My argument here is really for consistency. The beauty of a REPL, for a new user, is being able to hack through a few lines of code, see how they work in practice, and then use them confidently. It really speeds learning a new package, understanding a language feature, or just coding something new.
I'd be perfectly happy if, in the REPL, when I call `myplot = plot(x,y)`, I get back a plot object that I then need to display with `display(myplot)`, or by adding `plot(x,y,show=true)`, or by explicitly defaulting show=true with `pyplot(show=true)`. If that happened in the REPL, I'd quickly learn Tom's paradigm, `plot` "constructs a visualization of the input data" - and I would be benefiting from the REPL as part of my toolkit, and I'd be happy. But when there's a difference between how the REPL works and how the same command works in a code block, then I lose the benefit of the REPL, worse, the REPL teaches me one way of doing things that I then have to unlearn in an non-intuitive way. Whether it's done by default one way or another isn't important if consistency is maintained. I think the disconnect in this discussion is rooted in the inconsistency of behavior between REPL and code block. And I believe that will be a pain point for an untold number of future users. In any case, lovely discussion here. :)
