eliaperantoni opened a new issue, #14429: URL: https://github.com/apache/datafusion/issues/14429
### Is your feature request related to a problem or challenge? In #13664 we introduced the `Diagnostic` type and `DataFusionError::Diagnostic`. They allow enriching errors with messages meant for consumption by end users of an application built on top of DataFusion, by providing rich information and context that directly references locations in the SQL query. They enable features like:  See `datafusion/sql/tests/cases/diagnostic` for examples on how to extract and use diagnostics: https://github.com/apache/datafusion/blob/d5428b21d6486e5b7db525d314ee9119a512c397/datafusion/sql/tests/cases/diagnostic.rs#L132-L140 In that PR, we only implemented diagnostics for: - Unresolved table references - Unresolved column references (qualified and non) - Non-aggregate expressions missing from `GROUP BY` clause - Ambiguous column references - Wrong number of columns in set expression (e.g. `UNION`) - Incompatible types in binary expressions This issue is about using `Diagnostic` in more places, and adding related tests to `datafusion/sql/tests/cases/diagnostic`. We think we should at least implement the following, but suggestions are welcome and encouraged: TODO ### Describe the solution you'd like The implementation should follow the steps of #13664, by calling `DataFusionError.with_diagnostic` to attach a `Diagnostic` to an error that is currently being returned. Tests should be added to `datafusion/sql/tests/cases/diagnostic` for each newly supported scenario. For some of these items, it might be necessary to enrich the logical types with the `Span` information coming from the parser. This should be done using the `datafusion::common::Spans` type (note the "s"), introduced in #13664 to add span information to `datafusion::common::Column`. It is desirable that the implementation is as little invasive as possible, in that it shouldn't require changing tons of function calls and types, unless absolutely necessary. The public facing API shouldn't change. The `Diagnostic` should be attached as soon as possible to the creation of the wrapped `DataFusionError` (i.e. deep in the call stack) and every error should ideally have just one `Diagnostic`. ### Describe alternatives you've considered _No response_ ### Additional context _No response_ -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
