Thanks for linking to the OTN post – that was an excellent example of how to present a problem.

I’ve had a look at the problem, and written a couple of notes – I may write a little more on my blog over the weekend. I think an interesting point here is that the “unparsed” SQL can show the join order and join conditions, but (in line with my prevous comment about semi-joins) can’t indicate all the limitations that apply as you go from table to table. In this case I think there was a hidden side-effect of the order of the FROM clause that made the default rewrite mechanism block (or introduce an error for) a particular index.

The problem I was referring to was with an ANSI-style outer join — there was a thread about it with a reproducible test case https://forums.oracle.com/forums/thread.jspa?threadID=2394716

It was giving a bad plan, and rewriting it in the traditional style solved that problem. However, I was surprised to see in 10053 trace file for the ANSI-style query that the “unparsed query” was nearly exactly the same as the rewritten one. So at least in that particular case “unparsed query” was very misleading.

I don’t have a definitive answer on the unparsed query. I know that it is not (always) the whole truth because, for example, you can have an unparsed query that shows something that appears to be a join when the actual plan will have to use a semi-join. I just take it a possible clue about how the optimizer is working and where it’s heading, but I won’t assume that it’s an infallible guide line (at least for us outsiders) to the optimizer’s final destination.

BTW what is the significance of “unparsed query text”? Once I was tracking down a problem due to an ANSI outer join, and took this “unparsed query text” and ran an explain plan on it — it gave me a different plan from what the optimizer eventually arrived in the 10053 trace file. There weren’t any binds in the query, so apparently after it gets to this “unparsed query text”, the optimizer does more things to it, which we don’t know (and can never find out about). Is this correct or am I missing something?… And if this is correct, then is there any use in “unparsed query text” at all?

Best regards,
Nikolay

Thanks !