Issue found on page 'Order Preservation'

[soerenwolfers]

It's quite unclear to me what

For many operations, DuckDB preserves the insertion order of rows, similarly to data frame libraries such as Pandas. The following operations and components respect insertion order: The CSV reader

means: A user could be forgiven to conclude from this sentence, e.g., that aggregate functions are fed inputs in the "original table order" (see duckdb/duckdb#14755), that sorting is stable (duckdb/duckdb#14941), or generally that arbitrarily complex and nested queries on data respects that original order (at least for the CSV reader), none of which I don't believe is the case.

I also wonder why the paragraph starts with "For many operations preserve..." and then ends with "The following operations and components respect ..: <EXACTLY ONE COMPONENT AND NOT A SINGLE OPERATION> The CSV reader".

Page URL: https://duckdb.org/docs/sql/dialect/order_preservation.html

[soerenwolfers]

@szarnyasg I think this is one of the biggest causes for confusion about duckdb, so a few words about SQLs general approach and duckdb's actual specific enhancements would be super useful. I can't help out in this case because I genuinely don't know what duckdb's philosophy here is not what the current specifics look like.

[szarnyasg]

Hi @soerenwolfers, great point. A fix is coming. Still a lot more could be said about the topic but this quick patch should clarify the topic.

[soerenwolfers]

@szarnyasg This is great and offers way more powerful guarantees than I expected! I'll make a PR that adds explicit answers to the two questions/issues I linked above