RE: [mylar-dev] mylar filter and make less interesting are brokenin0.4.2

["Mik Kersten" <beatmik@xxxxxxx>]

Thanks for seeing this through Eugene and identifying the regression
(0.4.2.1 is now available on the update site).  It turned out that the
'improved' support for the predicted interest of errors was causing a
serious performance issue on workspaces with lots of resources and markers
by doing a marker check on every element when deciding whether or not to
filter it (in general marker checks are slow).  Also, it was making it
impossible to reduce the interest of the elements with markers.  It wasn't
an issue on my workspace or on the test workspaces, so the lesson learned
here is that we need to start adding performance tests that would identify
these kinds of regressions (new bug 116487).  Eclipse has some great
infrastructure for that.

The existing support for predicted error interest has always resulted in too
much Package Explorer refresh, and that's now indicated on the preference
option that toggles it.  This will get revamped for 0.4.3 (bug 115795).

Mik

> -----Original Message-----
> From: mylar-dev-bounces@xxxxxxxxxxx [mailto:mylar-dev-bounces@xxxxxxxxxxx]
> On Behalf Of Eugene Kuleshov
> Sent: November 15, 2005 10:28 AM
> To: Mylar developer discussions
> Subject: Re: [mylar-dev] mylar filter and make less interesting are
> brokenin0.4.2
> 
> Mik Kersten wrote:
> 
>   Thanks Mik. Some comments below.
> 
> >What I'm understanding is that the "make less interesting" behavior is
> >what's blocking you from using 0.4.2.  Once that bug is fixed I will put
> a
> >build into the "test" update site and put a note on the corresponding bug
> >report.
> >
> >But since I can't reproduce this, and since that implementation hasn't
> >changed since 0.4.1 and you seem to only be seeing the failure on 0.4.2,
> >what would help get this resolved is if you could make a new task context
> >using 0.4.2, verify whether the action works, then go to the task context
> >that it was failing on, and post on the report what it works on and what
> it
> >fails on (e.g. fails on a file, works on member).
> >
> >
>   I just tried it on a different PC with 3.2M3 isntalled and got the
> same effect of random crazy scrolling trough the package explorer and
> expanding bunch of nodes druring that.
> 
>   The only commonality I see is that task context has resources from two
> projects and these projects are in a different working sets (my package
> exprorer is in a working set groupping mode) and there aer lot of other
> projects between these two... Hope this helps.
> 
>   I also noticed that there is some huge performance degradation when I
> activate mylar filter on a package explorer. It is not onlys crolling
> and randomly jumping upon activation deactivatino but also freezes the
> workbench window for quite significant time (up to 10 seconds). I have a
> really big workspace here (about 50 projects not counting closed ones).
> 
>   regards,
>   Eugene
> 
> 
> >>-----Original Message-----
> >>From: mylar-dev-bounces@xxxxxxxxxxx [mailto:mylar-dev-
> bounces@xxxxxxxxxxx]
> >>On Behalf Of Eugene Kuleshov
> >>Sent: November 15, 2005 8:50 AM
> >>To: Mylar developer discussions
> >>Subject: Re: [mylar-dev] mylar filter and make less interesting are
> >>brokenin 0.4.2
> >>
> >>Mik Kersten wrote:
> >>
> >>
> >>
> >>>> It looks like there is something seriously wrong with 0.4.2 release.
> >>>>
> >>>> Now "make less interesting" action does not work at all.
> >>>>
> >>>>
> >>>This implementation of this hasn't changed, so I'm wondering if you're
> >>>seeing https://bugs.eclipse.org/bugs/show_bug.cgi?id=108309 which
> >>>
> >>>
> >>results
> >>
> >>
> >>>from a currently selected element becoming selected again right after
> >>>
> >>>
> >>you
> >>
> >>
> >>>hit Make Less Interesting.  The usual work-around is to hit Make Less
> >>>Interesting twice.  If that doesn't work please comment.  I've bumped
> >>>
> >>>
> >>the
> >>
> >>
> >>>priority of this bug to P1.
> >>>
> >>>
> >>   No, twice, triple or 10 times does not remove items out of
> >>interest, at least in the package explorer. The same for keyboard
> >>shortcut and popup menu.
> >>
> >>
> >>
> >>>> I also noticed some weird scrolling in package explorer on filter and
> >>>>task activation. The worst pa
> >>>>
> >>>>
> >>>This is almost certainly multiple editors being opened active on task
> >>>activation while "Link with Editor" is pressed on the Package Explorer.
> >>>If you could verify that the fix should be very straightforward (to
> >>>temporarily suppress Mylar's Package Explorer management while
> >>>
> >>>
> >>activating
> >>
> >>
> >>>the task).  New bug is:
> >>>
> >>>
> >>https://bugs.eclipse.org/bugs/show_bug.cgi?id=116449
> >>
> >>
> >>
> >>>> Only those issues make Mylar unusable.
> >>>> So, rolling back to 0.4.1 :-(
> >>>>
> >>>>
> >>>Assuming that the former is not a regression, and the latter is
> >>>
> >>>
> >>requested
> >>
> >>
> >>>behavior that can be toggled off, if I'll avoid cutting a service
> >>>
> >>>
> >>release.
> >>
> >>
> >>>But I would appreciate your trying 0.4.2 again to verify.
> >>>
> >>>
> >>   Not sure what exactly you want me to verify?
> >>
> >>
> >>
> >>>> Also note that if I exclude resource from interest it does not get
> >>>>removed from the mylar's commit set. That was happening in
> >>>>synchronization that runs only on a subset of workspace (working set).
> >>>>
> >>>>
> >>>Created a bug: https://bugs.eclipse.org/bugs/show_bug.cgi?id=116450
> >>>The work-around should be to deactivate and reactivate the task
> context,
> >>>which will cause the commit set to get recreated without the
> >>>
> >>>
> >>uninteresting
> >>
> >>
> >>>resource.  But could you specify on the report precisely what you mean
> >>>
> >>>
> >>by
> >>
> >>
> >>>"exclude resource from interest"?
> >>>
> >>>
> >>   Yes, it is "make less interesting". I somehow managed to push it
> >>trough the above issue (once) and file did disappear from the filter
> >>but not from the commit set. Another thing is that commit set is not
> >>always removed on task deactivation. So, I see an empty commit set in
> >>some cases. Can't give more details now because I can't work with
> >>0.4.2 in its current state.
> >>
> >>   regards,
> >>   Eugene
> >>
> >>
> >>
> 
> _______________________________________________
> mylar-dev mailing list
> mylar-dev@xxxxxxxxxxx
> https://dev.eclipse.org/mailman/listinfo/mylar-dev