[Aurora-sparc-devel] Install Cd's?

[Dean Anderson]

On Tue, 8 Apr 2008, Tom "spot" Callaway wrote:

> On Tue, 2008-04-08 at 14:34 -0400, Dean Anderson wrote:
> > Otherwise, you have a unique build environment with unique and
> > undocumented software and undistributed sources. Which is what you
> > have now, and is why aurora builds aren't reproducible, yet fedora
> > builds are reproducible.
> 
> Most of the Aurora builds were done years ago, literally. 2.99 is built
> from a chroot on odyssey, my T2000.
> 
> It has no custom packages, no special rpm macros, no magic scripts. I
> use pungi to compose the tree, and I use rpmbuild to build those
> packages.

I don't doubt your good intentions. But I do doubt that odyssey is as
generic as you assert.  Obviously, until just fairly recently, it had
kernel sources from outside of the kernel src.rpm. Otherwise you too
would have had problems building the kernel from the src rpm, just as I
did, and reported recently. (few months ago I think).

If you want to install a general machine and build the aurora from
nothing but source, and post the terminal log of the entire operation,
from install to distibution cd's I'd be more confident.


For example, since you bundle mkisofs 2.01, which can't build sun boot
disks. It can only build solaris-type iso's, with a sunOS partition
table, which in fact, you do construct. However, that deviates from the
ordinary installer instructions, e.g.

http://www.linuxworks.com.au/linux/redhat/redhat-installer-howto.html

So you do have some magic scripts that build your distributions.

In fact, you might consider something similar for aurora.  It would 
probably encourage more people to rebuild from source, which obviously 
picks up a lot of problems.


As too whether you have changed things in the several rpm config files
or not (various rpmmacros files, .rpmrc, etc) , I'll say that I haven't
checked recently.  The last version I was able to successfully check was
1.0.  Since then, there has continuously been something wrong with the
source distribution (2.0, 2.0beta, 2.98, 2.99).  Perhaps this latest is
good, but I can't say yet.  However, I can say it hasn't been good for a
long time, since 1.0 until perhaps current.

Indeed, it used to be the case (2.4 kernels) that boot kernels were
different from installed kernels, so a build script _had_ to build each
with different RPM options (as both claimed to have the same spec file).  
This isn't the case with 2.6 kernels, so perhaps there isn't any special
options anymore.  But I'd need to have a listing of your rpmbuild macros
to compare with a generic install to know for sure, and to trust that
there have been no changes.



As an aside: In fact, if I were changing rpm, I'd make sure it couldn't
have conditionals on source and patch definitions, and would include at
least a signature of the distribution rpmmacros environment, so to flag
rpms whose build environments varied from the distribution.  Then I'd
probably also include the variances (or maybe the complete environment)
in the rpm file so that rebuilders would know what the build environment
was originally.

		--Dean



-- 
Av8 Internet   Prepared to pay a premium for better service?
www.av8.net         faster, more reliable, better service
617 344 9000