bug#61962: 30.0.50; New trouble with symbols with positions

[Mattias Engdegård]

12 mars 2023 kl. 21.42 skrev Alan Mackenzie <acm@muc.de>:

> I'm not at all happy with the changes you've made.

Don't worry, Alan, we're writing software, not carving granite. Anything can be 
changed, and I'm not out to get you!
There seems to be some misunderstanding on both sides so let's clear that up.

It was far from obvious that your making `safe-copy-tree` a user-visible 
utility arose from a particular need rather than just being easy to do -- it 
wouldn't be the first time.

If you really have a good reason to add safe-copy-tree then I'm certainly not 
against it, but you haven't made much of a case for it. It seems to have made 
its way in on the slip-stream of a bug fix. I found no justification for it, so 
please forgive me for assuming there wasn't any.

The status quo is something everybody can agree is an improvement: a bug has 
been fixed and that's it. Now if you want Emacs to come with a standard 
circular tree copy function then it needs to go in on its own merits: please 
argue why and how. Concrete examples would be useful.

If added, your `safe-copy-tree` (wouldn't `copy-graph` be a more accurate 
name?) should not be forced to share the implementation of what is now 
`bytecomp--copy-tree`; that would just introduce unnecessary compromises for 
either use.

> would you please identify exactly what the bugs were.

I only know that some of my (probably insufficient) tests failed for the old 
implementation, for example ((a . #1=(b)) #1#) -- I didn't dig deeper.