Recently I created a spike of a view engine, in which views are plain classes, and the content is created by using funny using-scope blocks.
The code together with a simple sample site is available at http://code.google.com/p/sharp-view-engine/
Here I'd like to hear your opinions regarding such an idea. Is it completely weird or maybe someone likes it?
I would actually not like that.
I can agree with DSLs (such as a Parser-Combinator or for generating XML Nodes in a data-context), but in this case I think that too much is being put in code that. And, in the end, this just complicates boundaries and leads to hard-to-maintain code. (You can already do the same, but with more verbosity just using the "standard" Web Controls. You can always use {subblock} in C# to limit a variables scope.)
The approach I prefer to use is templates with bindings (but no "code in templates"). That makes it easy for the "designer" (hopefully not me, or the next person to come along and) edit the layout of the view how they see fit. However, the core logic (the available controls and bindings) are kept in the code -- uncluttered. (Another advantage with the templates is that if they externally housed they do not require a recompile for every little change.)
Simplicity and maintainability are like ... zen.