From kde-core-devel  Thu Jan 05 01:38:30 2006
From: Maks Orlovich <mo85 () cornell ! edu>
Date: Thu, 05 Jan 2006 01:38:30 +0000
To: kde-core-devel
Subject: Re: KDE/kdebase/kicker/kicker
Message-Id: <200601042038.30856.mo85 () cornell ! edu>
X-MARC-Message: https://marc.info/?l=kde-core-devel&m=113642673921492

On Wednesday 04 January 2006 19:06, Waldo Bastian wrote:
> On Wednesday 04 January 2006 11:08, Maks Orlovich wrote:
> > >  * one for GCC 3.4/4.0 ABI
> > >
> > > In theory, the latter two should be identical (they both try to
> > > implement the Itanium C++ ABI specification (go to
> > > http://www.codesourcery.com/cxx-abi/ for a full documentation).
> >
> > In practice, even with identical versions of GCC, you'd find that many
> > Linux distributions break the libstdc++ ABI.
>
> Please be specific. Which distribution did break the ABI when and with
> which version of GCC? Vague statements like the one above tend to get
> repeated and gain alife of their own without any way to evaluate their
> truth.





   
On Wednesday 04 January 2006 19:06, Waldo Bastian wrote:
> On Wednesday 04 January 2006 11:08, Maks Orlovich wrote:
> > >  * one for GCC 3.4/4.0 ABI
> > >
> > > In theory, the latter two should be identical (they both try to
> > > implement the Itanium C++ ABI specification (go to
> > > http://www.codesourcery.com/cxx-abi/ for a full documentation).
> >
> > In practice, even with identical versions of GCC, you'd find that many
> > Linux distributions break the libstdc++ ABI.
>
> Please be specific. Which distribution did break the ABI when and with
> which version of GCC? 

Debian and derivatives (some of which have since resolved this), Gentoo, and 
potentially --- I did not have direct testing access to any of those, so this 
is indirect based on googling of diffs --- some of RH/Fedora systems, for 
some periods of time (likely to be reverted now). GCC version is 4.0.x. The 
issue is use of an incompatible default allocator for the STL, which implies 
that if two dynamically linked objects were built on systems with and w/o a 
change, and they pass an STL container around (think of a function returning a 
vector --- aRts has those), destruction will crash. 
Mandriva and SUSE are compatible with FSF GCC.

>Vague statements like the one above tend to get repeated and 
> gain alife of their own without any way to evaluate their truth.

I am not Zack or Aaron. No one cares about what I say ;-)