An experimental version of Yaki for Google App Engine, published as sample code.
I initially decided to do this as a quick and dirty test to see whether it would be viable to use the NoSQL datastore for raw markup and very basic metadata, and ended up realizing that this would be a great chance to finish refactoring Yaki as an MVC/WSGI app with a well-defined set of constraints regarding storage, task queues, etc.
That and try out a PaaS, which I've never really done (I favor sysadmining my own machines, but there's only so much time in the world). GAE isn't perfect, but it's well documented and a good fit for my usual development pattern.
So, among other things, this version aimed to do away (almost completely) with the yaki package and standardize the internals a bit more.
However, after finishing that, I realized that it was prohibitively expensive to crawl the remote data store for changes, so I decided to put this on ice, strip away the GAE bits, and move development to a private Bitbucket repo.
This is pretty usable on its own, provided you're willing to write a way to manage the content store rather than sync it with a cloud store service -- it's certainly feasible to run a self-contained blog engine atop GAE using this as a basis, but a filesystem is just more convenient for me.
The base premise of Yaki is that all my content is stored as flat files -- rather than generate a static site out of five thousand files and keep them all around, I'd rather have Yaki keep track of metadata and render pages as necessary.
In order to work inside App Engine, I tossed out direct filesystem access and replaced it with Dropbox (soon MEO Cloud) API access -- so Yaki is, in effect, acting as a caching proxy to the API, storing pages in the datastore as users navigate the site.
One of the main constraints I have is that I need to support multiple markup formats (such as Textile, which is still superior to Markdown for a number of use cases).
Were it not for that (and the superior feature sets of most Python markup renderers) and I would probably be rewriting this in Go.
Given that I'm re-using a fair amount of old code, the coding style is uneven and there is an uneasy mix of singletons, static classes, functional code, rather excessive memoization, etc. One of the priorities here was to minimize datastore and cloud store requests.