I'm probably missing something here, but if the parent of bslib-page-main is not fillable, then AFAIK this doesn't need to be a fill item?

Maybe not? But I know for sure that, without having to consider parent containers, we want to break fillability for children by making sure the container isn't a fill container. We could probably also set item = fillable, but then we have to consider all of the parent layouts. page_main_container() isn't just used in page_sidebar(), it's also used in navbar_page() with a sidebar where the DOM structure is quite different from page_sidebar().

I know for certain that removing html-fill-container does what we want it to do in this case and that that's correct in all layouts. That's also why I used .html-fill-container for min-height/min-width above. My general take is that .html-fill-item is upward-indicating, as in it makes the child filling when the parent context is fillable. .html-fill-container is downward-facing and communicates intent downward to its children.

My take is that we could probably either make .bslib-page-main a fill carrier when fillable = TRUE or add no filling properties when fillable = FALSE, but we could certainly work with the fill container property. So this choice mostly came down to 1) proving we want container = fillable is easy; 2) proving we want item = fillable requires enumerating all the layouts and making sure that's right.

Ah ok, I was missing that navbar_page() context, thanks for writing that out. Although I'll probably find the code confusing to revisit, I like the careful approach your taking. Maybe consider linking to your comment here in the code?

Well, in the mean time I convinced myself that we should go wholesale as_fill_carrier() when fillable=TRUE and just wrap things in the <main> tag otherwise.

I think the reason I used as_fill_carrier() was in case we use page_main_container() in other situations; the fact that that also adds display: flex was the unintended side-effect.