test results, Tuesday Call

Reminder:  the lsbsi call is now Tuesday at
11:00 MST - that is one hour earlier than
previously, in the time slot immediately 
following the lsb-futures call.

lsbsi update:

(1) beta bits are available, we just need to
make the announcement.  I think the webpages
are okay, although they could be restructured
to be more useful.

(2) I did a crosscheck build on a not-Debian
PPC distro, and it built okay. Have not been
able to test, but I was interested mainly in
the ability to build.  This means each of the
three lsbsi's has been built successfully on
at least two different distros. Yay!

(3) I've stamped the cvs tree for lsbsi-1_3b1
and moved the current version to 1.3b2 so
there won't be any confusion (doesn't mean
we'll actually have a beta 2, however).

(4) two base packages have "unofficial" versions
with interesting status:

findutils is at 4.1.7 ("released" is 4.1)
and tar is at 1.13.25 ("released" is 1.13).

Both have been in this "alpha" state for a
very long time, and all distros I have access
to, even old ones, are using these versions.

Benefits of "alpha" versions:
1. eliminates need to use lsbsi-specific patch
2. eliminates all lsb-runtime test failures in
   these two packages
3. matches what everyone seems to be using (less
   any vendor-specific patches, of course)

My proposal is to use these two interim versions
in lsbsi-1.3, but I'm open to persuasion. It
doesn't *quite* follow our rule of released
upstream packages only.

(5) on IA64, the "fold" utility from textutils
core dumps.  This is not unique to the lsbsi,
as it happens on the reference system I have
as well. On the ia32 lsbsi, from the same source,
there is no problem, which hints at a problem
of the class of "integer used as pointer".

The successor to textutils is coreutils (which
also incorporates fileutils and sh-utils), and
the version of "fold" I build from coreutils
runs correctly.

I don't propose to shift to coreutils, since
coreutils has never been officially released,
and this is stretching the rules a bit too much.
It does imply a patch could be generated, 
although I have not taken the time to examine
how much trouble this would be.

(6) There's still som discussion of tests to
exclude from the lsb-runtime-test because they
require commands that are not part of the lsb.
The cutoff date is Wed, when the (hopefully)
final lsb-runtime-test will be cut.

ed, ex, vi and nm are clearly not part of the
lsb, and it looks like these tests will be
eliminated.

iconv was failing due to using a c compiler,
not part of the lsb; however, a patch has been
applied which moves the compilation step to the
test suite build stage, so this should now pass.

gencat also uses the c compiler, there's no
workaround yet, apparently.

setlocale is also showing some "compilation"
problems, looks like some binary is missing
but I don't have details.  Maybe something
to do with localedef?  Investigating.


The remaining 46 failures, mostly from textutils
programs, I'd expect to leave alone and just
document in the release notes.