs early in case certain combinations of type filters can be
proven to never match an EObject. However, this assumes (and this may be
an assumption too risky) that we can determine full subclass trees ahead
of time.
That's the background on why the NavigationStep implementation tries to
compute the set of all of a class's subclass which then triggers the
load of all packages.
I'm fine with excluding all NavigationStep-based tests from Hudson right
now. Shall I remove the respective launch configurations or Adolfo, can
you do this (I'm currently traveling and have very unreliable Internet
connections unfortunately; back in office on Friday).
Sorry, thanks and best,
-- Axel
On 5/19/2011 6:42 AM, Ed Willink wrote:
Hi
See https://bugs.eclipse.org/bugs/show_bug.cgi?id=346368.
Users of the IA need a fix since reverting pivot.jar is not the
solution; it just demonstrates that the algorithm is unsound in larger
modeling environments.
Regards
Ed
On 18/05/2011 22:01, Ed Willink wrote:
Hi
This is wierd.
Surely caused by the changed of classpath for
https://bugs.eclipse.org/bugs/show_bug.cgi?id=346272.
But how can the pivot classpath affect impact analyzer tests, and why
just 15 out of 250 odd?
?? Is some access to the global package registry affected ??
Regards
Ed
On 18/05/2011 21:36, hudsonbuild@xxxxxxxxxxx wrote:
See<https://hudson.eclipse.org/hudson/job/buckminster-mdt-ocl-core-3.1-nightly/502/>
-----
No virus found in this message.
Checked by AVG - www.avg.com
Version: 10.0.1375 / Virus Database: 1509/3645 - Release Date: 05/18/11
