bug#63103: 30.0.50; nconc compiler optimization breaks user packages

[Daniel Mendler <mail@daniel-mendler.de>]

On 4/27/23 11:47, Mattias Engdegård wrote:
> Although the utility of accepting dotted lists is very slight and requiring proper list for all but the last argument would have made it easier for us, this obviously needs to be a well-reasoned change if made at all. I'll remove the troublesome compiler transform right away.

Thanks for fixing the issue! I agree that improper lists are of limited
utility. It wasn't the best idea to use them for the
`completion-all-completions' API since this makes manipulation of the
result for API users less convenient from my experience.

>> Due to symmetry reasons it is not far-fetched to assume that when
>> `nconc' can be used to turn a proper list into an improper list, that
>> the inverse works too.
> 
> I don't think that's a valid assumption -- when the documentation says that arguments are lists, then they should be proper lists. The behaviour for improper lists needs to be specified explicitly, in particular in this case when the function actually overwrites arbitrary information in the input, not just a terminating nil.

Well, my assumption didn't completely stand on its own. Given that the
last argument can be a non-list suggests that the last argument is not
inspected or mutated, but just overwrites the cdr of the second to last
argument. This then implies that the second to last argument could also
be an improper list. Given that CL and Elisp have behaved like this for
a long time, it seems better to preserve this property. I think it is
kind of nice that `nconc' can be used as a tool to turn a proper list
into an improper list and vice versa. It may be good to document this in
the `nconc' docstring and the Elisp manual.

Daniel