On 12/11/05, Greg Schafer <[EMAIL PROTECTED]> wrote: > Jeremy Huntwork wrote: > > > I just wanted to report on the status of the alphabetical branch as it > > currently stands. For all intents and purposes, I believe it produces a > > stable environment. I have built many, many packages on top of it and > > it's working wonderfully.
> IMHO, a very important aspect of these proposed changes is to check > whether Ch 6 can reproduce itself reliably. There is only one way to > properly measure reproducibility and that is to do some binary diffing a > la ICA (Iterative Comparison Analysis). My techniques (detailed in the > gsbuild scripts) are not perfect but they have certainly been good enough > over the years to detect quite a few bugs. On this note, I will say that I attempted to do the ICA on the lfs-alpha branch. Unfortunately, I borked it being my first time, so not much conclusive came out. I've been meaning to write you back about it, but I didn't have too much meaningful info. Anyway, here goes. I built the thing word for word out of the branch in Jeremy's home directory. Built fine. I only did the tests on the Ch. 6 toolchain. glibc and gcc had a few failures. Some were known, like the gcc mudflap failures. I'll post them soon, but I don't have access right now. After this, I built a few BLFS packages (dhcp, links, net-tools, openssl, openssh) and booted. Ran fine. At this point, I set up the ICA build. By this I mean I would completely rebuild Ch. 6 after copying the contents of mostly everything except /dev, /sys, /proc, /tmp and /home I think. I actually just implemented Greg's do_ica_prep function manually. Problem here is that to do a good diff, I shouldn't have built the BLFS packages unless I was going to build them again, and I shouldn't have booted the new system (particularly running a different kernel). For instance, one stupid thing was that net-tools overwrote the coreutils hostname binary, then I rebuilt and didn't suppress coreutils from installing hostname again. Stupid, stupid, stupid. Well, one thing that happened the second time was that the some of the glibc test failures were gone. Makes me think that Ch. 5 isn't up to snuff. So, after the binary diff (using Greg's do_ica_work function manually), there were tons of differences. Unfortunately, I can't be certain that they were different because: 1) My environment had changed 2) The alphabetical changes had caused issues 3) The current LFS build is flawed So I never posted anything. I plan to try this again, BUT on the current LFS. It seems that no one's done the purity test on LFS in a long time. At least, I'd like to know what the situation is before making wholesale changes and trying to figure out what the new differences are. Jeremy, earlier you said that the test would be to build the whole lfs-alpha, then rebuild the whole lfs-alpha. I thought about this, and I think this is one worthy test. However, it only really highlights the ability of the Ch. 5 temporary toolchain to separate from the host. I know Ryan Oliver has said many times that he's used a busted up RedHat 6 machine he has to build LFS from. This is only one data point, but it seems that the bootstrapping ability of LFS is relatively good and you should be able to build LFS Ch.5 and Ch.6 from most sane toolchains. To find out whether the final system (Ch. 6) is not subtly broken goes beyond rebuilding LFS from the floor or building packages in BLFS. Passing testsuites is certainly a good indicator and the ICA also seems to be an excellent way to test the final system. I'm not saying this to rant against what you're doing. I think testing the build order of the packages is good if what you're doing is finding out whether it builds a better system or not. But I'd like to see a whole lot more testing before major changes to the build order are done. This order was known at one time to rebuild itself byte-for-byte. We should first find out whether this is the case before changing it. That's all my opinion. If it's deemed that the alphabetical order is the way the group wants to go and that enough testing has been done, then I won't say another word. Anyway, I applaud your effort on this even though I'm not behind it right now. -- Dan -- http://linuxfromscratch.org/mailman/listinfo/lfs-dev FAQ: http://www.linuxfromscratch.org/faq/ Unsubscribe: See the above information page
