Personally, I probably wouldn’t bother looking into assoc_with_index, the usage makes pretty clear what it does. I’m also not saying that I would necessarily rewrite the code like this, I just wanted to show that the reduce was causing the most complexity here, in my opinion.
I guess my issue with the original code is that you need to thread the socket through all the layers in order to use it in the innermost reduce. That complicates it a lot for me, because the creation of the names and modification of the socket are mixed together. So maybe a version with a few flat_maps that generate the values first and then a reduce - basically my for example, but without the for - would be a pretty good solution as well.
