[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Shared libraries, dependencies and symbols files


On Tue, Oct 02, 2007 at 10:09:01AM +0200, Raphael Hertzog wrote:
> On Mon, 01 Oct 2007, Raphael Hertzog wrote:
> > In other cases, most of the architectures are in sync and some
> > special architectures are creating troubles. It looks like
> > hurd-i386 is currently causing me lots of troubles for example on pango:
> > http://qa.debian.org/cgi-bin/mole/seedsymbols?pkgname=libpango1.0-0

First off, many thanks for this work, I believe it is very valuable to
the Hurd port (well, and Debian in general as well, of course).

> > Pango export the same set of symbols across architectures but apparently
> > hurd-i386 is an exception that makes my script believe otherwise.
> > 
> > (m68k has a different md5sum too, but the same set of symbols, it's only
> > the version which are different since m68k didn't manage to build the
> > exact same version historically... however hurd-i386 has more symbols
> > in this case, it looks like it exports symbols that should be hidden)
> Can some hurd people check and explain me why hurd-i386 libs export
> private symbols ?

In general, the Hurd port uses the same tools, so the most likely
explanation are configure checks which mean "gnu/glibc" but actually say
I'm trying to look into pango now.

> Because IIRC (but I didn't check), the various "_pango*" are marked as
> "hidden" with gcc's visibility attribute (same goes for the "_hb_*" and
> "_HB_*" symbols).

> This difference is undesirable and complicates somewhat the work of
> maintainers who will have to deal with arch-specific differences when
> there shouldn't be any.

Agreed.  Would it be easy to have a list of those libraries which differ
on hurd-i386, unless it's almost every lib we're talking about anyway?

I checked a couple, and it seems glib2, gtk2, and gnome-vfs2 seem to
have various issues.

Reply to: