[postgis-devel] Why delegating MakeValid to GEOS ?

[Sandro Santilli]

On Thu, Dec 10, 2020 at 10:19:30PM +0100, Sandro Santilli wrote:
> On Thu, Dec 10, 2020 at 12:48:08PM -0800, Paul Ramsey wrote:

> > If you don't then we instead end up at a place where swapping implementations results in sudden behaviour change, which is Also Bad.
> 
> True. I'll check how does the GEOS side behaves, chances are
> it already does NaN stripping (it doens't help that the _unit_
> test for GEOS makevalid was left out of autotools build scripts).

I checked: GEOS won't even accept NaN ordinate values, so that part
of the work will still need to stay in PostGIS.

The only difference right now between going with GEOS or doing it
fully inside PostGIS seems to be the stripping of EMPTY components,
without which GEOS may crash.

--strk;