So it’s not available in 11.1.0.7.

There are references to 7.2 and 8.1 in the note – so it has been around a long time; I think it must have been hidden for the last few years.

Thanks for that link it was very helpful – I’ll have to see if I can get a backport to 11.1.

I’ll see if I can check the SQL some time next week – the first thing that springs to mind with the repeated occurrence of 82 is that it is 1% of 8168, suggesting a filter predicate like “function(column_value) = constant”.

The optimizer assumes that the number of rows produced by the table() operator will be roughly the same as the number of bytes in the default block size

I do remember reading a note on the Metalink saying that the number is derived as the number of 100 bytes length rows in 100 blocks. Still can’t find this note in MOS, so this may not be true.

BTW, the VIEW operator in the step 44 has cardinality of 82 – and IIRC this number is also kind of “magic” in a way that after instance restart you may see it changed to something different (I was told this by co-worker; and don’t remember all the details, but I’m pretty sure there’s some oddity in this number too).

Good point – I’ll have to check to see why it didn’t. Whether it does or not, though, there’s an important principle that the optimizer seems to be doing something very strange in this particular code path. The reported paths are NL and HJ, but the arithmetic looks like Cartesian Merge Join.

Thanks for the Tom Kyte reference

At the moment the best I can say is that it existed once, and appeared to be a deliberate note to explain the hint and its use.

Right now the best I can do (apart from the reference to Tom’s comments supplied by Narenda) is a couple of bug reports where a user has called in a bug relating the the cardinality() hint and not been told: “unsupported feature”. For example, bugs: 7758650, 9024541, 4137713 (which is a multi table example), and 9499541 (where a suggested workaround is to use the cardinality hint).

I’m at the UKOUG conference right now with some of the heavy hitters from the Optimizer team, so I’ll see what they say about it.