I'm a grad student that works in Racket and fwiw I find that I prefer emacs for editing but DrRacket for testing and debugging. It has great tools for documentation, locating errors and macro expansion and of course it all works out of the box. I also use Geiser, but I've found it buggy and unreliable (including instantly shutting down my OSX machine if I tell emacs to kill it when I quit, plus I still haven't figured out how to limit its memory consumption).

Just to be nitpicky, Racket is not a Scheme dialect (hence the name changing from PLT Scheme to Racket a few years ago (already!)). It's not because it not only adds a lot to Scheme but also has somewhat important differences. For instance, `cons` pairs are immutable in Racket: there's no `set-car!` or `set-cdr!` as in Scheme (you can have mutable cons pair using `mcons`, `set-mcar!`, and `set-mcdr!` though). Also the macro systems that Racket uses (`syntax-parse` and co) is quite more advanced than the current Scheme rNrs one.

Well, #lang racket is not scheme, but one can run rnrs in the racket runtime. There are many languages that run on the racket runtime.

IIRC, Arc/HN runs on mzscheme because they rely on mutable cons cells that newer versions (racket) don't provide. Could they run under new rackets by just telling racket to use a proper scheme?

As a Racket author, I can tell you that we provided them with a hack to enable mutation on arbitrary conses, so they can upgrade whenever they wish, and I don't know whether they have or not.

There is an mzscheme language mode, but it specifically does not include set-car! and set-cdr!.