bug#49776: 28.0.50; Incorrect username completion results reported by `read-file-name-interal`

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

On 7/30/21 3:12 PM, Lars Ingebrigtsen wrote:
> Lars Ingebrigtsen <larsi@gnus.org> writes:
> 
>> (all-completions "~/.config/~notexist" #'read-file-name-internal)
>> => nil
> (read-file-name-internal "~/.config/~notexist" nil t)
> => nil

Yes, this is what I expect. I see the same.

> (read-file-name-internal "~/.config/~man" nil t)
> => (#("/.config/~man/" 0 13 (face completions-common-part)))

I see this too and it may be correct behavior. But then I wonder why the
completion boundaries are placed as they are.

> and note that this is an internal function not to be used by users.

I am not looking at this as a user. I am investigating odd completion
results in my Vertico completion UI (on GNU ELPA), when completing file
names. If you type "~/etc/~man", you see these odd candidates
"/etc/~man", where "/etc/" is actually part of the shadowed path. So to
say it differently, I don't expect the shadowed path to take part in the
completion.

Maybe Stefan can chime in here and explain some of the background. I've
corresponded with him before a few times regarding how the completion
machinery works internally.

Daniel