invalid continue label

[Dan Kortschak]

Can someone explain why this is illegal (./cluster.go:40: invalid
continue label ATTEMPT)? Should I just be using goto here? My use here
feels sane, and continue indicates what I mean, but it seems that the
label must immediately precede the for loop - is this the case? It seems
restrictive.

thanks
Dan

func ClusterFeaturePairs(p []*pals.FeaturePair, epsilon float64, effort int) (pc [][]*pals.FeaturePair, rf []*feat.Feature) {
km := kmeans.NewKmeans(Pairs(p))

values := km.Values()
cut := make([]float64, len(values))
for i, v := range values {
l := epsilon * (v.Y() - v.X())
cut[i] = l * l
}

for k := 1; k <= len(p); k++ {
ATTEMPT:
var e int
if k == 1 {
e = 1
} else {
e = effort
}
for attempt := 0; attempt < e; attempt++ {
km.Seed(k)
km.Cluster()
centers := km.Means()
for i, v := range values {
dx, dy := centers[v.Cluster()].X()-v.X(), centers[v.Cluster()].Y()-v.Y()
ok := dx*dx+dy*dy < cut[i]
if !ok {
continue ATTEMPT
}
}

pc = make([][]*pals.FeaturePair, k)
rf = make([]*feat.Feature, k)
for ci, c := range km.Clusters() {
rf[ci] = &feat.Feature{Location: p[0].A.Location, Start: int(centers[ci].X()), End: int(centers[ci].Y())}
pc[ci] = make([]*pals.FeaturePair, centers[ci].Count())
for pi, i := range c {
pc[ci][pi] = p[i]
}
}

return
}
}

panic("cannot reach")
}

[David Symonds]

Put the label before the for loop instead of inside it.

[Dan Kortschak]

But that's not what I mean. It is before the relevant loop. If I replace
continue with goto I get the behaviour I intend.

[David Symonds]

It has to be immediately before the loop. If you want it where you have it, use goto instead.

[Dan Kortschak]

OK. That was changed in d4f8b77723db?

[Russ Cox]

On Sun, Oct 7, 2012 at 7:09 PM, Dan Kortschak
<dan.ko...@adelaide.edu.au> wrote:
> Can someone explain why this is illegal (./cluster.go:40: invalid
> continue label ATTEMPT)? Should I just be using goto here? My use here
> feels sane, and continue indicates what I mean, but it seems that the
> label must immediately precede the for loop - is this the case? It seems
> restrictive.

Yes, this is the case. I don't know that it is restrictive, it's just
trying not to be error-prone. Continue is an operation on for loops.
When you say continue, it refers to the innermost for loop. If you
want to specify a different loop to continue, you have to label the
loop, not a statement inside the loop. Requiring that the continue
label actually label a loop avoids errors guessing what loop is meant
when the label points elsewhere. For what it's worth, I believe this
is the same behavior as in Java. This has not changed in many years.
The revision d4f8b77723db that you found was fixing a corner case
where in

L:{}
for x := range y {
break L
}

the L label was still treated (incorrectly) as labeling the for loop.
But a label inside a loop body has never been considered to label the
loop holding that body, and that revision predates the public release.
The label behavior has not changed since the public release.

If we allowed the label to go either inside the loop body or before
the loop, as in

for x := range y {
L:
f()
for z := range w {
continue L
}
}

then if we delete the call to f, the label becomes ambiguous.

Russ

[Dan Kortschak]

Looking at my repository more carefully (it was around June the last
time I touched this code, so there is no reliable memory of what I was
doing) I think the situation is that I had made the change of including
the conditional block between the label and loop and then moved on to
something else without testing (the change was not committed).

Thanks for taking the time to explain.

Dan