On Sun, 9 Jun 2002, Ton Hospel wrote:

> In article <[EMAIL PROTECTED]>,
>       "Sean O'Rourke" <[EMAIL PROTECTED]> writes:
> > From the rules:  "A word contains at least one alphanumeric character
> > (letter or digit), and optionally some punctuation (. , ; : ' " ( ) & /).
> > Note: This means that punctuation characters can never occur alone. But
> > except for the period, which must end a sentence, punctuation can occur
> > anywhere in a word."
> >
> > I took this to mean "punctuation, including '.', can occur within a word,
> > but since '.' can end a sentence, only non-'.' punctuation can end a word.
> > This seems like a more careful reading which, if universally followed,
> > would have resulted in a lot more solutions with /\S*\w\S*/ in them.  But
> > the refs, in writing the tests, either did not test this corner case, or
> > interpreted the rule the other way.
> >
>
> ....the period, which must end a sentence...
>
> So periods end a sentence. if there is a period in the middle of a word,
> it's now two words, one on the end of the last sentence, and one at
> the beginning of the next.
>
> How else do you decide where a sentence ends in fact ? whitespace/newlines
> were optional between sentences.

I take them to be required: "A paragraph consists of one or more
sentences, separated by sequences of whitespace characters (spaces and/or
at most one newline)," and "a sentence consists of one or more words,
separated by sequences of whitespace characters, where the last one ends
in a period."  Combined with the above rule, I take this to mean that
end-of-sentence is "/\.(\s+|\z)/".  Unfortunately, while neither of us
appears to have any doubts about the rules, we don't quite agree...

> Nope, if dots in the middle were allowed, the whole -056 concept would have
> broken down.

You're right here -- I was on crack.  Maybe this could be next month's
challenge ;).

> > hour, what should have been done?  A test suite is unambiguous -- you
> > either pass it or you don't.
> >
>
> Nah, that would be unfair.
> [potential score-shaving measures]

I would say that in these cases either (1) the tricks would clearly
violate the rules, and the refs would therefore notice them, or (2) the
tricks would exploit corner-cases the refs hadn't thought of, in which
case it's not necessarily clear what the rules intended in the first
place.

> A purely testcase based approach would also make challenges with random
> numbers like in the ircnet golf almost impossible. It invalidates
> the "either you pass or you don't".

How did you automatically test the entries?  The score program's no longer
on the web.

> Also have look at the winner of the lapm 1 challenge. It will
> succesfully finish the testcase, but will not finish until long after
> the sun becomes a white dwarf. Reject or not ?

If it can't finish the test case, I think either it should be rejected, or
the golfer should argue that the tests be changed.

> I don't really think there is a "simple enough that the intent is clear"
> in general.

I think Kolakoski and Cantor both qualified here, as will most other
math-derived holes.  Not that I want to restrict our hole selection to
just math.  In some sense, every spec will be ambiguous, but clearly magic
number's ambiguity is nothing like interlinear's.

/s

Reply via email to