Even Rouault

> Actually SetActiveSRS() could be seen as a particular case of a more general SetLayerDefn(). actually, no... SetActiveSRS() is about on-the-fly "reprojection". Whereas if we would allow SetLayerDefn() to modify...

> I think in an ideal world it would be possible to override the type of one field, and have the rest auto-detected. Brainstorming: if we go for the open...

> it looks like json-c implements [RFC 6902 JSON Patch](https://datatracker.ietf.org/doc/html/rfc6902#section-4.3): https://json-c.github.io/json-c/json-c-0.18/doc/html/json__patch_8h.html. In theory, this would allow fine-grained patching, like 'replace the `type` of field 1 with "Integer"'. I'm not sure...

> So how can user know that GDAL makes a bad guess? That might be a "Warning 1: Integer overflow occurred when trying to set 32bit field." (cf https://github.com/OSGeo/gdal/issues/10938), which...

> I meant that somehow just the field definitions would be saved into the schema file I was thinking you meant the user would just edit the output of "ogrinfo...

Draft RFC103 about that topic: https://github.com/OSGeo/gdal/pull/11071

For good measure CC @ptitjano as a [QDuckDB](https://gitlab.com/Oslandia/qgis/qduckdb) contributor (I unfortunately don't know the github handle of Florent Fougères)

> I was thinking of a much more lightweight integration, which is just supporting DuckDB with a vector driver. Obviously! I jumped right away to the stretch goal of having...

> The downside of this is of course that you may end up needing a pretty beefy machine to actually compile it. I would expect most packagers to actively hate...

https://github.com/OSGeo/gdal/pull/11003 adds a ADBC driver that can use libduckdb to read Parquet or DuckDB datasets. Before you get too excited, no spatial support for now :-) but should be doable...