Matt Richards

Hi @aleczoeller, thanks for the pull request on this. Please have a look at the comment I've just left in #2951 - I don't think this approach is the best...

I'm going to close this in favour of #3599, I can see however this was attempting to solve do more than that. If a minimal reproducible example were added to...

This unfortunately isn't new as such, it's just that the `GeoDataFrame(..)` call hid this for the most common case - where we had a geometry column called "geometry" due to...

Aside, I was trying to think about how to do this properly, and thought that a Categorical is the best analogue in pandas: ```python import pandas as pd df =...

Just adding as a note that @jorisvandenbossche picked up on the warnings PR, this case also regresses: ```python crs = "EPSG:4326" df = GeoDataFrame( { "geometry": [Point(0, 0), Point(1, 1),...

There's actually a much older regression at play here too. ```python import geopandas from geodatasets import get_path nybb = geopandas.read_file(get_path("nybb"))[["BoroCode", "geometry"]] crs_orig = nybb.crs # add a new row nybb.loc[5]...

> It is not possible to roundtrip nullable dtype via GPKG as GeoPackage file format does not support it. The only way to write a numeric column with missing values...

I just had a quick look at #3088, @martinfleis I think you mean that we should do this for both GeoSeries.crs and GeoDataFrame.crs, correct? (that PR is currently just GeoSeries)

Going to close this as there is nothing substantial included in the PR. Future contributions on this are welcome as comments on #2856 or #2818 though.

We could potentially raise a warning for this, but I do not think we should attempt to fix the character or throw an error, as QGIS for instance will read...