jhalfs experimental and ICA
dbn.lists at gmail.com
Thu Mar 9 08:01:58 PST 2006
On 3/8/06, George Boudreau <georgeb at linuxfromscratch.org> wrote:
> Inclusion of ICA in jhalfs-X is going to take some thought as ICA
> requiress 3 builds of the chroot phase for comparison. Do we call the
> main Makefile multiple times with from a script, a stand alone app?
I think it's important to state this now. ICA does not depend on running
the build 3 times. It is a comparison of any number of recursive builds.
3 is usually picked for this reason: if you see a difference between the
first and second build, you can see if it's persistent to the third. The
libstdc++ stuff always differs because there is something about adding
random bits to the binaries that Greg found out. 3 builds allows you to
confirm this. I know Ken Moffat has occasionally run 4 or 5 builds.
When I started doing ICA in my scripts I used the same line of thinking.
if [ "$RUN_ICA" = Y ]; then loop_build_3_times; fi. However, I believe
it's much better to handle a generic situation where you want to loop the
build. Some people actually build Ch. 6 twice to get that squeaky clean
feeling regardless of doing any comparison or not.
In that case, I think it's best if ITERATIONS=x and RUN_ICA=x are
separate. The ICA stuff can be easily tacked on to the end of the
Makefile. I think this is a better situation because it will allow the
design to focus on the trickiest part. ICA is a simple tool that, IMO,
should be separate. My experience with jhalfs is almost nil, but to me,
it seems that the easiest way would be to call the main Makefile multiple
times. You guys are the experts, though.
More information about the alfs-discuss