Messages in this thread | | | Date | Mon, 8 Feb 1999 00:17:38 -0500 | Subject | Re: Kernel interface changes (was Re: cdrecord problems on | From | tytso@mit ... |
| |
From: hpa@transmeta.com (H. Peter Anvin) Date: 6 Feb 1999 01:19:58 GMT
Let me do a translation of what I think everyone is saying here:
* The Linux module interface is designed to minimize overhead. That means raw linkage into the kernel, which means when the kernel changes, so does the module interface.
This isn't true. There are plenty of changes which don't require changing the module interface. Changing the module interface happens when (a) function signatures change, and (b) structure fields or sizes change. During a stable kernel release, the times when such changes are absolutely necessary should be relatively few in number.
During the 2.0 kernel series, there have been a few changes where folks have disputed whether or not they really needed to happen during the 2.0 kernel series, or whether they could have waited until development series to make such changes. (For example, when structure elements were re-ordered in the name of improving cache alignment. Did that really need to happen in the stable kernel series?)
* A stable ABI requires some kind of layering to isolate it from changes to the kernel code (otherwise the kernel would be effectively barred from any improvements.)
Again, it's important to note that folks are asking for stabilty in the stable kernel series. We can still make lots and lots of improvement in the development series of the kernel. The argument that a request for stability in the stable kernels means that we can't make any more improvements in the Linux is very much overstating the case.
The problem I am having is that it's hard to convince MIT developers to do things like develop an layering API, when no such layering interface is necessary in NetBSD; NetBSD simply uses a little bit more care in reserving such changes for their "current" branch of the tree, and keeping their "stable" branch stable. My perception is that they don't do it just because it makes life easier for AFS, as much as they are simply more disciplined in their release engineering.
- Ted
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |