ENH: Add index_as_fid option to write_dataframe?

[theroggy]

There is an fid_as_index option in read_dataframe to be able to... read the fid. However, there is no explicit option to e.g. index_as_fid in write_dataframe to write it again.

For e.g. shapefiles writing fid's is not relevant at all as they don't save fid's.

For Geopackages there is a workaround: if you add the index as a column named "fid" to the dataframe you will write the fid.

Not sure if an explicit option is needed... an alternative could be to just document the "workaround" or use this issue as documentation :-).

[jorisvandenbossche]

Do you know if GDAL exposes some information or metadata about this? For example so that we can raise an error if the file format actually does not support it?

[theroggy]

Do you know if GDAL exposes some information or metadata about this? For example so that we can raise an error if the file format actually does not support it?

Yes, if OGR_L_GetFIDColumn() returns "", the file type cannot save fid's. If it returns another string, that's the column name used to store the fid.

Examples:

• for shapefile "" is returned.
• for Geopackage "fid" is returned: that's the default name of the column the fid is stored in.
• for sqlite files I never tested/used it, but according to the gdal doc the default column name for the fid is "OGC_FID", so I suppose that OGR_L_GetFIDColumn() will return that and I suppose that you'll have to call the column like that in the GeoDataFrame if you want the fid's to be retained.

[jorisvandenbossche]

Nice, in that case this sounds as a good option to me.

[theroggy]

I just encountered another way to save the index of the GeoDataFrame as "fid" (primary key) in a GPKG: if the name of the index is "fid" it will also be saved as "fid" in the file.

PS: Fiona treats this the same way.

xref: geopandas/geopandas#3217

[jorisvandenbossche]

(moved this to the next milestone)