Re: [SSI-devel] Re: libc problem in initrd

[Brian J. W. <Bri...@hp...>]

En Chiang Lee wrote:
> I've included the following in the RPMS/base directory of the tarball:
> 	glibc-2.3.2-27.9.7.i386.rpm
> 	glibc-common-2.3.2-27.9.7.i386.rpm
> 	binutils-2.13.90.0.18-9.i386.rpm
> 
> With these included, the install and boot happens fine. binutils is
> included because it is not installed as part of a "server" install of
> RH9, and it is needed by our version of mkinitrd (for the strip
> command).
> 
> For the next release, we could leave these rpms in the RPMS/base
> directory, or add them to the dependencies for the kernel and mkinitrd
> rpms.

Including the newer glibc-* and binutils packages is nice for ease of 
installation, but it bloats up the release tarball from 25MB to 46MB 
(84% increase). That's not so friendly to our users who don't have 
broadband. I'm inclined to do the following instead:

1) Create a supplemental release tarball (like Jai does with Lustre and 
DRBD) that contains these packages under RPMS/base, and has a stripped 
down install script that only installs them, and doesn't call 
ssi-create, install SSI packages or anything extraneous like that. Also 
useful would be a README that explains the purpose of the tarball and 
tells the user to simply run ./install.

2) The download section of our website for this release should tell the 
user to check the versions of these packages. It should provide the 
specific commands (e.g., `rpm -q glibc') and expected version numbers. 
It should also provide the download link for the supplemental package, 
and tell the user to either download and install it or get the most 
recent packages from some other source.

3) README.upgrade should also say the same thing.

4) The kernel-ssi-smp package should depend on the most recent versions 
of these packages. If the user completely fails to notice the previous 
admonitions, at least a failed kernel installation should catch their 
attention.

Regards,

Brian