Re: [bug-gnu-libiconv] Problems building a 64-bit iconv on Solaris x86

[Dr. David Kirkby]

On 08/10/10 08:44 PM, Bruno Haible wrote:
> Hi David,

Hi.

Sorry for the delay in replying. I had not noticed your first message, only the 
second one you sent today.

> Note that you had started a similar thread in
>    <http://lists.gnu.org/archive/html/bug-gnu-libiconv/2010-05/msg00000.html>,
> in which case you said that you wanted to try a newer gcc.
>
> Also, in
>    <http://www.mail-archive.com/maintainers%40lists.opencsw.org/msg06385.html>
> you said that switching to a newer gcc fixed the problem.
>
> > A little bit of information is given on a build failure on another machine
> > (hostname disk.math.washington.edu), which runs OpenSolaris.
> >
> > http://trac.sagemath.org/sage_trac/ticket/9405
>
> The compilation ends with
> Text relocation remains                         referenced
> aliases_lookup                      0x1ab1e     .libs/iconv.o
>
> This suggests that you should either try a newer gcc (see above), or
> simplify the storage-class specifiers of the function 'aliases_lookup'
> from lib/aliases_syssolaris.h so that it only it 'static __inline',
> or maybe even only 'static'.
>
> Additionally, your log file says
>    "gcc driver version 4.3.2 (GCC) executing gcc version 4.2.3"
> I would not do this. I would use a gcc driver of the same version
> than the cc1 that is being executed. What's the benefit of using
> a gcc driver 4.3.2 for reading the specs file of gcc 4.2.3? Sounds
> dangerous.

I've actually concluded disk.math was badly set up, and gave up with that.

> > A log file of a 64-bit failed build on host 'fulvia' running Solaris 10 update 5
> > 05/2008 can be seen here
> >
> > http://trac.sagemath.org/sage_trac/attachment/ticket/9718/iconv-1.13.1.p2.log
> >
> > The config.log can be seen here.
> >
> > http://trac.sagemath.org/sage_trac/attachment/ticket/9718/config.log
>
> This one is already using gcc 4.5.1, so I suggest to you to report a bug
> to the GCC developers, like we discussed in
>    <http://lists.gnu.org/archive/html/bug-gnu-libiconv/2010-05/msg00004.html>.
>
> Bruno

As I remarked in my earlier email, it's not a gcc bug, as it is possible to link 
the library properly by choice of the right options to the linker. But it 
appears libtool is not sending those right options.

I think this is a libtool bug, which I put on my other mail, so I'm not going to 
repeat here.

Dave