Showing an FIR filter is NOT generalized linear phase: 'the angle changes' is not a valid reason
To show h[n] is not GLP (H = R(w)e^{j(alpha - M w)}, R real), factor out e^{-jMw} with M=(N-1)/2 and look at the remainder, e.g. {-1,3,1} gives 3 - 2j sin w. The tempting one-line reason 'its angle changes with w' is wrong as stated: a GLP response's angle also changes, by jumps of pi wherever R changes sign. The correct reason is that the angle varies continuously (not just by pi jumps), i.e. the remainder is not a real function times a fixed e^{j alpha}. Equivalently, the ratio Im/Re is not constant. When a style or brevity pass shortens these verdicts, keep the 'not just by jumps of pi' clause. Related convention trap: under the strict definition, linear phase (H = |H|e^{-jMw}) needs alpha = 0 AND R(w) >= 0 for all w. So symmetric filters like {2,1,2} (R = 1+4cos w, R(pi) = -3) or {2,1,1,2} are GLP but not linear phase. Check R for sign changes rather than concluding 'symmetric, so linear phase'. In a GLP phase plot, solid vertical segments of height pi are sign changes of R (zeros of H on the unit circle, so H there = 0). Dashed 2pi steps are only wrap-arounds.