Putting an end to this awfull unknown type problem (autocast doesn’t work all the time - “null pointer error”, “error during deserialization” and others, perhaps the most unstable node of all the nodes => avoid when automation), string conversion neither, nor expression. The fact is “unknown type” should be a string and after that you do a choosen type conversion.
The brute force way would certainly be one approach for an issue lingering around so long.
BUT, and this is literally a big BUT, when mergin two different data type, you want something that tells you that you made a mistake. Just converting silently to string could cause quite a some repercussions.
Then there are situations where the data ingested cannot be assigned a data type:
Personally, I’d love to see an improvement on converting non-native more easily as the string manipulation nodes cannot cope with non-native and so do many other nodes too.
Anyhow, you still got my vote but the brute force approach requires some through thought as once implemented, rolling back becomes challenging.
I do agree, but throwing an error will be better than allowing the creation of an unknown type column. The string option could exists anyway…as an option. Doing the node auto-cast more reliable could be an option too, because it’s behaviour seems to be heratic in general and a kind of unpredectible in automation and we dont even speak using it in a loop structure.