Add write_csv to DataFrame

[matthewmturner]

Which issue does this PR close?

Closes#1777 task 7

Rationale for this change

What changes are included in this PR?

Are there any user-facing changes?

[matthewmturner]

@alamb would you mind checking this out to see if going in right direction? I took the logic for writing csvs and put it in new function plan_to_csv within src/physical_plan/file_format/csv and moved testing there. Then just used that function in both ExecutionContext and DataFrame

currently failing because of the signature of write_csv having impl AsRef<str> which i took from the existing implementation on ExecutionContext but doesnt work because trait object cant have method with generic type parameter. Is there a reason we cant use &str? More details here https://doc.rust-lang.org/error-index.html#E0038 and i dont think any of the proposed workarounds are good for our use case.

[alamb]

I took the logic for writing csvs and put it in new function plan_to_csv within src/physical_plan/file_format/csv and moved testing there. Then just used that function in both ExecutionContext and DataFrame

Looks good to me 👍

currently failing because of the signature of write_csv having impl AsRef which i took from the existing implementation on ExecutionContext but doesnt work because trait object cant have method with generic type parameter. Is there a reason we cant use &str?

I think taking &str is totally fine in the DataFrame trait

[matthewmturner]

@alamb thanks!

[jimexist]

Hi thanks for the contribution

I wonder if we can use an extension trait for Csv writing method so that users can choose to use the method if they import the trait - same goes to other potential methods like write json files, etc.

This way we can reduce the minimal dataframe struct size

[matthewmturner]

@jimexist im not opposed, but would the idea be to do this only for methods that are writing? or should this be generalized to all IO methods? Im just wondering where we draw the line between core dataframe methods and extensions.

[matthewmturner]

@jimexist what would you think about merging this as is and then tackling that point as part of convos on #1712 where there is interest in moving DataFrame from trait to struct.