Re: "Near Miss" error handling?

Gwyn Judd wrote:


> I'm writing (modifying actually) a compiler for my final undergraduate
> project and I've come across a feature I've never seen in a production
> compiler. basically when the compiler comes across an identifier it
> hasn't seen before, it will go through the list of known identifiers
> and try to determine which is the closest so it can then make a
> hopefully helpful suggestion on how to correct the error. I didn't
>


Most of these routines are dramatically simple, and the technique is
covered in the compiler construction books. Given an undefined label,
the checker tries to check for:


1. One character wrong.


2. One character missing.


3. One character added.


4. Two adjacent characters transposed.


It was suposedly based on a study of common identifier errors. You
need to except one and probably two character identifiers from the
system, since they will yield false results. The routine I use for
this is pretty small, it just takes the match string, mangles it in
the above ways, then runs a search again and iterates for every
character, based on the idea that having an error is not common, so
does not need a blindingly fast algorithim. You might also want to
flag the labels as respelled, to cut down on double errors.


--
Scott A. Moore is samiam@cisco.com


I don't have a solution, but I admire your problem.

Return to the comp.compilers page.
Search the comp.compilers archives again.