[postgis-devel] Much slower processing on GEOS 3.9.0 versus 3.8.0 for geodesic area calculation - Upgrade to PROJ 7.2.1 causes issue

[Marco Boeringa]

"The cache can hold up to 128 transformations, which should be enough 
for most cases"

With 60 UTM zones, that works out nicely even if someone wants to loop 
through all of them and do back-forth transformations ;-)

"I'm not against backporting it, but I'm not sure the gain justifies the 
risk and effort."

If not backporting, the fix will be in a 3.1.2 release?

Marco

Op 15-5-2021 om 03:38 schreef Raúl Marín:
> On 15/5/21 1:58, Paul Ramsey wrote:
>
> > The fix is a little invasive, I'm "so so" on backporting it into the 
> stable branches. Interested in comments regarding that.
>
> Adding fcinfo everywhere was a pain, so it's only fair that removing 
> it would be too. I'm not against backporting it, but I'm not sure the 
> gain justifies the risk and effort.
>
> I like those tricks to get Postgis schema without the fcinfo. If I 
> recall correctly somebody (mobilitydb?) had some issues related to it 
> because they were calling Postgis from C without passing the fcinfo. 
> This change should avoid that, I think.
>
> > It works cleanly for me locally and evades the start-up cost of 
> proj. I haven't done anything to make the cache access any faster yet, 
> but it's a low priority since most people only have a handful of 
> projections in their cache anyways.
>
> The cache can hold up to 128 transformations, which should be enough 
> for most cases, and has some basic intelligence to remove the least 
> used objects when it's full but it might need to be improved if being 
> full ends up being the norm. It's really simple so I don't expect any 
> performance issue coming from that unless the cache size is increased.
>
>
> Regards,
> Raúl.
>
> _______________________________________________
> postgis-devel mailing list
> postgis-devel at lists.osgeo.org
> https://lists.osgeo.org/mailman/listinfo/postgis-devel