RE: ECPG bug fix: DECALRE STATEMENT and DEALLOCATE, DESCRIBE

Dear Horiguchi-san,

Thank you for reviewing! I attached new version.
Sorry for delaying reply.

> Since we don't allow descriptors with the same name even if they are
> for the different connections, I think we can set the current
> connection if any (which is set either by AT option or statement-bound
> one) to i->connection silently if i->connection is NULL in
> lookup_descriptor(). What do you think about this?

I tried to implement. Is it correct?

> connection is "conn1" at the error time. The parser relies on
> output_statement and friends for connection name reset. So the rules
> that don't call the functions need to reset it by themselves.

Oh, I didn't notice that. Fixed.
I'm wondering why a output function is not implemented, like output_describe_statement(),
but anyway I put a connection reset in ecpg.addons.

> Similary, the following sequence doesn't yield an error, which is
> expected.
>
> > EXEC SQL AT conn1 DECLARE stmt STATEMENT;
> > EXEC SQL AT conn2 EXECUTE stmt INTO ..;
>
> In this case "conn2" set by the AT option is silently overwritten with
> "conn1" by check_declared_list(). I think we should reject AT option
> (with a different connection) in that case.

Actually this comes from Oracle's specification. Pro*C precompiler
overwrite their connection in the situation, hence I followed that.
But I agree this might be confused and I added the warning report.
How do you think? Is it still strange?

Best Regards,
Hayato Kuroda
FUJITSU LIMITED