Re: Help with recursive destructive function

[Eric Abrahamsen]

On 05/15/18 01:16 AM, Michael Heerdegen wrote:
> Eric Abrahamsen <address@hidden> writes:
>
>> For backwards compatibility, we need to be able to handle lists that are
>> quoted, or that start with the symbol `list'. This will sound familiar
>> to you... In the non-destructive version, it was easy enough just to
>> return (cdr thing) instead of thing.
>
> What will this code do when the saved list was really quoted, or was a
> list with the symbol list as first element?  Wouldn't the backward
> compatible version cause errors?

Yup. But no more errors than it already would. This will also be an
issue when I try to unambiguously identify lists-that-should-be-objects.
The only surefire way around this that I can think of is writing an
eieio-persistent version number into the files.

>> In the destructive version, this means that `handle-refs' would need to
>> first edit and *then* traverse the cons, which is not what it's set up
>> to do, obviously. I could probably cheat and move the backward
>> compatibility into some other part of `deep-edit' itself, but I was
>> trying to avoid that because that function could be useful elsewhere as
>> part of the general library.
>
> My implementation so far doesn't perfectly fit that case (as you
> noticed).
>
> Do you expect something occurring like '1 or (list) - i.e. cases where
> the remaining thing is an atom (like 1 and nil above)?

Yes -- there's a check in the current code precisely for the (list)
case.

> If not, I would just tweak `handle-refs' so that it calls the modify
> function _and_ additionally pushes the replacement to the stack when
> some condition is met.  You could make it so that `edit-func' returns a
> special value for this case, like in
>
> #+begin_src emacs-lisp
> (defun edit-func (proposed-value)
>     (cond ((and (consp proposed-value)
>                 (eq (car proposed-value) 'list))
>            (list t #'cdr))
>           ((and (consp proposed-value)
>                 (eq (car proposed-value) 'quote))
>            (list t #'cadr))
>           ((stringp proposed-value)
>            #'upcase)
>           (t nil)))
> #+end_src
>
> where (list t FUN) would mean that we want to modify the value with FUN
> and additionally push the resulting (remaining) tree to the stack.  Not
> much magic here...

I had been treating it as magic :) I'll try to loosen up a bit and
experiment with something that might also handle the (list) case.