The slice indexes must be within the length of the slice, not its capacity. Please read the spec more carefully: https://golang.org/ref/spec#Slice_expressions

For arrays or strings, the indices are in range if 0 <= low <= high <= len(a), otherwise they are out of range. For slices, the upper index bound is the slice capacity cap(a) rather than the length.

Note that it says "upper index bound". You can do:

a := make([]int, 0, 5)
fmt.Println(a[:5])

That is, you can use a slice expression to extend length up to capacity. But arbitrary slice expressions outside of the length range aren't allowed. Perhaps the spec could be made clearer, but I think it's clear enough as it is. I'll leave that to @griesemer.

low<=high
low = cap(a)

2. why the following code works fine:

a := make([]int, 0, 5)
fmt.Println(a[5:5])

and

a := make([]int, 0, 5)
fmt.Println(a[5:])

does't

I have an impression that the low index in s[low:high] must not be larger than len(s) for a long time.
I just rechecked Go spec, I was wrong.

fmt.Println(a[5:]) should work fine, it is a bug at least since Go SDK 1.7.

Then one section in my article should be wrong. And one SSA optimization in gc is too conservative.

Ah, sorry, it is misunderstanding. The compiler is right.

fmt.Println(a[5:]) doesn't work for it is equivalent to fmt.Println(a[5:0]). :)

But my former impression for the slice rule was really wrong.

fmt.Println(a[5:]) doesn't work for it is equivalent to fmt.Println(a[5:0]). :)

Are you sure? I think omitting the high index just means the length. The spec mentions it:

a[2:] // same as a[2 : len(a)]

I'm not sure what to think of Go 1.7 changing the behavior of this code. Perhaps earlier Go versions behaved correctly and have since been broken. Or perhaps they misbehaved and have been fixed.

Reading the spec again, now I think I agree with @soslanco's reading of it. My initial replies were muscle memory, as I think I've encountered very similar panics in the past.

Are you sure? I think omitting the high index just means the length.

Ah, len(a) == 0, forgive my slowness. I think you're completely right, and we can close this issue.

Yep!
A missing low index defaults to zero; a missing high index defaults to the length of the sliced operand.

Perhaps the runtime panic message could be a bit more explicit. In particular, the issue here is that we're slicing [5:0], which breaks the low <= high property.

Aside from that, I don't see what else there is to do in Go. The program is wrong, and the runtime is right to panic. I'll leave the issue open for now to see if anyone has any opinions on improving the runtime panic messages.

I'll leave the issue open for now to see if anyone has any opinions on improving the runtime panic messages.

I don't think we should change the runtime panic messages, for two reasons:

(1) I'm sure that people are relying on the text of the existing messages in their programs. They shouldn't be, but they are, and it's not worth breaking them.

(2) Adding additional information (such as the bounds involved) has a cost in binary size and execution. With the current ABI, passing bounds to the runtime panic routines requires stack space, which might prevent elimination of the function prolog, which can have a substantive performance cost in hot code. And there's always a binary size impact.

I imagined that would be the case (2), but I wanted to make sure :)

Thanks for all your advice.