In his e-mail message of 1993/02/24, Jean Michel writes
    The new ranges are an improvement, but...
    I find the fact that <high>-<low> must be divisible (the error message
    says divisable which is a misspelling) by <inc> to be slightly
    annoying. I had in my code thinks like:
        [1..QuoInt(n+1,2)]*2-1
    to specify the odd integers smaller than n, and
        [1..QuoInt(n,2)]*2
    to specify the even integers less than n. I hoped to be able to rewrite
    these as
        [1,3..n] and [2,4..n]
    but the first one works only for n odd and the second for n even!
    One has to use
        [1,3..n-1+(n mod 2)] and  [2,4..n-(n mod 2)]
    which is less pleasant and efficient.

What is the rationale for the restriction?

The rationale is that a construct such as

for i in [1,3..n] do ...

has a visual clue that <n> is in fact a member of the range, and that it
will be the last value of <i>. That is a programmer might
(unconsciously) follow that <n> is in the range (in the membership
sense), because it is in the range literal (in a textual sense). But
this visual clue could be misleading if GAP would be silently rounding,
because then in the above example <n> would be a member of the range only
if it is odd.

I have to admit that this is a somewhat weak argument. If one would
apply it consequently, '[<i>,<j>..<k>]' would also have to generate an
error if '<k> < <j>', because in this case <j> is not in the range, even
though it is in the range literal.

Actually, I don't really care one way or the other. If other users would
also prefer silent rounding, I am quite willing to change the definition
and the code accordingly (which is not a big deal).

Martin.

-- .- .-. - .. -.  .-.. --- ...- . ...  .- -. -. .. -.- .-
Martin Sch"onert,   Martin.Schoenert@Math.RWTH-Aachen.DE,  +49 241 804551
Lehrstuhl D f"ur Mathematik, Templergraben 64, RWTH, D 51 Aachen, Germany