On Mon, 9 Jan 2006, Dan Nicholson wrote:
I'm pretty sure the perl, vim and nscd differences are time stamp
related. Ken, farce does quite a bit more analysis than Greg's ICA
functions. Would you mind looking at these results and commenting on
differences between the ICA report and the farce report?
I'm a bit short of time at the moment. Bug me in two days if I haven't
responded, please.
Not quite sure what you mean by "timestamp" - for files where file's
description contains ' shared ' (executable programs, shared libraries),
farce first runs 'strip --strip-all' on copies. If they then match,
they are accepted 'after --strip-all'. If they still differ, the
stripped versions are subjected to the regex processing. If they now
match, they are accepted 'after --strip-all and processing'.
At this point, they definitely differ after embedded date/time strings
have been converted to tokens. The only exceptions are incorrect
regexps and new date/time formats (three for the groff-1.18.1 manpages
and postscript files!). I've added a test to disassemble fresh copies
of the files, to see if the code sections are the same. This is
something of a comfort blanket - the files differ, I don't know why, but
at least they have the same code so they might not be too bad.
In practice, further testing shows that these files with the same code
do not show up when I compare the second to third builds. Also, some
files move from 'acceptable' to 'identical'. But, we now know that the
toolchain is not being built correctly, I'm hoping that fixing it will
address some of this.
At the moment, perl is proving a particular problem for me - yesterday,
my first (regular) build for my fourth cycle had different code from the
subsequent builds. All the previous cycles, and yesterday's in-place
rebuilds, compared acceptably against other builds. Today, I tried two
regular builds, in one reverting the /etc/hosts change for perl, and in
the other reinstating the removal of arpd (I'd accidentally left the
iproute2 sed in for the earlier cycles). Neither made any difference,
perl built acceptably both times, but these were the only apparent
changes in my buildscripts between the third and fourth cycles! So, at
the moment I have an aberrant build of perl in the first run of the
fourth cycle, and at the moment I can't replicate it.
Will try again later, using the script as it was, but keeping the perl
source around.
Still working on a couple nits concerning e2fsprogs and the placement
of gettext. farce shows "differ but same code when disassembled."
Ken, could you comment on this new feature of farce?
See above. Less bad than different code, but probably still wrong.
Right now I'm running a build with the DIY toolchain adjustment in
place that forces the use of the correct glibc for gcc and binutils.
I want to see if it affects anything.
Should be interesting.
Ken
--
das eine Mal als Trag?die, das andere Mal als Farce
--
http://linuxfromscratch.org/mailman/listinfo/lfs-dev
FAQ: http://www.linuxfromscratch.org/faq/
Unsubscribe: See the above information page
