better error when using `=` in boolean position

[shriram]

This has got to be a not-uncommon error:

n = 1

if n = 0:
  print "yes"
else:
  print "no"
end

whether for people coming from other languages, or from typos (I know I've made it a few times). Is a better error possible?

[jswrenn]

I very curious about the frequency with which this error is encountered, but I don't have good data here because it's surfaced as a generic parse error.

To catch this, we could treat let bindings as an expression (rather than a statement) and immediately curtail it using a well-formedness check. This might be as simple as moving let-expr from stmt to expr (pinging @blerner). However, this would amount to extending the grammar for the sake of error reporting.

[blerner]

IIUC, the problem here could be caught by adding the EQUALS token to the bad-expr production. I think this would allow parse-pyret to recognize the token much like it does other bad tokens, and raise a custom error. It would make definition statements locally ambiguous, but since bad-expr is not derivable from program, that's ok, it won't affect overall ambiguity. (That production is a hack to get the parser state offset just enough that we can sometimes get a better error message, and works solely because we're triggering off a single unexpected token.)