On 2/11/19 12:09 PM, Bruce Dubbs via lfs-dev wrote:
On 02/11/2019 11:29 AM, Pierre Labastie via lfs-dev wrote:
Hi,

I already wrote (at a wrong place) that I was worried about
the number of test failures in some LFS test suites. Here are
my findings about the bash test suite:
- Two locales are missing: Ja_JP.SIJS and zh_HK.big5hkscs. Can be
   created with:
   "localedef -i ja_JP -f SHIFT_JIS ja_JP.SIJS"
   "localedef -i zh_HK -f BIG5-HKSCS zh_HK.BIG5-HKSCS"
- Several tests generate something like "/dev/null: not a directory".
   This is because nobody's HOME is /dev/null. The tests rely to the
   ability to run "cd", which would change directory to $HOME.
   adding HOME=/mnt to the command seems to be enough.
   not that it cannot be /tmp, because one test uses "pushd /tmp",
   which prints ~ if /HOME=/tmp, but of course the test expects
   /tmp...
- Some tests complain about chmod, touch, /usr/bin/printf, mkdir, ... not found.
   The reason is that for some tests, the PATH is hardcoded to
   /bin:/usr/bin:/usr/local/bin, but those programs are only available
   in /tools/bin... Actually, on the dependency page, we have that
   coreutils should be built before bash, but 3 years ago coreutils
   was moved after automake (because at the time it needed
   aclocal, AFAIU). Maybe coreutils should be moved back to be
   built before bash? Another possibility is to (temporarily) symlink
   those programs to /usr/bin.
- Some tests seem to not find /dev/tty. I've not investigated those
   yet...
- One test outputs: "bash: cannot set terminal process group: Inappropriate ioctl for device"
Not investigated yet.

We are going into package freeze in a couple of days.  Probably Wednesday.  When we freeze, I normally update any outstanding tickets and do a full reference build with all tests.  I check for errors at that time and update the book as needed.

In some cases a package update can break tests for other packages.  An example is bash-5 breaking many tests in autoconf. That has been reported upstream, but there is little indication of a release to fix the problem.

We can certainly add the newer locales.  We can also add symlinks to Section 6.6.  That is relatively painless, but I'm not sure we really need to fix every issue.  Just documenting that specific tests are broken in the chroot environment should be sufficient.

I don't understand the /dev/tty issue.  That should be available via the mount -v --bind /dev $LFS/dev in Section 6.2.

Besides the locales, more specific recommendations are welcome.

While we're here, can we please add the C.UTF-8 locale? That will resolve issues with the Gjs test suite in BLFS.
--
http://lists.linuxfromscratch.org/listinfo/lfs-dev
FAQ: http://www.linuxfromscratch.org/faq/
Unsubscribe: See the above information page

Reply via email to