bug#55205: 28.1.50; completion--replace illegally mutates completion candidates

[Daniel Mendler]

On 5/2/22 18:34, Stefan Monnier wrote:
>> Completion takes text typed by the user and produces strings that the
>> user could possibly mean by typing what he/she typed.  Some part(s) of
>> the candidates can legitimately come from what the user typed, some
>> other part(s) could be invented by the completion machinery more or
>> less out of thin air.  Why should anyone expect this meat-grinder to
>> refrain from destructively modifying any of the involved strings?
> 
> Because the overwhelming majority of strings are never modified.
> It's very unusual to modify a string by side effect (as opposed to
> creating a new string object via `concat`, `substring`, ...).
> This is true in most languages, AFAICT, but it's definitely true in
> ELisp.

One should add that *by definition* of the completion API, mutations of
the supplied candidate strings don't take place. For example you can run
completion on strings from the obarray:

(completing-read "Symbol: " obarray)

It would be quite harmful if the symbol names are destroyed in the
process. I don't see the completion API as a meat-grinder. It is all
quite well-defined.

Daniel