Ken Moffat wrote:
  As the book correctly says: "Five tests are known to fail in the
LFS build envronment, but all tests pass if rechecked after Chapter
6 is complete."

  I took a look at this, because it implies the build order might be
inadequate for the current version.  After looking, I suspect it
shows two things:

1. These tests want automake (more specifically, they use aclocal).

2. The invocation of am_missing_run *might* be buggy - the autotools
are always hard to grok, but I sort-of expected that am_missing_run
might report that automake or aclocal is required.  The tests now
use 'the autotest harness', perhaps the error gets ignored there.

  Trying to debug the differences between good and bad runs is hard,
because the directory for any test which succeeds gets deleted
immediately after that test.  Also, I thought I could edit the
Makefile afterwards to remove most of the earlier tests so that I
didn't have to wait around, but it continues to run all of them.  I
interrupted during test 123 and got some information :

(i.) the initial failure in test 122 is repeated in all five tests:
/bin/sh: no: command not found
  and that was from
CDPATH="${ZSH_VERSION+.} && cd . && no -I m4

(ii.) for test 123, the Makefile for the test included
ACLOCAL = no
but in the rerun at the end of chapter 6
ACLOCAL = aclocal

(iii.) The relevant line in the configure script for this test is
ACLOCAL=${ACLOCAL-"${am_missing_run}aclocal-${am__api_version}"}

(iv.) The tests which fail are from standalone.at and subproject.at.

Summary:

  Anybody feeling clever and want to disable these tests ?

More effort than it's worth IMO.

  Or could we say that these 5 tests need automake which cannot
be installed until later ?

What you are describing is a circular dependency, at least as far as the regression tests go. We could change the wording of why the 5 tests fail, but does anything else provide the user anything of value?

  Also, should we add automake to the dependencies for the test suite
of Libtool in Appendix C, in the same way that automake is a
dependency of the autoconf tests ?

Yes, that is reasonable.

  -- Bruce

--
http://lists.linuxfromscratch.org/listinfo/lfs-dev
FAQ: http://www.linuxfromscratch.org/faq/
Unsubscribe: See the above information page

Reply via email to