Brian Groenke

No. There is a loop there for a reason. As I pointed out in the other comment, it iteratively reconstructs `m` for each column. It's not a `map`-like operation. There...

Ok, I'll try `foldl`.

It makes it type stable but does not have any substantial impact on performance, at least on this test case that I tried: ```julia m = Model(s1) xs = Tables.columns([(val=5.0,bounds=nothing),(val=4.0,bounds=nothing)])...

It's also worth noting that indexing tables with `Tuple` is not standard practice. I don't think `Matrix` or `DataFrames` support it.

I just started using this branch in my "real world" use case and it's.... not good. Like almost a full minute to compile on a deeply nested model with about...

It's probably going to be `reconstruct` :(

I guess we should come back to this at some point.... I am no longer convinced that there is a strong enough use case for type-stable table updates that is...

@rafaqz @briochemc As mentioned in #39, I would push back against using type matching to accomplish this. This has the (imho significant) disadvantage of hard-coding into your types what is...

It's also worth noting that any custom subtypes of `AbstractParam` would need to (redundantly) follow the current structure of `Param` by at least providing a field `parent` with a type...

> That redundancy is common to all inheritance in julia. I don't think so. That depends on how you use inheritance. I typically use inheritance to group related types together...