Isn't this the case for postgresql data in general? ie, on updates, new data rows are written containing all the field values, whether or not they have changed.
It's more they case that when you want to reach into an existing stored document and change a single value you have to re-write the entire document from your application.
On the reading side there are functions and operators that allow you to reach into stored JSON and extract the parts you want. What would be nice would be to be able to do something similar for updates - although this is clearly more complex than reading, so I can see why it has been done this way.
Edit: I guess the most general solution would be to directly support something like JSON Patch:
Yes, but you don't have to specify all the other columns that aren't being updated.
With json/jsonb, you have to provide the entire object graph each time you are updating it. You can't update one field in the graph. Which could be a pain if you have concurrent updates.
Hopefully we'll be able to update parts of the jsonb columns sometime.
You're right, because MVCC works on immutable rows. The structure can be broken up to accommodate large values[1], which saves updating parts that don't change, but json data is a single column and updates need to copy the entire document.
