> Is this is vanilla emacs, or are you using some icomplete-mode or
> fido-mode?
> I.e. can you post the entire Emacs -Q recipe?

It is really simple to reproduce.

0. Run ‘emacs -Q’ in a directory with two files: “1” and “foo1”.
1. Enable fido-mode.
2. C-x C-f 1 RET.

Emacs will open “foo1” even though I would expect it to open “1” as that
is an exact match.  But, I really think this is a pointless discussion.

The issue *is not caused by fido-mode* but by the mechanism of substring
(and therefore, flex) completion.  You can also trigger it without
fido-mode by invoking minibuffer-force-complete-and-exit.

    (completion-flex-all-completions "1" '("1" "11" "1122") nil 1)
    (completion-substring-all-completions "1" '("1" "11" "1122") nil 1)

Both return completely the nonsensical result of

    (#("1"
       0 1 (face completions-common-part completion-score 0.0))
     #("11"
       0 1 (face completions-common-part completion-score 0.0)
       1 2 (face completions-first-difference))
     #("1122"
       0 1 (face completions-common-part completion-score 0.0)
       1 2 (face completions-first-difference))
     . 0)

(Why are all the completion-score values 0?)  Applying the attached
patch changes the result to

    (#("1"
       0 1 (face completions-common-part completion-score 1.0))
     #("11"
       0 1 (face completions-common-part completion-score 0.5)
       1 2 (face completions-first-difference))
     #("1122"
       0 1 (face completions-common-part completion-score 0.25)
       1 2 (face completions-first-difference))
     . 0)

which I hope you would agree makes a lot more sense.

> M-j instead of RET in these one-letter completion cases. Or even C-u
> M-j, if that doesn't work.

Sure, but my muscle memory opposes that.

Best regards,
Dario