Ken Moffat wrote:
> On Fri, Aug 17, 2012 at 12:52:40PM -0500, Bruce Dubbs wrote:
>>
>> Looking at the date code in coreutils, I think the only things that use
>> the zoneinfo data is when TZ is set or /etc/localtime is a copy of a
>> file in the zone DB. I'll be interested in what you find.
>>
> I do wish you hadn't said that ;) We always create /etc/localtime
> (at the end of glibc). Anyway, for my first set of testing I'm
> looking at _all_ the packages which had test failures. The only
> changes are that chapter 6 has been completed, and I've temporarily
> removed /etc/localtime. I _think_ that should probably be based on a
> regular (i.e. posix) file, but it was using a 'right/' file because
> that was the only variant installed.
>
> sed : I've got to revisit this, my results don't make sense. The
> tests use the ru_RU.UTF-8 setting to test manipulations of 'да д'
> (da d in cyrillic letters). When I firest *re*ran this, the output
> was garbage: ÿÿа ÿÿ, but after stepping through the first test
> trying various things including editing the Makefile.tests, and then
> reverting that, it seemed to work. For those who haven't seen this,
> it's all four utf8-{1..4} tests, and they have been failing for some
> time in my builds. NB - not related to tzdata! I'll probably defer
> further review until the tzdata is sorted.
>
> e2fsprogs-1.42.5 - the 9 failures remain.
>
> coreutils-8.17 - the 2 failures have gone. They were
> test-getlogin - insufficient ACL support
> test-parse-datetime - assertion failed
>
> Either removing a (possibly wrong) /etc/localtime did the trick, or
> else something I built later fixed coreutils.
>
> perl-5.16.1 - again, the 2 failures have gone.
> cpan/IO-Compress/t/105oneshot-zip-only
> cpan/Time-Local/t/Local
> (they were originally both 1 second different from the expected
> results) Same conclusion as coreutils.
>
> automake-1.12.2 - I delayed this until last, and it's still running.
> NB - 91 failures, I don't expect tz issues to be behind all of
> them, but I normally blow away the build when the package has been
> installed, so I've no idea what is going on until I've rerun it.
>
> flex-2.5.37 - still the 2 failures, they don't seem to be related to
> date/time consideration : test-bison-yyl{loc,val}. Again, I'll
> probably defer these until the tzdata is a bit clearer.
>
> patch-2.6.1 - again, the failure has gone
> patch --backup --set-utc 1
> like the perl, that was 1 second different when I first ran it.
>
> Doing without /etc/localtime is clearly not the correct solution,
> so I'm going to leave the automake tests running, and then I'll look
> at changing the tzdata and retesting.
I'm rerunning a full build right now. I intended to check all tests,
but it's not done yet. I did leave /etc/localtime set at GMT, but in
some cases have set 'TZ=UTC0 make check' or a variation.
We probably should be doing the zone date the same as Debian, but TZ
should override /etc/localtime.
This seems to be one of the more complex upgrades to LFS in some time.
I don't know if glibc did us a favor or not.
-- Bruce
--
http://linuxfromscratch.org/mailman/listinfo/lfs-dev
FAQ: http://www.linuxfromscratch.org/faq/
Unsubscribe: See the above information page
