Re: libtool pre-1.5b tests fail on 9 debian arches

[Scott James Remnant]

On Fri, 2003-09-26 at 20:46, Robert Millan wrote:

> The libtool upstream maintainers are preparing a new 1.5b release. On their
> behalf I've recently attempted to test a snapshot from CVS branch-1-5 on all
> architectures Debian supports (or pretends to support) that I had access to.
> 
Actually if it was branch-1-5 you were testing, that'd be the new 1.5.0a
(1.5.1) release.  1.5b would be on HEAD (as far as I understand the
esoteric version numbering upstream use) and a pre-release of
libtool 1.6 (which we know Gary wants to get out of the door alongside
Autoconf 2.58 and Automake 1.8).

> For testing libtool branch-1-5, get a CVS snapshot while passing
> "-r branch-1-5" to the checkout command, and run:
> 
>   export CVSROOT=:pserver:address@hidden:/cvsroot/libtool
>   cvs -z9 co -r branch-1-5 libtool ; cd libtool
>   ./bootstrap && ./configure && make && make check
> 
From a Debian point of view, this is a totally wrong thing to do and
will result in a *very* broken version of libtool being used!

Use the Debian libtool package, not only do I currently track CVS rather
than use the pure 1.5 release, there are additional Debian patches added
to make it work on some of the architectures.

Getting these patches accepted upstream is tricky though, I've sent some
bug fixes through.  A few days ago I decided to have a go getting some
of the portability patches (some of which are large) accepted, I mailed
the smallest (yet one of the more far-reaching ones) to -patches;
haven't had any follow-up yet though.

FWIW, I've run the same tests on Debian unstable with the Debian libtool
1.5-2 package.  I've also tried to ascertain *why* tests are failing, it
would've been nice if you could've done the same (or perhaps you're
working on that?)

        There's a slight bug in the Debian package, as
        tests/mdemo2-*.test are added in the .diff.gz, they never get
        made executable...

        Before running the tests, chmod a+x tests/mdemo2-*.test

        This will be fixed in the next Debian libtool package.

And now, as they say, the results...

[i386]
> gluck         i386    ok
> 
All tests passed with Debian's libtool package.

[powerpc]
> voltaire      powerpc ok
> 
All tests passed with Debian's libtool package.

[ia64]
> merulo                ia64    17 tests failed
> 
16 tests failed (out of 101) with Debian's libtool package.

They all fail because of:

PASS: demo-static.test
PASS: demo-make.test
FAIL: demo-exec.test

address@hidden:~/libtool-1.5/tests$ ./demo-exec.test
=== Running demo-exec.test
Executing uninstalled programs in ../demo
Welcome to GNU Hell!
Welcome to GNU Hell!
./demo-exec.test: cannot execute ../demo/helldl
address@hidden:~/libtool-1.5/tests$ ../demo/helldl
Welcome to *modular* GNU Hell!
found file: libhello.a
found symbol: hello
found symbol: foo
found symbol: nothing
Illegal instruction

It's the call to the function pointer obtained that generates the
illegal instruction.  An ia64 guru might want to look at this?

(mdemo and pdemo likewise fail)

[mips]
> casals                mips    ?? [3]
> 
All tests passed with Debian's libtool package.

[arm]
> debussy               arm     ok
> 
All tests passed with Debian's libtool package.

The upstream libtool (even CVS) can't compile (amongst other libraries)
GTK+ because the file_magic deplibs method is broken ... Debian includes
a necessary patch to change linux* to pass_all, which fixes this.

[alpha]
> escher                alpha   ok
> 
All tests passed with Debian's libtool package.

[s390]
> raptor                s390    3 tests failed
> 
All tests passed with Debian's libtool package.

[hppa]
> paer          hppa    4 tests failed
> 
1 test failed (out of 101) with Debian's libtool package.

PASS: demo-nopic.test
PASS: demo-make.test
FAIL: demo-exec.test

Entirely unsurprising given that hppa is the most fascist platform with
regards to PIC code...  You can't have Non-PIC shared libraries on hppa.

This should be in XFAIL_TESTS for hppa, or skipped (patched Debian
libtool appropriately).

[m68k]
> crest         m68k    ?? [3]
> 
All tests passed with Debian's libtool package.

[sparc]
> vore          sparc   ?? [3]
> 
All tests passed with Debian's libtool package.

[amd64]
> ravel [1]     amd64   25 tests failed
> 
(in i686 mode on pergolesi)
All tests passed with Debian's libtool package.

(in amd32 mode on ravel)
All tests passed with Debian's libtool package.

(in amd64 mode on ravel)
25 tests failed (out of 65) with Debian's libtool package.

This is most likely either amd64 isn't supported by libtool properly yet
(it's a very new architecture), given that the failures are all in the
*-make tests:

gcc-3.3: cannot specify -o with -c or -S and multiple compilations

[mipsel]
> williams      mipsel  ?? [2]
> 
Both vaughan and williams are still down :(  Though theoretically
there'd be no difference from the mips result.

[sh]
> ??            sh      ?? [4]
> 
sh isn't even vaguely approaching usefulness yet.

Scott

[0] On Debian libtool --version will report "1.5.0a" not "1.5"
-- 
Have you ever, ever felt like this?
Had strange things happen?  Are you going round the twist?

signature.asc
Description: This is a digitally signed message part