On 06/13/2014 12:11 AM, Nico Kadel-Garcia wrote:
The core CentOS deverlopers have, in my personal experience, never
been willing to discuss or expose the details of their internal build
environment.
Times, they are a-changin'.
I'm afraid I tried to find out details a few times, such as whether
they use 'mach' or 'mock', and did not get very far.
They use mock, and they published their mock configs a really long time
ago. I was able to get some details for my C5 IA64 rebuild project last
year, and the CentOS core devs were very helpful when I asked them
nicely for the info. I've let the IA64 stuff languish for a few months,
though; I may wait on 5.11 to rebuild those updates. Doing the rebuild
myself was an eye-opener, as many comments made by CentOS core devs in
the 6.0 timeframe are made very clear indeed when you find out just
exactly what was required to get 5.6 out of the door. Rebuilding
manually with mock and getting the dependencies just right was not easy
with 5.6, in my experience.
I don't recall exactly which packages that were affected by this, but
there was a particular library uprev from 5.5 to 5.6 that happened to
impact another package, which wouldn't build correctly with the new
version of that library, and thus was built against the older version
(but there was another package that wouldn't build correctly with the
older version). If there is enough interest I might be able to dig that
information up, but finding the particular build order for the 5.6
update set (not all of the packages for which were issued in-order as
separate updates, again IIRC) took iterative time, most of which was
spent waiting on the machine to do the builds. And there really wasn't
any way to parallelize what I saw with 5.6, but I reserve the right to
be wrong, too. Nor did I know what order would work until a build
completed and passed tests, at which point it was done anyway. It was a
lot like solving a Rubik's Cube; I can give you some basic sequences and
algorithms, but each solution is going to require a different order of
sequences of moves and I won't know what that order is going to be until
it's solved.
Although I had an epiphany of sorts afterwards, but I haven't spent the
cycles to try that idea out on the 5.5->5.6 upgrade. (Short version of
the epiphany: rpm -qp --queryformat "%{BUILDTIME}\n" $source-package
). When the update is issued (or populated to the upstream ftp source
directory) is totally meaningless and unimportant; build time though
does give you at least a possibility of finding out the contents of the
buildroots (barring unpublished packages being used by the builds, which
is always a possibility, and might very well have occurred for the 5.6
package set, and barring hand-built packages with ad-hoc mock configs)
at a given time.
I've not tried to dig into Scientific Linux's build environment.
That's actually a good question, though a separate one. Are the
details of the the Scinetific Linux build system publicly available?
I've not personally gone digging.
IIRC SL6 uses koji (see
http://listserv.fnal.gov/scripts/wa.exe?A2=ind1011&L=SCIENTIFIC-LINUX-DEVEL&P=R531&I=-3
).
Now, for 7, much of the details of those conversations are found on the
CentOS-devel mailing list right now, with patches being accepted from
the community, and two CERN devs being accepted into the CentOS Core
group (to do work on koji, again IIRC; but you can read the thread in
the -devel list yourself). The change to git.centos.org as being the
place where RH is dropping source is a real game-changer, and may make a
koji-driven rebuild much more friendly (koji can rebuild from source
RPM, but it prefers to build out of a repository, not a directory full
of src.rpm's).
These are certainly interesting times.
And, of course, Russ has all sorts of good stuff online for clefos. You
should check it out.
