bug#22320: Overlays with an 'invisible property break stacking of overlay faces

[Eli Zaretskii]

> Cc: 22320@debbugs.gnu.org
> From: Clément Pit--Claudel <clement.pitclaudel@live.com>
> Date: Thu, 7 Jan 2016 19:27:30 -0500
> 
> 1. The overlay does not intersect with the invisible area. No problems here.
> 2. The overlay covers a few characters before the invisible area and the 
> first few characters of the invisible area, but not the last ones.

In this case, Emacs doesn't care where the overlay ends, that buffer
position is skipped without checking overlays.

> 3. The overlay covers a few characters after the invisible area and the last 
> few characters of the invisible area, but not the first ones.

Likewise.

> 4. The overlay is a subset of the invisible area.

If it starts on the first invisible character, Emacs will take note;
otherwise it won't.

> 5. The overlay is a superset of the invisible area.

This is the case with all your use cases.

> The current solution doesn't think in these terms; instead, it ignores all 
> overlays for the ellipses, unless the first hidden character has the same 
> face as the preceding one.
> 
> Always applying the face of the first character to the ellipsis means 
> applying to the ellipsis the faces of overlays in categories 2, 5, and 4 if 
> they cover the first character. You pointed out that the special case of 4 is 
> confusing.
> 
> Another option (the one I was trying to explain above) is to apply to the 
> ellipsis only the faces of overlays in category 5. I feel that this should be 
> rather uncontroversial (don't you think?).

Well, now that I've committed what I believe is a better solution, I
think we can put these issues to rest.

> IIUC, the current implementation computes faces without taking invisibility 
> into account for the first character before and after the left boundary of 
> the invisible section, and then compares them. At this point it's too late to 
> merge certain faces but not others.

Yes.  More accurately, the display engine always considers faces
before it considers invisibility, so it always sees the face of the
first invisible character.