Re: kernel-2.4.0-test11 crashed again; this time i send you the Oops-message

[Peter Samuelson <peter@cadcamlab.org>]

[Albert D. Cahalan]
> The infamous LINK_FIRST infrastructure was sort of half-way done.

I disagree: it could handle all cases I could see that we might
reasonably care about.  I challenge anyone to come up with a
non-pathological case that could not be taken care of with a single
LINK_FIRST and/or a single LINK_LAST.

The worst I can think of is something like "all PCI drivers must come
before all ISA drivers" which would require listing all of one set or
the other.  But when you see a case like that, it often means "this
directory really needs to be split", because you have two different
classes of things in a single directory.

> It would be best to cause drivers with an unspecified link order
> to move around a bit, so that errors may be discovered more quickly.

That was the plan -- in 2.5.  (The 2.4 version did not disturb any
order at all, unless you explicitly put a LINK_FIRST declaration in the
individual makefile.)  Now that LINK_FIRST is officially dead, none of
this will probably happen at all.

> LINK_FIRST is pretty coarse. One would want a topological sort, or at
> least LINK_0 through LINK_9 _without_ anything else.

Too complex, no easy upgrade path (read: too different from status
quo), very little benefit over LINK_FIRST + LINK_LAST.  For the
topological sort in particular I'm interestested in how it's even
possible to do in a non-intrusive and maintainable way.

Peter
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
Please read the FAQ at http://www.tux.org/lkml/