Right, but the functionality is implemented with lm(@formula(y ~ x), data::DataFrame). So the question is more, why is it only defined for a Tables.jl data-format? It should be dead-simple to implement the method for lm(x::Vector, y::Vector), right?
Well of course using GLM without DataFrames is niche - the usage is not documented!
Not sure why using lm is relevant - it simply calls fit(LinearModel), right?
I would argue that univariate regression is the most normal, but that is simply a matter of different day-to-day encounters with regression. To me, fitting some mathematical equation to relate two physical sizes, say voltage and current, is all I do.
I would also say that by not prioritizing support for simple usage (univariate, no DataFrame), one creates the need for packages like EasyFit.jl, instead of having a single package that can do it all. If there is one thing the Julia ecosystem does not need, it is more splitting into packages that do similar things slightly differently. And it seems like GLM is more than capable, which is why I think it is a shame - GLM would simply need a little more catering towards simple usage, to make it the GOAT of linear fitting in the Julia ecosystem.
(I am rather new to linear fitting, so take my statements on the state of linear fitting in the Julia ecosystem with a grain of salt. But this is the way it currently seems to me.)