another semi-mass-rebuild (Was: Re: rebuilt for unwind info generation, broken in gcc-4.1.1-21)

[Jakub Jelinek]

On Mon, Oct 02, 2006 at 10:02:31PM +0200, Thorsten Leemhuis wrote:
> Thorsten Leemhuis schrieb:
> > Guys watching the rawhide report will notice that there are a lot of
> > packages listed today that have a Changelog entry which lists
> > 
> > [...]
> > - rebuilt for unwind info generation, broken in gcc-4.1.1-21
> > [...]
> > 
> > We probably need to rebuild some packages in Extras, too. FESCo
> > currently discusses when/how to do that.
> > 
> > Some background: It looks like gcc 4.1.1-21 introduced a bug that wasn't
> > fixed until gcc 4.1.1-26, so there's a window from Sept. 8 until Sept.
> > 26 within which any binary package  would have been built with the bad
> > compiler.

September, 18th actually, through September, 26th.
gcc-4.1.1-2{1,2} weren't ever added into rawhide tree, only 4.1.1-2{3,4,5}
were, the first one on Monday, 18th.

> Some additional details jeremy provided:
> 
> Essentially, the result is that backtraces in gdb won't (necessarily)
> work and that any app which calls backtrace() is likely to segfault.
> There are a few other potential ways that things can go wrong, but
> suffice to say that, yes, binary packages built in the window will need
> to be rebuilt :-/

The unwind info is also used for pthread_cancel and C++ exceptions BTW.
For more info about the bug see e.g.:
http://sources.redhat.com/ml/binutils/2006-09/msg00284.html
http://sources.redhat.com/ml/binutils/2006-09/msg00292.html
http://gcc.gnu.org/PR22313
http://gcc.gnu.org/PR29132

	Jakub