The next thing to tackle in the new codebase is the image directory. The current implementation is hard to maintain and clearly not scaleable, so I'll outline my proposed implementation here and ask for feedback.

Also, I'd like it if there were some convenient way for me do grab a tarball of Wikipedia's image directory--maybe an FTP logon?

Anyway, here's the implementation: First, I've already implemented [[image:name.jpg]] links, so that users won't have to know about the internal image directory structure, and so that wikitext will be more transportable.

I think the internal directory structure should be de-flattened at least one level. We can't count on all installations using ReiserFS, so a single directory with hundreds of files is a performance problem. It might even pay to go two levels. This means that images will be placed under directories based on the first letter or two of their name, e.g., "/upload/c/chessboard.png" or even "/upload/c/ch/chessboard.png".

The database conversion tool will have to add a function to find all the existing image links and convert them.

I've added a database table which contains for each image its name, size, the time it was uploaded, and by whom. So we'll be able to get a listing of images sorted by name to make maintenance possible.

The upload function should enforce some reasonable file name conventions, and it should warn the user who tries to unsupported file formats or very large files (though perhaps the user should be able to override these warnings).

It would be _very_ nice to have a feature to search for pages that use a particular image. I can't think of any way to do that but a fulltext search, but possibly that could be made reasonable by enforcing naming conventions (for example, requiring a minimum length, forbidding problematic characters).

Any other issues I haven't thought of? 0