[Xastir] Memory question

[Tom Robson, ve7did]

On September 29, 2004 10:53 pm, Tom Russo wrote:
> On Wed, Sep 29, 2004 at 09:47:36PM -0700, a Mr. Richard Feyler of Fort Lee, 
New Jersey <ve7did at dccnet.com> writes 'Dear Rosanne Rosannadanna':
> > > Are you running the CVS version of xastir?  When was the last time you
> > > built it?
> >
> > Yes, I am using cvs... built on Sept 13th.
>
> Hmmm.  Odd.  That rules out the old dbfawk leaks, then --- those were
> cleared up months ago.
>
> > > There could be a new memory leak in some aspect.  How many shapefiles
> > > do you load?  Do you use dbfawk or do you configure with
> > > --disable-dbfawk? Have you tried running with a lighter-weight desktop
> > > than KDE?  Is the behavior the same?
> >
> > I don't routinely run shapefiles because the repositioning is a bit slow,
> > but use them quite often when I want some street detail etc.  I have run
> > about 6 shapefiles but they are about 20Mb each (the *.shp files).  I do
> > use dbfawk for gathering street names.
>
> That's a very small number of shapefiles to take up that much core.  I have
> all the tiger shapefiles for my state, and frequently zoom out to state
> level.  Most are under 10MB, but a few are as large as 14MB, but there's
> 33 of them.  Even when I'm just zoomed in to my own area for a long time
> I've still got about 10 shapefiles on screen (my county shapefile, a bunch
> of USGS data, the National Atlas wilderness area map, a USFS trail map, and
> assorted point shapefiles).  I've never seen the code go as high as 262MB
> even when dbfawk was leaking core like a sieve.
>
> Are your 6 20MB shapefiles all visible at once or are some of them
> completely off screen?
>
> > I should have mentioned when the maps consume the memory initially and if
> > I deselect them (memory useage does not drop), then select them again the
> > memory useage does not go any higher.  It is as if the maps have reserved
> > the memory space for each map.
>
> Well, that's certainly not *explicitly* the case in xastir --- the maps are
> loaded one at a time, painted into the fixed pixmap, the map is closed, and
> then after all maps are drawn into the pixmap the pixmap is blatted to the
> window.  But it does, of course, have to allocate some memory as it reads
> things in.   Some of what you're seeing (the bit where select, deselect,
> select again neither reduces nor increases memory use) makes sense --- when
> maps are first loaded the code asks for memory from the OS, gets it, uses
> it, and "frees" it, but the memory isn't actually returned to the OS --- 
> it's just marked as free so that when the code asks for more later it gets
> bits of the freed memory first before the OS gives more.  But there's no
> way that accounts for the degree of bloat you're seeing.
>
> I'm interested to see whether your behavior is significantly different with
> a light-weight desktop environment like fvwm or icewm.  My experience with
> KDE is not extensive --- just enough to know it's too bloated for my
> tastes.

Built the latest cvs and no change.

Ok, think I found the problem.  It looks like kde is causing most of the 
problem.  Obviously (now) the one I tried yesterday was not lightweight.

With these large shapefiles KDE is consuming 45M of additional memory whereas 
fvm is consuming a bit more than 1M with the same xastir parameters.  Even 
though fvm consumes little additional memory it still does not release the 
memory when the maps are deselected.

Does yours do the same?

I let kde run overnight and the memory useage seemed to have climbed quite a 
bit.  Not too sure, but I'll keep closer tabs on it over the next few days.  
I think because kde consumes so much memory this slow increase eventually 
filled things up.

Tried gnome, and it was worse than kde.

Gees.. hate to give up kde as I really like it.....  perhaps prettier and more 
built in functions  is not always better...

...73 Tom


>
> Looking for other possible culprits: what versions of shapelib and pcre are
> you using?  If you disable dbfawk at configure time (--without-dbfawk) does
> the behavior stay the same?  Has it done this for very long, or did you
> just notice it?  If you cvs update to today's version does it still bloat?