Hi @ankrgyl, we use the same function to auto-detect the schema in our JSON reader. DuckDB is case-insensitive, and duplicate column names are not allowed in tables/structs, so this behavior is the result of a fix (#11271).

The json_structure function is just a function that exposes this automatic type detection, so its behavior also changed.

There are two options, we can either:

1. Throw the error (like we do now)
2. Ignore the error and return {"a":"UBIGINT","A":"UBIGINT"}, but you will not be able to use this as an input to json_transform (like before)

I'd happily receive your input on what you would prefer here.

Thanks for the clarification. In our case, we actually use json_group_structure without using json_transform. Specifically, we:

• Call json_group_structure to find the distinct set of fields
• Populate a UI dropdown based on those fields
• If the user selects one of the subfields, run a query using json_extract (which is case sensitive) to project / group by it

So my preference would be option 2. One suggestion could be that json_transform could error if you try to use it with something like {"a":"UBIGINT","A":"UBIGINT"}, or you could have a flag to json_group_structure which allows you to pick your adventure.

Thanks, I've sent out a PR that fixes this for json_structure specifically, without changing the behavior of our JSON reader.

Appreciate it!