Support for nonlinear geometries

[snorfalorpagus]

GDAL 2.0 introduced support for nonlinear geometries, such as CIRCULARSTRING and CURVEPOLYGON. There are also some functions converting to/from their linear approximations.

https://trac.osgeo.org/gdal/wiki/rfc49_curve_geometries

I don't think it would be too difficult to add support for this in Fiona.

[kannes]

I am currently mentoring a project where we would love to use Fiona instead of the official GDAL Python bindings. Our data has lots of curve geometries though so this is a blocker for us.

Is it worth trying to simply uncomment the types in https://github.com/Toblerity/Fiona/blob/master/fiona/_geometry.pyx#L18 and hope for the best? Or are there other places in the code where will is going to fail spectacularly?

[sgillies]

@kannes That is one change that we will need. A big missing piece is that OGR_G_GetLinearGeometry will need to be called before https://github.com/Toblerity/Fiona/blob/master/fiona/_geometry.pyx#L163 in GeomBuilder.build to coerce curve geometries to their linear approximations. We can't represent curve geometries as GeoJSON.

https://trac.osgeo.org/gdal/wiki/rfc49_curve_geometries

[snorfalorpagus]

I think there is a debate to be had re: returning the curved geometries vs their linear approximation. From my understanding curved geometries are still just a type and some coordinate sequences. The objects returned by Fiona are GeoJSON-like structures only. I would expect we could return structures for these, otherwise round trip operations are not possible. I do think we should make the linear approximations available thought.

[kannes]

And I just realised that of course we use Shapely for actual geometric stuff. D'oh!

Still, looking forward to the day when Fiona supports this :)

[jwilson8767]

Currently this is breaking things for me as prior to gdal 2.2 curve geometries were automatically approximated, post-2.2 fiona just throws a "UnsupportedGeometryTypeError". While I agree that preserving curves would be awesome for round-trip operations, in the short term I suggest we implement the suggested GetLinearGeometry to restore the previous behavior. Perhaps we could also check for non-linear geometries and show a warning that the geometry will be approximated.

[snorfalorpagus]

I made a start on full support for curved geometries not long after my last comment. It was going quite well until I realised we need to decide on an exact format for some of the more complicated geometry types. I wasn't aware that Fiona was already interpolating geometries with earlier versions of GDAL.

[jwilson8767]

Should we maybe create a separate issue for fixing the silent approximation vs actually full supporting curve geometries, especially as it sounds like full support may take more work than originally expected?

[sgillies]

Yes, two issues makes the most sense. Let's let this one continue as the issue about full support, including JSON representations of non-linear geometries.

[jwilson8767]

Was just looking at RFC 49: Curve geometries and realized that GEOS (and thereby shapely) is specifically mentioned as not supporting non-linear geometries at this time. A quick glance over the GEOS wiki makes it look like curve geometries are not even planned, but I could be wrong. -- Anyone know how to support getting non-linear geometries implemented in GEOS?

[sgillies]

Fix for #617 is pretty much done in #643.

[sgillies]

I'm going to close this in favor of #1345.