[en] The possible slowing down of cosmic acceleration was widely studied. However, judgment on this effect in different dark energy parameterizations was very ambiguous. Moreover, the reason of generating these uncertainties was still unknown. In the present paper, we analyze the derivative of deceleration parameter $q^{\prime }(z)$ using the Gaussian processes. This model-independent reconstruction suggests that no slowing down of acceleration is presented within 95% C.L. from the Union2.1 and JLA supernova data. However, $q^{\prime }(z)$ from the observational $H(z)$ data is a little smaller than zero at 95% C.L., which indicates that future $H(z)$ data may have a potential to test this effect. From the evolution of $q^{\prime }(z)$, we present an interesting constraint on the dark energy and observational data. The physical constraint clearly solves the problem of why some dark energy models cannot produce this effect in previous work. Comparison between the constraint and observational data also shows that most of current data are not in the allowed regions. This implies a reason of why current data cannot convincingly measure this effect.