Why is the y_offset negated?

[CryZe]

I've been trying to figure out why my Arabic rendering is wrong, and it turns out that the y_offset of a glyph needs to be subtracted from its y coordinate, as opposed to x_offset which always needs to be added. This fixes my code, but turns out cosmic-text does it exactly like this internally too. This is very confusing, seemingly not documented and possibly a bug (that cancels itself out internally):

cosmic-text/src/layout.rs

Lines 77 to 78 in 6aadfad

	(self.x + x_offset) * scale + offset.0,
	libm::truncf((self.y - y_offset) * scale + offset.1), // Hinting in Y axis

One way I could explain this to myself is that the fonts use a coordinate system where Y Up points in the opposite direction. If this however is the case, then y_advance is not properly subtracted throughout the codebase, such as here:

cosmic-text/src/shape.rs

Line 1279 in 6aadfad

y += y_advance;

and here:

cosmic-text/src/shape.rs

Line 526 in 6aadfad

y_advance += glyph.y_advance;

[jackpot51]

Don't have a good answer other than the coordinates that were coming from fonts were flipped. If we flip it now, it would be a breaking change.

[CryZe]

I'm guessing it's indeed a difference in Y Up (fonts) and Y Down (images). I'm fairly convinced a purely vertical font / script, where y_advance would be non-zero, would probably be messed up, cause there's no reason to believe that y_advance would use a different coordinate system than y_offset when they are emitted from the underlying rustybuzz right next to each other. But then again, a vertical font is probably broken in many other ways in terms of text layout anyway.