Wenchen Fan
Wenchen Fan
why do we remove a working feature? what's wrong with `to_csv` generating non-standard but pretty strings for these values?
So we make a breaking change simply because we don't like the object address in the string? can we fix it?
You never know how people will use Spark and I'm not sure why it's necessary to ban this use case. Having inconsistency between scala and python is bad. Can we...
I'd prefer to fix the inconsistency. Can we use `ToPrettyString.eval` to generate pretty strings for these types in `to_csv`?
It's ok to not support round-trip. array/struct/map is not CSV standard anyway. I think the principle is to avoid breaking changes if possible. Since `to_csv` already supports array/struct/map, let's keep...
If no objections, I'll revert this tomorrow. We can have a new PR to add pretty string for array/struct/map in `to_csv`.
I agree that the current behavior is buggy, but we should have more discussion about the expected behavior. For `CREATE TABLE IF NOT EXISTS`, I think the user expectation is...
thanks, merging to master!
thanks, merging to master!
thanks, merging to master/~3.5~!