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

[Stefan Monnier]

Eli Zaretskii [2022-05-02 19:53:08] wrote:
> Why is it justified in this case?  How is this case different from any
> other case?

I don't think it's different from any other.  It's generally considered
a bug when a function uses things like `sort`, `nconc, or `nreverse on
a list it received as argument (unless the docstring explicitly
mentions that the arg may be modified destructively).  When a user
reports this behavior as unexpected the answer is almost always to fix
the funciton so it doesn't have this side-effect any more.

The same usually holds for text properties on strings.
The fact that `completion--replace` is a function used in the completion
machinery was largely irrelevant to the decision of how to fix
the problem.  Its docstring says:

      "Replace the buffer text between BEG and END with NEWTEXT.
    Moves point to the end of the new text."

and nowhere does it warn that NEWTEXT might be modified along the way,
so removing properties on that string was just a plain and simple bug.


        Stefan