A custom query class allows to add more methods on
queries (read: "cursors").
This custom query class especially adds a .sort with a
calling convention exactly like the mongo one.
Makes a lot of existing code happy!
sql/fake.py contains some fake classes and functions to
calm the rest of the code base. Or provide super minimal
implementations.
Currently:
- ObjectId "class": It's a function mostly doing
int(first_arg) to convert string primary keys into
integer primary keys.
- InvalidId exception
- DESCENING "constant"
- This adds a new SQL table field type for path tuples.
They're stored as '/' separated unicode strings.
- Uses it to implement a MediaFile table.
- Add relationship and proxy fields on MediaEntry to give a
nice media_files "view" there.
- Let the converter fill the MediaFile.
A bunch of functions on the db objects are really more like
"utility functions": They could live outside the classes
and be called "by hand" passing the appropiate reference.
They usually only use the public API of the object and
rarely use database related stuff.
Goals:
- First, simple: Share the code with the SQL objects, so
that the code doesn't need to be duplicated.
- Second, it might unclutter the db models and make them
more into "model only" stuff.
- Doesn't really hurt.
Inside the mongo db backend, use the mongo
MigrationManager. This is hopefully the last reference to
the generic MigrationManager reference on db.util.
1. Use the new setup_connection_and_db_from_config in the
sql backend.
2. Use sql and mongo specific functions wherever
appropiate instead of the generic "db.X" one. This makes
the converter more indepedent of the current backend
choice.
When initializing the database connection the current mongo
based setup checked for new migrations and warned about
them. This was mongo specific so factor'd it out into a
more generic check_db_migrations_current function in the
mongo backend.
Also created a dummy one in the sql backend.
This is a shortcut to adding the object to a session (if
needed) and giving a commit on the session.
In reality, calling code should probably utilize the
session on its own and call commit in an appropiate place.
MediaEntry now has a get_uploader (property) loading the
appropiate User object for the MediaEntry (and caches it).
MediaComment has the same for author as get_author.
While creating the new SQL model, the "state" field of
MediaEntry was left out. Currently using a plain unicode
string for it.
Maybe should use sqlalchemy.types.Enum?
sqlalchemy gives autoloading (hopefully caching) link to
other objects as properties. So turn get_uploader on the
current mongo based stuff into a property to ease
transition.
Instead of creating the email verication key on the db
model as a default for the field, create it in the
registration view.
Now all verification key generation is only in
auth/views.py!
1) MediaComment's author method conflicts with the author
field. So rename it to get_author.
2) Turn it from a normal function into a python property.
That means you call it by ".get_author" not by
".get_author()". This is exactly what sqlalchemy gives
us free of charge.
Starting to move the mongo specific stuff into db/mongo.
And create thin "from db.mongo.Y import z" wrappers in
db/Y.py.
Why?
1) Will make it lots easier to switch to sql for
testing/developing.
2) The mongo stuff needs to stay around after moving to
sql, because the converter needs it.
In trying to ease the migration to SQL, created an
interface to sqlalchemy that looks a lot like the interface
that is currently in use.
*WARNING* Work in progress
This is just a start at a Migration tool from Mongo to SQL.
It fills all currently available SQL models with data from
MongoDB. A few fields in the SQL tables are left out,
because some data format migrations are needed (notably:
queue_file_name).
This thing lives in mediagoblin/db/sql/convert.py because
it has a lot of stuff hardcoded and is not, repeat not for
end users!
Hard coded:
- output database: ./mediagoblin.db (sqlite)
- Mediagoblin config: ./mediagoblin.ini
Run bin/python mediagoblin/db/sql/models.py and watch the
create tables on a memory sqlite db.
Also unicode strings need unicode defauls. Warning by
sqlalchemy.
The .uploader() method conflicts with the uploader database
field. As we're moving to .FIELD for db field access, this
is a relevant conflict.
So renaming .uploader() to .get_uploader()