bug#55395: What does (1 2 3 . #2) mean?

[Stefan Monnier]

>> The main goal is to avoid inf-looping and the #NNN chosen is
>> somewhat arbitrary.
> It seems to have sort-of worked before and was broken, inadvertently and
> with the best intentions, by a later change.
>
> Additionally the #N value appears to be correct for other object types. For 
> example,
>
>   [a [b #1=[[c #1# d]]]]
>
> is printed as
>
>   [a [b [[c #2 d]]]]
>
> which is consistent with the manual.

I do have the impression that it used to be "correct", but I can't
remember of a single time where I actually made use of that NNN.

And even if you know what it means and it works correctly, it's pretty
hard for a normal human to correctly count the depth starting from
the root.

I suspect if we want to "do better", printing "the beginning" of the
object to which we're looping (and saying explicitly that there's
a cycle) will be a lot more useful to the average user.
[ Or we could even print "the period" rather than "the beginning".  ]

>> I'm personally more bothered by the fact that those #NNN use exactly the
>> same syntax as used with `print-circle` yet they don't have the
>> same semantics.
> The syntax isn't exactly the same (#N vs #N= and #N#) but annoyingly close.

Indeed, it's not as bad as I remembered.

> For that matter I would have preferred numbering the other way, starting at
> the bottom going up, like de Bruijn indices.

I love&hate de Bruijn indices as much as the next guy, but they're not
human-friendly (even more than the "like de Bruijn levels" we currently
use).


        Stefan