On Wed, 2006-02-15 at 15:50 -0500, Jakub Jelinek wrote:
On Wed, Feb 15, 2006 at 09:47:25PM +0100, Ralf Corsepius wrote:
> > gcc-4.1.0-0.27
> > --------------
>
> > * Tue Feb 14 2006 Jakub Jelinek <jakub(a)redhat.com> 4.1.0-0.26
> > - update from gcc-4_1-branch (-r110903:110978)
> > - PRs fortran/20861, fortran/20871, fortran/25059, fortran/25070,
> > fortran/25083, fortran/25088, fortran/25103, fortran/26038,
> > fortran/26074, inline-asm/16194, libfortran/24685,
> > libfortran/25425, target/26141, tree-optimization/26258
> > - ABI change - revert to GCC 3.3 and earlier behaviour of
> > zero sized bitfields in packed structs (Michael Matz, PR middle-end/22275)
>
> Do I read this correctly - ABI change?
Yes.
> The whole FC5 mass rebuilds were in vain?
No, packed structures containing zero width bitfields are extremely rare.
OK,
nevertheless, something to keep an eye on.
Should somebody find such a struct in a core part of a central library
(very low level packages (kernel, glibc, drivers) or GUI-toolskits seem
likely candidates, to me), ....
And if something really cared about the exact layout of those
structures,
it wouldn't rely on this, as GCC < 3.4 behavior differed from 3.4/4.0.
This
shouldn't matter here, because the rebuild was initiated by the GCC
4.0->4.1 transition.
Ralf