I personally tend to use 2-3 weights when building applications: Regular and Fill weights to denote statefulness (active/inactive) of icons at a larger size, and often using Bold for small indicator icons. This works well for my use case, and I like the comfort of how it is set up now.

I'm not totally against your proposal, but I'd want to hear it from more people before agreeing to a big breaking change like this, even in Elm where breaking changes are pretty easy. How many icons are you using in your application that the extra cost per icon is too much? Can you share some numbers? Would love to see real-world savings, as I not sure this is the source of any real issue.

To your other point though, I'd be happy to move the codegen to elm-codegen. This was initially set up at the same time as some of our other JS libs and it made sense at the time to copy and tweak the stuff I had already written.

I see how if you regularly change the weight then the current API is nicer.

I wonder if you could simply have the best of both worlds by having the per-weight modules in addition to the current single module.

I'm cool with that, if you'd like to make the change! Are you thinking to expose as a separate or child module so there aren't namespace collisions?

I'd call them Phosphor.Regular, Phosphor.Duotone, ...