I fear, implementation-wise the typical Common Lisp implementation quite a bit ahead of Haskell. In something like SBCL I can mix interpreted code, compiled code and optimized code freely. Stack traces are available in all modes, together with an advanced error handlich system able to deal with error conditions at runtime under program and/or user control. Creating new code at runtime compiled or interpreted is available all the time. Remote control (see Slime) of running Lisps via an IDE has been available for a long time.
On a Lisp Machine in the mid 80s, the error system would analyze the error, determine the author of the software system where the error happened and prepare an error mail with machine description, error description, extended stack trace ... to the maintainer's mail address. On a server machine, I would have it sending me such mails automatically...
