[postgis-devel] Topology Regression Failure

[Sandro Santilli]

Ok, I found the culprit.
The problem is wider than what shows up and may require reverting
commit r9072 or otherwise carefully going trough it.

Basically the issue is calling PG_FREE_IF_COPY too early.

When deserializing geometries we end up having LWGEOM objects
which still reference the underlying serialized representation.
PG_FREE_IF_COPY releases that underlying data.

This is still true on lwgeom_clone, which does not do a deep
cloning (also cloning pointarray). And lwgeom_clone is possibly
used by liblwgeom functions when they want to return a "copy"
of the input. This was the case for ST_Split, for example.

Freeing things _after_ serialization is fine.

I've committed a fix for ST_Split with r9095, which is fixing
the testsuite, but I'd be concerned about other such cases.

It would help if changeset were smaller and more focused.
The review of lwgeom_free and lwgeom_release didn't need
moving PG_FREE_IF_COPY around.

--strk;

On Wed, Feb 08, 2012 at 09:46:25AM +0100, Sandro Santilli wrote:
> On Wed, Feb 08, 2012 at 09:21:18AM +0100, Sandro Santilli wrote:
> > On Wed, Feb 08, 2012 at 09:05:23AM +0100, Sandro Santilli wrote:
> > > On Tue, Feb 07, 2012 at 04:32:40PM -0800, Paul Ramsey wrote:
> > > > Sandro,
> > > > Something in my cleanup to lwgeom_release
> > > > (http://trac.osgeo.org/postgis/changeset/9072/trunk) is causing
> > > > problems in topology...
> > > > 
> > > > 
> > > > --- regress/st_createtopogeo_expected   2011-11-13 13:55:16.000000000 -0800
> > > > +++ /var/folders/vQ/vQpcmrOZEXuINbyg+7Jj4k+++TI/-Tmp-//test_17_out
> > > >  2012-02-07 16:29:46.000000000 -0800
> > > > @@ -5,16 +5,16 @@
> > > >  ERROR:  Geometry SRID (0) does not match topology SRID (4326)
> > > >  T1|POINT(0 0)
> > > >  T1|1 nodes|0 edges|0 faces
> > > > -T2|SRID=4326;LINESTRING(0 0,8 -40)
> > > > -T2|2 nodes|1 edges|0 faces
> > > > +ERROR:  Invalid edge (no two distinct vertices exist)
> > > > +T2|0 nodes|0 edges|0 faces
> > > >  T3|SRID=4326;POLYGON((0 0,8 -40,70 34,0 0))
> > > >  T3|1 nodes|1 edges|1 faces
> > > > 
> > > > The problem seems to be in the topology creation, but it doesn't show
> > > > up directly and I don't know how to trace through and find out what
> > > > *particular* function has changed its behavior. If you can tell me
> > > > what function has changed, I can fix it, but at the moment I'm SOL.
> > > 
> > > Did you check cunit test under valgrind ?
> > 
> > I just did and it shows a lot of errors in TIN and SURFACE tests,
> > but nothing about topology. Will inspect further. May be better
> > NOT to cut next alpha until the regression and memory errors are
> > fixed.
> 
> First error I detect with POSTGIS_TOPOLOGY_DEBUG defined is here:
> 
> Good:
>  DEBUG:  Split by POINT(36 26): MULTILINESTRING((36 26,38 30))
>  DEBUG:  Split by POINT(38 30): MULTILINESTRING((36 26,38 30))
> 
> Bad:
>  DEBUG:  Split by POINT(36 26): MULTILINESTRING((36 26,38 30))
>  DEBUG:  Split by POINT(38 30): MULTILINESTRING((4.94065645841247e-324 6.94094528790257e-310,6.94094528790099e-310 38))
> 
> But doesn't look like the problem is in ST_Split, so there must be
> some memory corruption before that.
> 
> --strk;
> 
>   ,------o-. 
>   |   __/  |    Thank you for PostGIS-2.0 Topology !
>   |  / 2.0 |    http://www.pledgebank.com/postgistopology
>   `-o------'
> 

-- 

  ,------o-. 
  |   __/  |    Thank you for PostGIS-2.0 Topology !
  |  / 2.0 |    http://www.pledgebank.com/postgistopology
  `-o------'