Earlier quoted context omitted.
Yeah I would expect to see it compile to bytecode. It seems pretty wild to transpile to a language to compile to bytecode to run in a vm when there is already bytecode and a vm to execute against. Transpiling is common in js only because there isn't/wasn't a standard like webasm.
Similarly, despite the existence of LLVM and the simplicity of the various assemblies, tons of languages target C, mostly to facilitate easier interop with native libraries and because pretty much everything has a C compiler. Perhaps there were similar reasons to compile to Elixir, but I honestly can't think of any. It might just be for faster compiler development, at least until the featureset of Elchemy begins to s…
First of all thanks for all the interest and a good question! While obviously it would be possible to compile directly to Erlang, or even to Core Erlang it would just add a lot of additional work to do. Elixir adds a phenomenal amount of functionality that made making Elchemy easier. And thanks to building it on such a high level we can leverage most of the Elm tooling goodness as well as Elixir's.
Another very important factor is familiarity. People much rather like an environment they already know and it would be silly to try to recreate things like Mix or ExUnit yet again.
Another important aspect is the familiarity of the output. Elchemy's priority is producing fully readable Elixir code, which not only adds another layer for learning the language's characteristics as well as a 'belt and braces' for experimenting with it in existing projects - the time cost of removing Elchemy from a project written in it is way smaller than rewriting the logic from scratch