Joris Van den Bossche

Hmm, that's true. I did it that way because it's what we already do for the left index / right column name case. On master: ``` In [1]: from shapely.geometry...

On the other hand, it's also strange to have a result with a suffix applied to the right label, but not to the left. While in practice you would expect...

Testing what it would give to always add suffixes to the index, also if it is set back as the index of the result, I ran into some more corner...

OK, I revived this PR and I ended up doing a full rewrite of the index-handling logic. It is still quite some custom logic, but I think it should more...

So in summary, the main "bug fixes" / changes in behaviour are: - Preserve the index name of the right dataframe, it it had one (so not uses "index_right" in...

> Just a note on having the custom indexing code, it seems contrary to the consensus in #498 where there is an arugment to reduce complexity in sjoin by making...

(note, @m-richards, I started to write this before your post appeared, so posting this comment anyway, will look at your comment afterwards) > Decide what to do with results that...

> 2\. .... The change here would be that a return result with any columns of GeometryDtype would result in the return type being a GeoDataFrame (as opposed to currently,...

https://github.com/geopandas/geopandas/pull/2268 only enforces that you can't manually set the crs on a GeoDataFrame with no active geometry or create such a GeoDataFrame, but you can still end up with such...

> Supporting a left join > ... when doing a left join, geopandas.sjoin preserves the rows of the left dataframe that don't match with a geometry of the right dataframe....