The birds mouth on a rafter bearing, on a wall top plate, is not a notch in this discussion because of the way the stress act on the member at the point of what obviously looks like a notch to all of us. The eave overhand puts the member in compression at the reentrant corner of the notch, that is the effect of the overhang, a cantilever moment, and the shear stresses are fairly low. Thus, this notch does not have a tendency to split with (along) the grain, also the tension edge is on the top of the member at this location. The bearing surface of the birds mouth is supported by the top wall plate, in bearing or compression, so its likelihood of splitting along the grain is minimized. The birds mouth is o.k. as long as the horiz. seat cut is not longer than the width of the 2x4 or 2x6 wall top plate which supports it.
When the rafter has a long horiz. seat cut, so the tension edge of the rafter substantially overhangs the inter edge of the wall top plate, then the tension edge of the rafter is not supported and can split along the grain. If you extended the tension edge of the rafter to the wall, it might actually meet the wall below the wall top plates. This forms the notch DPR and I are talking about, and it is very likely more than 25% of the member depth, measured perpendicular to the lower edge of the rafter. The reentrant point of this notch is right at the upper/inner corner of the top plates, even if the seat cut is just a long horiz. cut. It is akin to a 2" notch at the bearing end of a 2x8 floor joist.
This type of notch is dangerous for several reasons. It is in the tension edge of the spanning member which is inherently dangerous, even though the bending stresses are very low at this location. The bigger problem at the member bearing has to do with shear stress, which is max. at the bearing or member reaction. Wood is notorious for splitting with (along) the grain of the member. Wood is actually quite strong w.r.t. a shear stress (shearing force, vert. reaction force) which tries to cleave it across the grain. But, at this notch point the vert. reaction force also produces what we call horiz. shear, equal in magnitude to the vert. shear stress, but acting along the grain, which tends to split the member with the grain. The horiz. shear stress is the killer at these notches, it is happening in a region of tensile stresses; and is maximized because it is acting on a smaller member at the bearing due to the notch. In my example above the 2" notch has effectively turned the 2x8 into a 2x6 which also means the shear stress effectively increases because of this loss in member area. Thus a shear stress of 135psi on the 2x8 becomes 180psi on the 2x6, a 33% shear stress increase, right at the notch.
On the other hand, the notch in the tension edge of a beam in the middle of its span is a killer because it interrupts the uniform (smooth) flow of the tensile stresses where they are at their max. It also reduces the depth of the member or its stiffness, and causes a reentrant corner which is prone to splitting along the grain. Although, in this area the shear stress is usually fairly low. So the notch at mid span is controlled by one set of stresses and conditions, while the notch at the beam bearing is controlled by another set of stresses and conditions.