Re: master ff3f17ca3cd: choose-completion: Retain the suffix after completion boundary

[Dmitry Gutov <dmitry@gutov.dev>]

On 18/05/2024 09:55, Eshel Yaron wrote:
> Dmitry Gutov <dmitry@gutov.dev> writes:
> 
>> Hi Eshel,
>>
>> On 17/05/2024 11:56, Eshel Yaron wrote:
> [...]
>>> I'm afraid there might be another regression, here's what I see:
>>
>> Thanks for keeping an eye out.
>>
>>> 1. emacs -Q
>>> 2. C-x C-f
>>> 3. C-a C-k to clear the minibuffer
>>> 4. Insert the path an Emacs checkout, e.g. "~/checkouts/emacs/"
>>> 5. Further insert "l/mi"
>>> 6. Hit ? to pop the completions list, which includes
>>>      "lisp/minibuffer.el" and "leim/MISC-DIC/", among others.
>>
>> FWIW, I don't see "leim/MISC-DIC" among completions, even if I first
>> evaluate (setq completion-ignore-case t).
> 
> Weird, perhaps you have read-file-name-completion-ignore-case set to nil
> and that takes effect rather than plain completion-ignore-case?

Indeed, it's nil. I started with 'emacs -Q' to avoid my settings' 
interference, and assumed your scenario used it as well. Anyway...

>> But the crux of the problem reproduces fine.
> 
> Great.
> 
>>> 7. Lean on M-<down> to cycle between completion candidates.
>>> After the last fix, I see M-<down> accumulate text after point in
>>> the
>>> minibuffer instead of replacing the part it previously inserted.  Do you
>>> see that as well?
>>
>> Please take the attached patch for a spin.
> 
> Nice, this seems to fix the regression at hand, and so far I'm not
> seeing any other issues.

Great! I've pushed it to master now.

>> It would be great if someone looked into writing integration tests for
>> this somehow. This is brittle stuff, and the variety of scenarios is
>> high enough.
> 
> I agree.  The default completion UI supports various interactions that
> can break in subtle ways, and that's without even configuring stuff.
> I'll try to code a couple of test cases when I have the time.

That would be very welcome, thanks.