Andrew Lamb

I will fix this

I agree the wrapping is confusing and I think in general the wrapping of errors is not valuable. We see similar wrapping obfuscation in IOx as well. The datafusion code...

> And we hope to do error recovering for some special error case and we need to know the root cause of the error. If an error is wrapped too...

> But it still depends on the programer to propagate the errors correctly in the execution path. And there is another case that we still could not catch the error...

> See https://github.com/rust-lang/rust/issues/58520 for work to stabilize a mechanism to "walk" the tree Yes, this is what I really want out of errors in DataFusion and Arrow

Filed https://github.com/apache/arrow-rs/issues/2725 to discuss better error handling in arrow (and thus also datafusion and ballista)

🤔 maybe we need to use `Decimal` type rather than `Float` for constants. When we initially ran TPCH we did not have support for this data type, but now thanks...

> This is a little weird. @alamb should we follow the behavior of Spark or Postgresql? In datafusion I think we should follow the postgresql model unless there is some...

> Also, how do you deal with the synchronization between the projects? for example, some DF issues are due to the decimal cast code in arrow-rs .. how do we...

That would be awesome @kmitchener -- and a long time coming 👏