[geos-devel] JTS/GEOS performance

Chris G. Nicholas cgn at globexplorer.com
Mon Jan 31 19:42:44 EST 2005


>  an enterprise api should have both smart pointers and ref-counting available as an option.  

Martin is certainly correct that smart pointers, reference counting and garbage collection are all related.  But one has to decide if a geometry API and in-memory features are fundamentally multithreaded animals, what their persistence mechanisms are, and what happens when one thread does wierd things while others are interested in a particular feature.  One can quickly open the whole "long transaction" can of worms and semantic questions like "when does an oil slick become more than one feature" ;-) .

I would argue that reasonable assumptions, i.e. simple thread safety and *s*p*e*e*d*, are much more preferrable than exotic special cases. Straightforward readers/writer locks should suffice; (not sure what the pthread equivalent is, but on Solaris it is rwlock(3thr) ) . 


Chris



More information about the geos-devel mailing list