Moving* back to /lib

Richard W.M. Jones rjones at
Fri Apr 9 18:35:39 UTC 2010

On Fri, Apr 09, 2010 at 01:42:54PM -0400, Adam Jackson wrote:
> On Fri, 2010-04-09 at 17:54 +0100, Richard W.M. Jones wrote:
> > On Fri, Apr 09, 2010 at 10:26:01AM -0500, Chris Adams wrote:
> > > Depending on fixed paths seems like a bad idea.
> > 
> > It depends on fixed paths because fixed paths are used to build the
> > appliance.  Therefore the dependencies tell us when something isn't
> > going to work at runtime, instead of having the package silently
> > broken by changes such as the one discussed in the OP.
> > 
> > Now you may think that this is a bad way to build an appliance, but no
> > one has come up with any better ideas for that so far.
> I thought I did suggest how to do this better:
> Is there a reason why that won't work?

It's not just libraries, the appliance gets built from many different
types of files.  To do this quickly, in 1/5th of a second, we start
with a list of filenames[0] (wildcards, actually) that we want to pick
up from the host system.  We use a C program[1], that for speed
reasons doesn't call out to any external programs, to generate the
cpio-format initramfs.

It turns out that binaries moving between /s?bin and /usr/s?bin, or
libraries moving, are quite rare events.  In the very few cases where
there are libraries that frequently churn we've added exceptions for
them.  We added an exception for more than a month ago,
and that's been it since.

Patches welcomed if you want to have a go at solving a very
challenging problem better.




Richard Jones, Virtualization Group, Red Hat
virt-top is 'top' for virtual machines.  Tiny program with many
powerful monitoring features, net stats, disk stats, logging, etc.

More information about the devel mailing list