[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Potentially serious problem with kernel-headers...



>If I understand the problem correctly (and I don't have the time right
>now to consider this carefully, so take this with a grain of salt),

You do not understand the problem correctly.  

(Sorry to be so up front about it :-)

>all you have to do is to make sure that you always install parallel
>versions of kernel-image and kernel-headers.  If you're using
>make-kpkg, you need to (at a minimum), always build and install both
>the headers and image packages:

The problem is that every binary built against a previous version of
the kernel headers (including, potentially, *libc*.  This is major.
Big red warning sirens should be going off here) will now no longer
work with new kernels and vice versa.  Changing the kernel headers and
interfaces breaks backward compatability for everything already built
(and means that 2.0.34 binaries will fail on 2.0.33).  Want to rebuild
everything on your box to upgrade from 2.0.33 to 2.0.34?  Didn't think
so :-) Thou shalt *not* modify kernel interfaces in a minor release
version.  Hell, one should only do it when there's a really damned
good reason, and then *still* not do it.

This is similar to the ctype.h fiasco, where the libc lookup tables
for various ctype operations are in the header (and thus compiled into
the binary, statically linked or not).  These lookup tables changed
endianness between libc 5.3 and 5.4, resulting in all binaries
dynamically linked against 5.3 breaking under 5.4 (if they used
ctype.h).  And, of course, made life more difficult for package
maintainers.  Installing new headers at this point won't help.

Also a caveat: We do not know yet for certain that this is the
problem.  It's simply a likely candidate.  We have established this is
not directly libc related (that is, some bug/change is causing
multiple versions of libc to break, at least 5.3.12, 5.4.43, 5.4.44
and glibc 2.0.7).  Note also that Linux users are complaining of other
major apps breaking between the 2.0.33->2.0.34 change (at least one
shell also segfaults immediately like cdparanoia... zsh?)

Monty


--  
To UNSUBSCRIBE, email to debian-devel-request@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org


Reply to: