Re: [Followup] Re: "Near Miss" error handling?

I suggest that you not, like Interlisp, proceed to allow execution of
the program once corrected. DWIM scored its corrections, and if they
were low enough scoring, it silently corrected the code. As a result,
the first time I used this system, I only realized half a day later
that the system was case-sensitive, since it had been silently
correcting the code as I enterred it. Sometime after lunch, it say
something like, "You seem to regularly be mispelling 'CAR' as 'car',
would you like me to record this change permanently?" Had I seen the
message earlier, I wouldn't have repeated the mistake. This
experience also suggests that multiple case errors counted as no more
than a single character error. I believe it was: case, consecutive
char transpose, char doubled, char dropped char inserted in that
order.


On a related note, I have seen systems regularly rebuilt with hundreds
of errors output, but the source was never corrected because it wasn't
necessary -- carriage returns while compiling allowed systems to be
successfully built and released. Such systems were still being
debugged and extended without ever fixing the reported errors. I
would suggest that you at _most_ offer to produce an equivalent
corrected source on output, so that the user can only proceed with a
cleanly compiling file. Interactively offering to correct a typo is
reasonable to help out a user at a keyboard, but leads to weaker
systems when extended to compiles from source files.


-Scott David Daniels
Scott.Daniels@Acm.Org

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