ENH: support mask keyword for compatibility with Fiona

[brendan-ward]

Current plan is to use bounding box of mask when reading features with GDAL, then filter using shapely.

The issue is that OGR will give different results if GEOS is or is not compiled into GDAL. Using shapely afterward ensures precise results.

[jorisvandenbossche]

Can we detect when the GDAL in use in compiled with GEOS or not? (I suppose if get_gdal_geos_version() is not None?) Because otherwise we could let it depend on that whether we do an additional check with shapely or not?

Other question is also whether we want to do this in pyogrio, or just in geopandas.read_file (where we could pass the bbox to pyogrio, and do the remaining filtering there).

[martinfleis]

Other question is also whether we want to do this in pyogrio, or just in geopandas.read_file (where we could pass the bbox to pyogrio, and do the remaining filtering there).

The more code that can live in pyogrio, the better I suppose. Then you know where to look when the issue comes up.

[theroggy]

I suppose it is, otherwise it probably wouldn't have come up, but is it a common case that gdal is built without geos? If not, just raising an "unsupported" exception or log a warning if geos is not available might be enough?

I had a look at the docker images, and if read correctly it is included in >= ubuntu-small and >= alpine-normal, so only not in alpine-small. Alpine-small is - I suppose by purpose - really small, so has quite limited support for anything, so not sure if that is the bar to aim for.