俱舍論記 5.192

5.192

「又應一法至立異終不成」者，此即別破異相。又應一法生已未壞名住，住已壞時名滅。理且可然，縱許住、滅也。異於一法進退推徵，理不應有。凡言異者，前後性別，非即是此法可言異此法。故說頌言：異相時法即是前住相時法，異不成。此即進責。若異相時法異前住相時法，即非一法。若住、異別法，有違宗過。此即退徵。是故說一切有部，於一法上立異終不成。

Regarding "Furthermore, it would follow that on a single phenomenon ... establishing alteration is ultimately unfeasible": This specifically refutes the characteristic of alteration. Furthermore, it might follow that when a phenomenon has arisen and is not yet destroyed, it is called "stationariness," and having abided, when it is destroyed, it is called "cessation." In principle this might still be tenable, granting stationariness and cessation. But pushing the inquiry back and forth regarding alteration on a single phenomenon, in principle it should not exist. Whenever one speaks of alteration, there is a difference between prior and subsequent natures; it is not that this very phenomenon can be said to be altered from this phenomenon. Therefore, the verse says: If the phenomenon at the time of the characteristic of alteration is the very phenomenon at the time of the previous characteristic of stationariness, alteration is not established. This is pushing the inquiry forward. If the phenomenon at the time of the characteristic of alteration is different from the phenomenon at the time of the previous characteristic of stationariness, it is not a single phenomenon; and if stationariness and alteration are different phenomena, there is the fault of contradicting your own tenet. This is pressing the inquiry backward. Therefore, for the Sarvāstivāda, establishing alteration on a single phenomenon is ultimately unfeasible.