Hi Gregory:

Please look at the attached patch where I use a modified version of your
line approach instead of pixels. Because I can't reproduce most of the
issues you mention.

This change is not permanent, but it is the only substantial difference
with your and my code; so lets try that and if this fixes your issue I
will check the pixels approach.

Thanks in advance,
Ergus

On Sat, Sep 19, 2020 at 10:30:13AM +0000, Gregory Heytings wrote:
>
>
>It seems to me that it does not work as well as my code, alas.  A few 
>points:
>
>- with C-x C-f the default completion, between angle brackets in the 
>original icomplete code, is not shown anymore, unless there is a 
>single candidate (for example in a directory with "foo", "bar1", 
>"bar2", "f" will give "f[oo]" but "b" will give "b[]"); in other 
>words, the "[]" after the point usually serves no purpose
>
>- likewise, with M-x there is a "()" after the point that usually 
>serves no purpose
>
>- if I type "M-x icomplete-m", I see the following:
>
>M-x icomplete-m(ode)
>|
>
>where "|" represents the cursor (on the next line!)
>
>- when the current line after the prompt is too long the prompt disappears
>
>- when completing a unique completion with TAB the point is moved to 
>the next line
>
>Gregory