Fwd: Bug#595496: still fails...

[Thorsten Glaser <tg@mirbsd.de>]

Hi again,

if someone is bored, maybe that’s something to hack.

bye,
//mirabilos
-- 
I believe no one can invent an algorithm. One just happens to hit upon it
when God enlightens him. Or only God invents algorithms, we merely copy them.
If you don't believe in God, just consider God as Nature if you won't deny
existence.		-- Coywolf Qi Hunt

---------- Forwarded message ----------
From: Kurt Roeckx <kurt@roeckx.be>
Message-ID: <20101024180924.GA14638@roeckx.be>
To: Thorsten Glaser <tg@mirbsd.de>, 595496@bugs.debian.org
Date: Sun, 24 Oct 2010 20:09:24 +0200
Subject: Re: Bug#595496: still fails...

On Sun, Oct 24, 2010 at 03:26:53PM +0000, Thorsten Glaser wrote:
> OK, more details:
> 
> linux-2.6 build-depends on libelf-dev which is built and installable.
> 
> elfutils is not installable because libasm1 isn't built (but that
> doesn't matter for compiling a linux kernel). This is probably
> what you were hinting at with saying it has never worked on m68k.

There is alot of code in elfutils that is arch specific, as far
as I know all related to dwarf debug information.  The source files
are in the backends/ directory and end up in the libdw1 binary package.
There is a minimal patch in the source packages:
debian/patches/m68k_backend.diff.  There are tests that for
instance try to read a compiled file with debug info, and they will
probably fail for m68k binaries.  So the libdw1 package is
probably currenly in a useless state on m68k, but the libelf1
binary package should not be a problem at all.

I see 2 ways to fix your problem:
- Disable the m68k backend completly and don't create an
  libdw1/libdw-dev package on m68k.
- Add proper support for m68k.  I think to be able to
  pass the regression tests you need atleast information
  about the return location of various types.  Atleast that's
  the error shown in the last buildd log I have.  And m68k
  is the only arch not having an *_retval.c file.

I would prefer the second solution.


Kurt