[Webware-discuss] MiddleKit and Postgres

[Graham H. <gr...@ly...>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

I've been looking through MiddleKit in the 0.5.1rc2 release
(incidentally, very easy to follow code; thank you!), and have a list
of things that would need to be changed to have a PostgreSQL backend,
something I would need to be able to use MK.  I'm willing to implement
this stuff, incidentally.

o You use /* a lot.  /*, despite being recognized by Postgres and
  MySQL both, isn't really proper SQL syntax: real SQL comments match
  the regexp `--.*$'.  I'm not going to change this, as Pg supports
  the C style block comments.

o Postgres has different syntax for drop database (no if exists
  clause, mostly) and changing to a new database, as well as IDs and a
  couple other minor things.  Not a big deal.

o You use enum (...), which isn't in ANSI; I assume this is a MySQL
  thing, as the normal way to do it in ANSI is `varchar (<max enum
  length>) check (<column_name> in (...))' which requires some sort of
  referential integrity checking.

o You use `bigint unsigned /* Video */' for foreign key referencing;
  ANSI doesn't specify, but assuming your IDs fit in an int4, `int4
  references Video (videoId)' will do the checking correctly.  If they
  don't, then I suppose an int8 would do the trick, but that's a lot
  of rows.

o Generate.py would of course need to be hacked slightly, but only
  slightly.

o MySQLObjectStore would need to be hacked but that's easy enough.
  Except: you have retrieveLastInsertID taking no arguments, assuming
  that the database will keep track of the last inserted ID on a per
  connection basis, I assume?  Postgres can do this through one of two
  methods: get the OID from the insert statement and look at that row,
  or get the last returned value from the sequence corresponding to
  that class.  But in one case I would need the return value for the
  statement and in the other I would need the object inserted.

  While I could just override commitInserts and do it myself, and I
  will have to end up doing that if I want this to work with the 0.5.1
  release, it would seem better to make the database layer a little
  more flexible.

o I'm pretty sure I know how to retrofit transaction support onto the
  object store, which speaks well for the design; I can even get
  revertChanges in and working.  But as revertChanges isn't
  implemented I'm a little unclear about how it is intended to work.
  If I work with a store and saveChanges a dozen times, and then
  revertChanges once, what does reverting mean?  Does it mean `go back
  to the state things were in when I started working with the store'
  or does it mean `don't do what I've been working on since the last
  save' or does it mean `undo what I did last save' or what?

o How do you run the tests?  I'd like to be able to make my changes
  and then run all those tests to make sure I didn't break something,
  but I'm not entirely sure how to get them to go.
- -- 
Graham Hughes <gr...@ly...>
PGP fingerprint: 1F1D 0027 B835 E114 3F5B  2C7C 64D1 83A0 C5C7 312A
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.0.4 (GNU/Linux)
Comment: Processed by Mailcrypt 3.5.5 and Gnu Privacy Guard <http://www.gnupg.org/>

iD8DBQE6zPkgZNGDoMXHMSoRAtBiAKDm8DEaM9nkPhdMYGLzfUzHjX1TfwCg6z2N
CgZm2c29aG5rmNB8c9LIr+o=
=EpI3
-----END PGP SIGNATURE-----