I noticed this when looking at Raku/nqp#685.

Why can we assume that the array is in a good state?

Why can we assume that the array is in a good state?

Well, I believe we’re allowed to act as if it’s in a good state. Modifying an array during iteration is not allowed AIUI. Also, the iterator keeps track of the index compared to the size of the array.

Is this PR still good? If so, what would be against merging it?

Modifying an array during iteration is not allowed AIUI.

I couldn't tell if this comment was purely about nqp level commitments or also about Raku level commitments.

For Raku, I'll say this. Shortly after 6.d shipped, jnthn commented under an SO answer by Vadim Belman. Based on my existing belief that modification during iteration was fine, plus Vadim's answer, plus jnthn's comment under it, I'd say it looks like it is "officially" allowed in Raku 6.d.

For nqp, I'll say this. The doc for nqp::iterator says (with my added EMPHASIS):

You can also use nqp::iterator() to iterate over a list, which is functionally equivalent to cloning the list and not having to know which form of shift to have to use.

my $list := nqp::list('a', 'b', 'c');
my $iter := nqp::iterator($list);

while $iter {
    say(nqp::shift($iter));
}

Using iterator this way proved to be not very well optimizable, so this form will probably be DEPRECATED in the future. Depending on circumstances one could use clone instead, or use an index variable...