graphene_matrix_init_look_at() inverted and upside down

[sta-c0000]

graphene v1.10.8

Hello and thank you for graphene.
I was trying it out this weekend. It appears graphene_matrix_init_look_at() actually produces an inverted view matrix, ok, but with a negated eye translation vector, which doesn't make sense to me. For an inverted view matrix I would expect the eye translation vector to not be negated.

here: /include/graphene-simd4x4f.h

m->w = graphene_simd4f_init (-eye_v[0], -eye_v[1], -eye_v[2], 1.f);

Unless I'm missing something? I wonder if it's a quirk related to GTK's 2D coords Y+ pointing down? But here in 3D all axes flip upside down; the look_at source states coords: X+ right, Y+ up, Z+ towards us (same as OpenGL). And normally we project to NDC first then flip Y if needed while converting to screen space.

To illustrate, see NOTEs in this barebones graphene 3D to 2D projection example: graphene-3d-svg.c

Seems it's been like that for a long time, I guess it won't change. I'm just noting it here for others that may bang their head. Don't know how used it is. Appears gthree has a single call to graphene_matrix_init_look_at, converting the resulting matrix to a quaternion, effectively throwing away the translation.

Thank you for your time.

p.s.: I started by prototyping in Python, but it segmentation faults whenever graphene returns an array with more than 4 elements. I understand not a graphene issue, but an introspection one (#11) . Confirmed with valgrind --tool=memcheck

import gi
gi.require_version("Graphene", "1.0")
from gi.repository import Graphene

rect = Graphene.Rect.alloc().init(-0.5,-0.5,1,1)
verts = rect.get_vertices() # malloc issue!

p_max = Graphene.Point3D.alloc().init(0.5,0.5,0.5)
p_min = p_max.scale(-1)
box = Graphene.Box.alloc().init(p_max, p_min)
verts = box.get_vertices() # malloc issue!

m = Graphene.Matrix.alloc().init_identity()
mat = m.to_float() # malloc issue!