Slightly nicer error message when a body is all definitions.

[stamourv]

Currently, this program:

#lang dssl2
def foo():
    def bar():
        3

raises an error in terms of the Racket expansion.

The PR catches these kinds of errors, and produces a slightly nicer error message.

Two issues that remain:

• I couldn't get it to print the actual DSSL2 surface syntax where the error is, but rather the AST version.
• I didn't add the above program as a test; do you have tests for syntax errors? I didn't find any.

[tov]

I don't think I do have syntax error tests…

[tov]

Is it necessarily wrong to end with a definition? Instead of improving this error message, maybe we should make the result None in that case.

[stamourv]

Is there a non-mistake situation where you have definitions, but no expression in the body?
I'd imagine that that error would catch a lot of oversights, e.g., missing recursing starting point.

[tov]

It’s never ultimately useful, but there are definitely cases where you could temporarily have a definition last in a block. And notably, Python allows it.

[stamourv]

Alright, either way works for me.
The more time-sensitive PR is the other one anyway. Any ETA on that one?

[tov]

So it turns out it wasn’t just def that did this—let-with-no-RHS, struct, class, and interface did as well. I implemented my solution on a branch, block-end-fix. If you’re happy with that then I’ll merge it.

[stamourv]

That looks fine, I'm happy with that.
No tests?

[tov]

Yeah, okay, a couple tests.