I'd like to know how one maintains a huge modular css code base without any sort of preprocessor.
Preprocessors are here because CSS sucks, and no standard committee seems to want to fix it. That's why people built these tools right, to solve a problem nobody was going to solve for them.
That's exactly the same thing with javascript. If javascript was decent we would not need all these compile to javascript languages.
These tools bring diversity and choice to a poor plateform with shamefull languages that are not evolving very fast. You cant fix it just by adding features without fixing the core. that's what preprocessors are for and it's ok to abuse them without shame.
EDIT:
my point is "it's good enough" isnt good enough, in a time where we are supposed to maintain complex front-end projects and asset pipelines, that work on different devices that have their own idea of how to implement (poorly written) standards.
The problem with CSS preprocessors is that they are simultaneously too powerful and not powerful enough. That is, things like SASS mixins and nesting make it easy to make mistakes (specificity overload and huge stylesheets are the two big ones). At the same time they don't change runtime semantics so you can't fix the biggest problems in CSS like specificity wars, finding dead selectors and namespacing.
I do like variables and expressions though (mainly for colors and grid units).
The way you maintain a big modular codebase is to make your stylesheets really stupid and build the smarts into whatever is generating the DOM. The tools we have for generating DOM are both far more mature and we have more control over the semantics of the runtime.
His mention of separation of concerns is, IMO a common misunderstanding of what that term means. The coupling has to be made somewhere after all and it's better to choose the more performant and more explicit way.