On Tue, 17 Jan 2006, Dan Nicholson wrote:
Seems to me like this whole issue with the .old libraries for readline
should just be eliminated. It's the only package that does this. DIY
has
sed -i.bak '/MV.*old/d' Makefile
Well, when it works, it looks good (update in place, decide you don't
like the new version, uninstall), and it _only_ affects in-place
updates.
I have a theory about where this symlink problem comes from, but until
tomorrow night I won't know if I'm on the right track.
Meanwhile, I've realised that some of the difference between
'identical' and 'allowable' files might be down to those which hard-code
the date-compiled (but not the time) - on the same day, these have a
chance of being identical, on any other day they can at best be
'acceptable' (after converting the date to a token). Obviously, the way
to get more-reproducable figures for how many become identical on a
subsequent build is to only do one build per calendar day (otherwise,
I'll have to explain why even after three builds, the number of
identical files is still increasing).
I think you might be right here. There may be some other oddness with
the date/time stamps. Trying to hunt down e2fsck problems was a real
pain because sometimes the test would return that the two binaries
were identical, sometimes acceptable, and sometimes totally failing.
And I'm 99% positive that I identified the only differences to be
embedded time stamps.
I'm sure that farce's failure to strip static binaries didn't help.
If I had the time, I'd look at the substitutions to try to identify how
many files store date without time (that's complicated by the
possibility of multiple different substitutions in each file), but I've
got a backlog of real-life things I need to attend to.
Ken
--
das eine Mal als Tragödie, das andere Mal als Farce
--
http://linuxfromscratch.org/mailman/listinfo/lfs-dev
FAQ: http://www.linuxfromscratch.org/faq/
Unsubscribe: See the above information page
