Breakage with bleeding edge host toolchains (Fedora)
dbn.lists at gmail.com
Wed Feb 28 10:16:46 PST 2007
On 2/28/07, Matthew Burgess <matthew at linuxfromscratch.org> wrote:
> On Wednesday 28 February 2007 04:44, Dan Nicholson wrote:
> > For this case, it looks like there's a couple different ways we can
> > attack this. But I think we agree that this is a good thing to do.
> > Let's give Matthew and anyone else to chime in.
> I'd say go with the '-B' switch too. It's being used for its intended
> purpose, and the 'non-executable ld' trick won't work as Bryan pointed out.
>  suggests it'll work on gcc-2.95.3 hosts too, so we won't be introducing
> any new host requirements either.
Greg pointed out that it doesn't work on gcc-2.95.3. The reason is
not because -B is supported. I don't have this to test, so I'll just
pass on what Greg said. In gcc-2.95.3, if you pass -B and it's not
needed for anything, it will issue a warning on stderr that says
gcc: file path prefix `/usr/bin/' never used
This is a problem because some of the feature tests in the
gcc/binutils configures check for failed return status by checking
whether there was any output to stderr. That's obviously dumb. I ran
into this issue when I was trying to compile with CC="gcc -v" so I
could get really verbose output during a bootstrap (-v sends its
output to stderr). I believe the problem was in the libiberty/
I don't know if this is an still an issue in gcc-4.1.2 and
binutils-2.17. Someone would have to check. DIY has support for the
older versions, and I'm pretty sure I was messing around compiling
gcc-3.4.6 at the time.
More information about the lfs-dev