Re: Bug#430249: ldbl128 transition for alpha, powerpc, sparc, s390
On Thu, Oct 25, 2007 at 03:01:11PM -0700, Steve Langasek wrote:
> On Tue, Oct 23, 2007 at 12:17:56PM +0300, Damyan Ivanov wrote:
> > [d-release CC-ed for oppinion]
> > [please CC at least debian-perl]
> > > > > On 2007-06-23 Matthias Klose wrote:
> > > > > > Package: libdbi-perl
> > > > > > This package has been indentified as one with header files in
> > > > > > /usr/include matching 'long *double'. Please close this bug report
> > > > > > if it is a false positive, or rename the package accordingly.
> It appears that libdbi-perl is included in the list for the ldbl transition
> because of the file /usr/lib/perl5/auto/DBI/dbipport.h. I don't believe
> that anything in Debian builds against this file, but then I also don't know
> why it's in the package at all. If other packages do build against this
> header, then there is at least potentially an ABI change that needs to be
> handled. If this header is dead weight, then no changes need to be made to
> the libdbi-perl package for this bug report (not even "depending on a perl
> that is compiled with the new glibc/gcc").
The libdbd-*-perl packages do build against this file, through DBIXS.h.
The file provides a compatibility API for older versions of Perl so
that XS modules can use features introduced in newer versions. See
Devel::PPPort(3perl) for more information.
In this case, the relevant block is
# if defined(USE_LONG_DOUBLE) && defined(HAS_LONG_DOUBLE)
# define NVTYPE long double
# define NVTYPE double
which provides the NVTYPE definition when building with an old Perl that
doesn't know about it. Now, 'perl dbipport.h --api-info NVTYPE' says
Supported at least starting from perl-5.6.0.
Support by dbipport.h provided back to perl-5.003.
so this isn't used on Debian. Even if it were, our perl is compiled
without the 'uselongdouble' configuration parameter (see #430322), so the
'long double' alternative wouldn't be used anyway.
I'm closing the bug, please reopen if I have missed something. In that
case, #430264 should probably be reopened too.
Niko Tyni email@example.com