BUG: DrRacket REPL corrupted (#%top-interaction: bad syntax) after error in definitions area

[sohang3112]

If a #lang rhombus file fails to compile (e.g., due to an unbound identifier), the DrRacket REPL enters a broken state. Any subsequent expression typed into the REPL fails with #%top-interaction: bad syntax.

Note: This only happens after an error in the definitions area, not from errors raised within the REPL itself.

Current Workaround: Running a file that compiles successfully resets and fixes the DrRacket REPL.

Steps to Reproduce

Make a file with the following code (deliberately having error) and run it in DrRacket:

#lang rhombus
 x

As expected REPL shows error when file is run. But after that, trying to run any valid code in DrRacket REPL also fails with syntax error:

x: unbound identifier in: x            
$ def a = 0      
. #%top-interaction: bad syntax in: (#%top-interaction . def)

[mflatt]

This is indeed a limitation of DrRacket. More generally, it's a consequence of Racket's multi-language approach. Approximate workarounds are possible, but it has not been clear working around the failure would be a good idea.

Specifically, it might seem that #lang rhombus puts DrRacket in Rhombus mode, and then the body of the module is processed. But #lang rhombus actually hands over control for the full meaning of the module to rhombus, and DrRacket's REPL sees only the result of that process. If module processing goes wrong due to a syntax error, then DrRacket has nothing to work with in the REPL.

It would be possible to have a #lang protocol that provides a fallback when module expansion fails, in much the same way that #lang protocols support syntax coloring and indentation of incomplete programs. There's a danger of that fallback providing misleading feedback, since it would not really match the module that failed, but it might be a good idea. Or maybe DrRacket should be more restrictive and refuse to allow REPL evaluation if module evaluation did not clearly succeed.

[sohang3112]

It would be possible to have a #lang protocol that provides a fallback when module expansion fails, in much the same way that #lang protocols support syntax coloring and indentation of incomplete programs. There's a danger of that fallback providing misleading feedback, since it would not really match the module that failed, but it might be a good idea. Or maybe DrRacket should be more restrictive and refuse to allow REPL evaluation if module evaluation did not clearly succeed.

IMO the first option is better and it's how other languages like Python (when run with -i interactive option) do it - in case of syntax error in file, print the syntax error and then start an REPL in a default Rhombus environment.

[rfindler]

DrRacket has attempted an approach like that before but there is no official way to get a "default language X environment" and attempts like "start the file with just #lang X followed by a newline" do not work in practice. For example, #lang racket/unit is a syntax error because of the missing import declaration. I think we would need to do more at the #lang protocol layer in order for DrRacket to handle this situation.

I do think it would be good to do something here to improve the current situation.