[postgis-devel] CGAL/SFCGAL

[Paul Ramsey]

> On Feb 28, 2020, at 1:54 AM, rmrodriguez at carto.com wrote:
> 
> Hi,
> 
>> Has there been some work in trunk to support newer versions of CGAL/SFCGAL and in particular (I think) do something special with invalid geometries being sent into CGAL/SFCGAL? It seems like the fact that PGDG had to peg their versions back down at 4.7/1.3.1 is worrisome, maybe we have a general problem with newer versions?
> 
> AFAIK, we haven't changed anything to add support to newer minor releases.
> 
> We have some issues open (#4157, #4513) that crash under
> `SFCGAL::algorithm::isValid`; I opened a ticket in sfcgal
> (https://github.com/Oslandia/SFCGAL/issues/194) but it hasn't seen any
> movement. To be honest, I've reached a point where I've spent 10x more
> time dealing with sfcgal issues (build it, checking bugs, etc) than
> using it (which is next to 0) so I'm not too keen to get into it
> again.

Well, short of completely tossing it, the 3.0 separation (thanks, Komzpa) has at least made it very highly optional. 
I find it odd that our bindings have moved to new versions whilst those self-same  new versions seems susceptible to crashing against long-standing components of our regression suite. Maybe the crashes are quite platform dependent. I will leave my versions pegged to old ones for now, and perhaps do some experiements with the very freshest versions to see if there’s any value in modernity. 

P.