[postgis-devel] Minimum number of points in a POLYGON ring?

[David Fuhry]

Chris Hodgson wrote:
> What if the two modes of operation were "reject invalid geometries" and 
> "automatically fix invalid geometries (in ways which are clearly 
> documented)"? This way we get to assume a standard level of data quality 
> that is slightly above "unknown", and (if we set rejection as the 
> default) we make sure that the user knows that we are fixing their data 
> for them. Also, this seems like it should introduce less of a second 
> path through the code, only a single "if it's broke, fix it" should be 
> required.

The problem is that the definition of "fix" is application-specific.  I 
may want to try to repair geometries with certain characteristics, and 
just NULL out others.  There is no general formula for what constitutes 
a fix in all cases.

> Or perhaps the ideal system has all three options, including these two 
> and the "we'll take whatever you got" option. But then we don't get the 
> nice assumption of a minimum quality level.

Currently, I think only an isvalid() check constraint gives you any 
level of quality assurance at all.

Thanks,

Dave