Oliver Jowett <oli...@opencloud.com> writes:

Again here, I think the word "compilation" is purposely fuzzy to be portable across DBMS. The message is:

if the driver supports <pre-compilation|pre-parsing|pre-planning|pre-whatever> then the PreparedStatement will be _sent to the database_ at object creation time then not sent again but referred to.

By the way that's precisely the reason why the SQL request string of PreparedStatement is documented as not mutable, to allow this optimization.

PreparedStatement are quite obviously meant to match the SQL "PREPARE" statement or the ODBC SQLPrepare() function
<http://msdn.microsoft.com/library/default.asp?url=/library/en-us/odbc/htm/odbcprepared_execution.asp> The name re-use is not just coincidence, don't you think?

OK, this discussion has a bit slipped to a general description of the implementation of PreparedStatement, but the main questions stayed explicit, simple and not so far away:

1. is it possible to get the same level of performance achieved by LIMIT using .setMaxRows() (question still in Subject:)

2. when is the SQL query actually sent to the database and "prepared"? This question came because it seems any .setMaxRows() coming AFTER this point will come too late to help the server, whatever the protocol changes.

Then I digressed to other PreparedStatement performance questions. You did not had to answer; thanks a lot for doing it.

It's functionally irrelevant, but this is a discussion about performance. From what I got, calling .setMaxRows() before vs after query parsing greatly changes the range of possible optimizations. This discussion is just about the ordering of events (= "hand-waving"?). I'm not even discussing about any protocol changes at this point.

Even if you delay the parsing on the first execution in order to wait for any helpful but late .setMaxRows(), then later .setMaxRows() on the next re-executions will come after parsing anyway. Unless you kinda re-parse every time...

I found very interesting to discuss about possible strategies to handle this issue, but you do not have to be interested. If no one else is interested then I'll stop noising the list.

I can understand that. Looks like we already saw that with the datatypes issue.