On Thu, 4 Jul 2002, Ashley Winters wrote:
: I was pondering how to implement the apocalypse 5 stuff (only pondering) and I
: was wondering if <Inf,0> could be legal, indicating a greedy match.
:
: * = <Inf,0>
: + = <Inf,1>
: ? = <1,0>
: *? = <0,Inf>
: +? = <1,Inf>
: ?? = <0,1>
We could autoreverse, but it'd be a bad idea. It doesn't work out
well when you want to parameterize it or write a code generator.
We don't do it for ranges either, for similar reasons.
: Speaking of the range assertion, is there anything other than <x,y>? There
: used to be discussion on the list about adding more possibilities, but I
: didn't follow it.
We also allow <$x,$y>, <x,$y>, and <$x,y>, but that's about it. There's some
talk about disambuguating from <$x> with a leading +:
<+$x,$y>
So possibly we could make
<-$y,$x>
count down instead. But if we're gonna go all general on this, it'd be
better to do it the same way ranges do.
<$y..$x:-1>
That's still ambiguous with <$x>, though. Perhaps the general form is
just
<* $start..$stop:$step >
That devolves nicely to
<*2>
to match twice.
But there's one clinker in the works. Currently
{1,10}
is a maximal match, and you write
{1,10}?
to get the minimal match. But if
<*$min..$max>
defaults to a step of 1, it's a minimal match, and we write the
maximal match with
<*$max..$min:-1>
That goes against the historic precedent that the maximal match is
the unmarked form. Much as I feel tempted to make minimal matching
the default across the board, I'm not sure we could pull that one off.
Too many neurons hardwired to think .* is greedy.
So I'd guess that we just don't talk about :-1, but rather say that
<*$min..$max>
is naturally greedy, and as with any quantifier you write
<*$min..$max>?
to get minimal matching.
But sigh, it would fix so many novice bugs to make minimal matching
the default...
Larry
