Eric W. Biederman wrote: >>> >> With "architectural" I mean "guaranteed to be stable" (as opposed to >> "incidental"). Sorry for the confusion. > > Ok. Then largely we are in agreement. To implement that the rule is simple. > If it isn't CTL_UNNUMBERED and the number is in Linus's tree, it is > our responsibility to never change the meaning of that number. > > If a new sysctl entry is introduced it should be CTL_UNNUMBERED until > it reaches Linus's tree to avoid conflicts. > > There is simply no point in having any kind of support for numbers > whose meanings can change. > > Which is why I removed the few cases of binary number duplication I > found. > Agreed. *Furthermore*, if the number isn't in <linux/sysctl.h> it shouldn't exist anywhere else, either. -hpa