Warner Losh wrote: >>giving. I'd really like to see namespaces and name prefixes defined >>that have stable/unstable meaning. I dropped this due to the very >>vocal opposition at the time. > > > Sounds like a good thing to me. If we document this (and the bit that > I sent in my other email) in sysctl(8), or somewhere more appropriate, > then I think we can do this. I propose that we put a single _ in > front of those sysctls that the author knows will go away and that no > code should be written to use them. this would give the unstable > namespace that you want and would be consistant with what _foo() > functions are for. Not sure if that means everything else is > permanant, however. I'm hoping this doesn't result in a bikeshed of > doom. > > Warner Sounds good. I guess what this means is that the entire sysctl namespace should be considered stable except for a set of defined areas. These defined areas will either be specifically designated and documented, or will have a '_' prefixing some element of their node name. Developers still have a responsibility to put proper thought into what they add into the stable namespace, and this space may be subject to change while the tree is marked "-CURRENT". However, once it is marked "-STABLE", it is stable and can only be changed by following appropriate deprecation procedures. For the purposes of documentation, it should be described that there are actually 3 types of nodes. There are stable, statically named nodes that will always exist on a running system. There are unstable, statically named nodes that may or may not exist from one release to another. And there are dynamic nodes that will exist depending on the devices, modules, and drivers present, and that these nodes may be stable or unstable depending on other attributes of the node name. ScottReceived on Tue May 23 2006 - 19:22:27 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:38:56 UTC