[geos-devel] Unclear objects lifetime and ownership issues in Monotone Chain components

[Martin Davis]

Mark Cave-Ayland wrote:
>
> Martin, would there ever be a case where a hidden internal object could
> be shared between multiple parent objects?
>   
Um...  I'll say no, this is not a pattern which is intentionally used in 
JTS *at the level at which user functionality is concerned*.  Of course, 
in the internal graph structures there is lots of cross-pointing, but 
the lifetime of these structures is always well-defined. 

In general, if the goal is to expose the methods on Geometry which are 
used in PostGIS, then it should be possible to free all allocated memory 
when the method completes (apart from any returned geometry, of 
course).  The one new exception to this is the PreparedGeometry classes, 
which of course have to hold onto memory for their internal structures.  
But their lifetime is well-defined as well.

Hopefully that helps.  Is there a possibility of supplying a single pool 
to each new Geometry method, and freeing it at the end of the 
operation?  Although how would you allow returned Geometrys to be 
allocated outside of this pool...  But maybe that doesn't matter - you 
can just free the pool at the end of the PostGIS function. 

This seems like it might be building GEOS to be a bit too specific to 
its use in PostGIS, however.  I think there's lots of people who want to 
use it in other environments.

Is GEOS really that far from being memory-tight?
>
> ATB,
>
> Mark.
>
>
> _______________________________________________
> geos-devel mailing list
> geos-devel at lists.osgeo.org
> http://lists.osgeo.org/mailman/listinfo/geos-devel
>
>   

-- 
Martin Davis
Senior Technical Architect
Refractions Research, Inc.
(250) 383-3022