I realised this approach is wrong.

I must do the quaternion sorting in otherway, probably using a separate
tree for each component (then what?).


Former-commit-id: 6de715751ddd41f6ffc72b274cadf58a765b3edd

qpms/pointgroups.c

@@ -7,7 +7,9 @@
   if ((a) > (b)) return 1;\
 }
 
+// THIS IS NUMERICALLY UNSTABLE! FIXME!!!!!!
 int qpms_pg_irot3_cmp(const qpms_irot3_t *p1, const qpms_irot3_t *p2) {
+  QPMS_WARN("NUMERICALLY UNSTABLE! DON'T USE ME!")
   PAIRCMP(p1->det, p2->det);
   const qpms_quat_t r1 = qpms_quat_standardise(p1->rot), r2 = qpms_quat_standardise(p2->rot);
   PAIRCMP(creal(r1.a), creal(r2.a));

qpms/quaternions.h

@@ -99,6 +99,8 @@ static inline bool qpms_quat_isclose(const qpms_quat_t p, const qpms_quat_t q, d
 
 /// "Standardises" a quaternion to have the largest component "positive".
 /**
+ * FIXME
+ * NUMERICALLY UNSTABLE. DON'T USE
  * This is to remove the ambiguity stemming from the double cover of SO(3).
  */
 static inline qpms_quat_t qpms_quat_standardise(qpms_quat_t p) {