Re: Feature branches review please

[Ergus]

On Thu, Nov 05, 2020 at 10:00:57PM +0000, Gregory Heytings via Emacs 
development discussions. wrote:
> > > This has been discussed at length earlier: it is in practice 
> > > impossible to calculate the height of the minibuffer, and to 
> > > calculate the size of the completion candidates list to insert in 
> > > the minibuffer.  Yet you need to do both to have a correct 
> > > solution with the approach of the branch.
> >
> > With the current display code on `master`, I don't think the 
> > behaviors you refer to can qualify as incorrect.
>
> Which is why I said, in the two previous mails, "not 100% correct" and 
> "not always correct".  I did not think it was necessary to repeat 
> "always" here.
>
> > You can argue that they are less often preferable than some other 
> > choice, but that's a far cry from incorrect, IMO, and then should be 
> > fairly uncommon.  So it's definitely not very high priority and 
> > shouldn't decide whether we install a particular version of the code 
> > right now.
>
> My point is that now that `(setq icomplete-separator "\n")' works (in 
> most but not all cases), there is no need for the specific 
> vertical-icomplete implementation anymore.  What is (or could be) 
> needed is an implementation that is "more correct" (correct in all 
> cases).
I don't totally agree is the same than the branch.

The completion like "compi{compilation" with fido mode is still there,
in the same line and is not intuitive; the ellipsis is not shown and the
{} and [] are still there and hard coded. icomplete-prospects-height is
not respected either because the number of candidates is still
calculated with the window-width and adding the candidates length which
actually makes no sense at all in vertical mode.

The pixel calculation is just part of the modifications in the branch.