🚀 go-pugleaf

In article <46jiqv$pl9@solutions.solon.com> diamond@jrd.dec.com (Norman Diamond) writes:
> In article <46cu3i$6fg@solutions.solon.com>, danielv@crt0.crt (Daniel Villeneuve) writes:
> >Note that `undefined result' seems different, as of my reading of the
> >standard, from `undefined behavior'.
>
> I agree.

Me too, but Clive D.W. Feather <clive@stdc.demon.co.uk> said, on the same
question:
] There is a separate term "unspecified" which means a valid, but not
] described, result. Undefined result is the same as undefined behaviour.

I cannot decide on that, because neither `unspecified result' nor
`undefined result' are defined in the standard in section 1.6.

> However, qsort() or bsearch() has to be called with an array in the first
> place.  It can be an array of structures but not a list of structures.
> So, I guess you're not trying to call qsort() or bsearch().  What do you
> mean by a conforming comparison function?

Maybe the expression `a conforming comparison function' is not clear.  In
the context of my previous article (keeping ordered sets of structures), I
wanted a comparison function that could be called by a sorting device (such
as an AVL tree interface) to order items unambiguously, given that no two
items were the same.  No more hypotheses are made on those items.

After having read the previous replies about that article, I agree that the
comparison operators < and >, if fed with invalid arguments, could lead
to `undefined result' that could be like random (if the result exists).

So my comparison functions, to be conforming, should rely on some unique
identifier in the structures (such as an integer telling the position of
the structure in the global list), instead of the pointer on each
structure, even though this pointer _is_ a unique identifier for the
structure.

As Stephen Baynes <baynes@ukpsshp1.serigate.philips.nl> said,
] It is a bit annoying that there is no full ordering on pointers as are a 
] number of occasions it would be useful.

--Daniel