[storm] Oracle back end now passing all tests in tests.store and tests.database

[Drew Smathers]

On Wed, Jul 1, 2009 at 10:37 AM, Jason Baker<jbaker at zeomega.com> wrote:
> As of rev 339, the Oracle backend we've been working on is passing all
> the tests in both tests.store and tests.databases.  I've been working
> in lp:~zeomega/storm/zeomega_storm, which is mainly a somewhat heavily
> modified branch of lp:~kov-alfaiati/storm/oracle-support with most of
> Drew Smather's patches (lp:~djfroofy/storm/oracle-support-patches)
> added.  It's definitely far from being in a perfect state (as you can
> see from the bugs that are open), but it's getting better.  :-)
>
> That said, here are some issues that may pop up if you're switching
> from the current Oracle branch:
>
>  * Binary data is no longer encoded as base64.  Instead, the data will
> be hexlified, as Oracle expects binary data.
>  * The back end now requires cx_Oracle >= 5.  This is because
> cx_Oracle 5 makes type handling much easier by way of input type
> handlers and output type handlers.
>  * For ORA-08177 (a DatabaseError), the back end will now raise an
> OperationalError with the message 'database is locked' as storm
> expects.
>  * The back end now keeps its own store of reserved words and escapes
> them in mostly the same manner as storm itself does.  I did this
> because the reserved word list in storm is a bit dated (from SQL 92)
> and contains 159 words that aren't reserved in Oracle 10g.  For
> reference, Oracle 10g has about 66 reserved words.
>
> Some other things I've added:
>
>  * Support for connecting using a TNS name.
>  * "Isolation contexts" - essentially a way to temporarily put the
> database in read committed mode temporarily using a with statement.  I
> did this because I couldn't get any of the DDL statements in the tests
> to run without using read committed isolation.  I'm not sure this is
> the best solution, so I'm open to suggestions.
>  * Support for NLS_LANG settings other than UTF-8.
>

This is awesome.  (Except having to change my configs due to the
subtle differences in our approach to supporting TNS names :(
However, overloading the host seems reasonable and maybe cleaner when
working with a URI than adding "dsn" option as I did in my branch.)

> I did also make some changes outside of Oracle support.  Fortunately,
> I didn't change too much of the actual code itself (and what I did
> change may even end up getting rolled back), but I did have to make
> some changes to some of the tests.  I was going to include those
> changes here, but when I started to type the list out, I decided it's
> better to save them for a separate email.  :-)
>
> At any rate, this branch is *far* from perfect.  What it really needs
> now is more real-world testing.  We're going to start using it for one
> of our own projects, but if anyone else is using Oracle, it would be
> helpful if you could test it out.  If there are any questions, let me
> know.
>

I should have this tested more within the next month or so.  Still
using my branch (twisted-oracle) and we're a code freeze right now.

-Drew