[gdal-dev] ogr - layer extent weirdness after geometry update

[Even Rouault]

Selon Simon Lyngby Kokkendorff <silyko at gmail.com>:

Yes indeed, you've precisely described the behaviour of the code of the OGR
shapefile driver. The shapefile format stores the extent of the shapes stored in
the shapefile.
When you append a new shape in a shapefile, the shapefile driver will extend the
layer extent with the shape bounding box. This works well and efficiently when
creating a shapefile from scratch.
The same algorithm is also used when updating an existing shapefile, and it's in
that case you can get the weird behaviour you see if you move a shape that was
touching the layer extent. In that case the shapefile driver should invalidate
the layer extent and recompute it entirely by iterating over all the shapes,
which could be costly if the shapefile is big. But the current situation is not
very satisfactory. And a an optimization, the invalidation could be only done if
the bounding box of the old shape touches the layer extent.
The deletion of a feature could also lead to the same behaviour.

Could you file a ticket about this in GDAL trac ?

> Hi List,
>
>   I have observed a puzzling behaviour of the reported extent of an ogr
> layer after geometries of features have been modified, with the file opened
> in update mode. I use gdal 1.7 via the pyhton bindings, open a dataset with
> ogr.Open(fname,update=True), then loop through layers and features and
> modify geometries like:
> geom=feature.GetGeometryRef(), new_geom=some_modification(geom),
> feature.SetGeometry(new_geom), layer.SetFeature(feature). However, ogr (and
> Qgis, which uses ogr for that I suppose) reports the new layer extent as,
> the "minimal bounding box" of the new and old extents. So somehow, the
> extent before geometry modification is still stored somewhere in the
> dataset. The same observations also hold for a dataset which is modified by
> hand using the editing tools in Qgis! Consider the following ogrinfo output
> from a file with one layer, containing one point:
> INFO: Open of `qgis_test.shp'
>       using driver `ESRI Shapefile' successful.
>
> Layer name: qgis_test
> Geometry: Point
> Feature Count: 1
> Extent: (0.077586, 0.577586) - (10.670690, 31.805172)
> Layer SRS WKT:
> GEOGCS["GCS_WGS_1984",
>     DATUM["WGS_1984",
>         SPHEROID["WGS_1984",6378137,298.257223563]],
>     PRIMEM["Greenwich",0],
>     UNIT["Degree",0.017453292519943295]]
> test: String (80.0)
> OGRFeature(qgis_test):0
>   test (String) = nothing
>   POINT (10.670689655172414 31.805172413793112)
>
> Even though the point clearly lies at (10.67,31.80) the previous position is
> still included in the extent. Is this a bug or is there some reasoning
> behind this behaviour?
>
> Best Regards,
> Simon. National Survey and Cadastre of Denmark
>