What is surprising, and bad, is that the data changes between stable
versions. Unfortunately, if the internals of a structure change there
is no way to correctly handle that for a module if it accesses that
structure. If the module were written to handle this case, it might work,
but the penalty would be huge. Imagine a file system that cannot directly
access an inode.
Any change that changes a _header_ file should be subject to extreme
scrutiny in stable releases. Unless there is no other way to fix a problem,
and the problem is critical, it should not be accepted. I.e. the public
interface should not be allowed to change...
Perhaps it would help to say 2.0.x0 is a stable tested release (i.e. 2.0.10,
2.0.20, 2.0.30 etc.) and 2.0.xy if a stable, somewhat tested release.
> Once again: there's no reason that source code can't be written
> to fit between the AFS module and the various linux kernel
> versions. This would be an easier task if AFS had been engineered
> for this approach, but it's doable even now.
probably not if critical structures have changed...
-gordo