When simply introduce a uniform scale to the modelview matrix, this scale should not affect lighting results if GL_NORMALIZE is enabled.
But now the light calculation is not correct if scale the modelview matrix when enable GL_NORMALIZE.
Created attachment 10047 [details]
In test case, a polygon is drawn with lighting parameters set in such a way that most values drop out and the resulting pixel color is a function of the normal vector.
The equation reduces to c = max(0, n dot VP) and (n dot VP) reduces to cos(theta).
Local light sources use the unit length vectors but have a w value of 1.0E-6. Using the w value makes the local light so distant that it is an infinite light for all practical purposes.
Fixed in git.
This bug exists with i965 driver. Reopen to i965 component.
fixed in git.
Mass version move, cvs -> git