Jorge>
The value of 0 should be used to fetch all rows, if we leave the default at 0 and handle internally that as 100 (and reporting it as 0) it is a wrong behavior.
Note that "unset" and "set to 0" might have different meanings for a good (or bad?) reason.
E.g. 1.7 times difference for fetchsize 1000 and 100 for fetching 2000 rows of 4 int4 over localhost connection.
Ideally, I would like the backend to support "fetch at most 10000 rows and at most 1MiB" kind of requests.
Unfortunately, PG does not yet support "byte size limited" fetches, so we might implement some "machine learning" trick: "fetch 100 rows, calculate average row size, then adjust fetch size so subsequent fetch would be close to the desired "fetch byte length"
Anyway, I think it makes sense to switch to some non-zero value for defaultRowFetchSize, then implement "fetch size autoscaling".
Vladimir
