Now that we have transitioned from Mongo to SQL, and sqlalchemy is providing our database abstraction, it does not make lots of sense to abstract away the database abstraction layer.

I seriously suggest ripping out Mongo. I have actually done that with no ill side-effects (besides not supporting Mongo anymore) in this branch:

WIP/RIP_mongo repo: ​git://gitorious.org/~spaetz/mediagoblin/spaetz-mediagoblin.git

This leads to quite some code simplification:

45 files changed, 1518 insertions(+), 3107 deletions(-)

the removing of an additional lookup layer and possibilities for further optimizations. (Such as transitioning away from the mongo-style alias "_id" to the "id" field that we actually use)

A user accepting "klingon, de-de, de, it" will currently be served english web sites, as we try to serve the 1st choice and fall back to english if not available.

(WHY don't we have a klingon translation by the way???)

We need to fetch a list of available translations and match that up again the browsers accepted languages coming up with the best choice to use.

For this we can use 2 options: werkzeug.accepted_languages.best_match(['de', 'en']) which would be good, but probably is not sufficiently clever: initial test seems to show that preferred: "de-de, "en" available: "de", "en" would negotiate "en".

Alternatively there is "babel.negotiate" which seems to be cleverer and return "de" if preferred "de-de, en" and available "de, en".

Investigate and fix.

The audio and video media types both write to temporary files instead of using the workbench... the workbench was designed for this though as a place to put files during processing that get removed when things are done.

The NamedTemporaryFile bits should be changed over to use the workbench unless there's good reason not to, and the end of these files' processing routines should call:

workbench.destroy_self()