> Peter Eisentraut wrote:
>> If you replace the third point by "maybe partition TOAST tables", replace
>> large object handle by TOAST pointer, and create an API to work on TOAST
>> pointers, how are the two so much different? And why should they be?
The reason they should be different is that (IMHO anyway) you don't want
the default behavior of SELECT * FROM ... to include pulling back the
entire contents of the blob. Indeed, we *can't* have that be the
behavior, unless we want to go back to the proposal that started this
thread of making the entire system safe for multi-gigabyte datums.
It's certainly possible that the underlying implementation could be
just TOAST, but we need some other API at the SQL level.
regards, tom lane