Re: [Evolution-hackers] vfolder performance problems

[Lee Revell <rlrevell joe-job com>]

On Tue, 2005-04-19 at 18:19 +0800, Not Zed wrote:
> > I don't see the need to assign copyright as it's basically three one
> > line fixes.  I would just like to be credited in the changelog for
> > fixing this bug.
> 
> Well in that case you should supply the changelog!  Changelogs are also
> part of the patch guidelines anyway.
> 
> So are "no c++ style comments" (or 'c89'), which your code includes.
> 

OK, I'll fix it.

> > The "hidejunk" fix is a temporary hack until someone implements it in an
> > efficient way.  I think it would be sufficient to add a "hasjunk" flag
> > to the folder summary so we don't have to check the junk flag on every
> > single message in every folder with every message list update.  Or make
> > "hidejunk" configurable and default to off.
> 
> Well, of course, until the full functionality is maintained, the patch
> can't be applied anyway.
> 
> I don't think the hasjunk idea is a very scalable one either.  Nor does
> it really help - we have to test every flag of every message every time
> we build the message-list tree model anyway.  So I'm wondering whether
> your patch is just hiding another problem (the lock contention one), and
> not really making the big performance difference it appears to be.

Then disregard the "hidejunk" part of the patch.  The impact of that
part is minor.  The massive speedup comes from not searching every
header for a null string on every message list update.

Lee