Re: New packages gnumach, hurd, glibc
On Mon, Jul 23, 2001 at 03:58:58PM +0200, Robert Bihlmeyer wrote:
> Marcus Brinkmann <Marcus.Brinkmann@ruhr-uni-bochum.de> writes:
>
> > Yes, there is no promise on our side to make sure the dependencies
> > are sane. I really don't think we should even make an attempt to
> > provide a sane upgrade path.
>
> I beg to differ.
You are welcome ;) I will try to let you know when I become aware
of such incompatibilities.
> > and we will definitely break the libc ABI completely anyway when we
> > do the switch from stdio to libio (and/or to pthreads). Then we will
> > have to recompile the whole archive, and any such clues become
> > obsolete.
>
> Couldn't we just change the package name? Every package using libc
> depends on libc0.2.
Of course, and we will certainly change the name. The problems are
other libraries (remember the libc5 -> glibc transition in Linux).
If you have a program compiled with the new glibc and readline4,
and install it on a system with a readline4 compiled against the old
glibc, the linker will suck in both glibcs (because the readline4
ABI did not change, this will appear to be okay). Usually this
results in segfaults.
This is why all libraries had to be renamed to libfoo1g during the transition.
After the ABI changed, the g could be dropped again.
I doubt it will be feasible to load this burden on the package maintainers
for a Hurd libc ABI change.
> I'm content with your notice, BTW -- I just wanted the reasons. I've
> also filed a glibc bug, since it already has a (wrong) dependency on
> hurd.
As I said, I am cool.
> > I have a random translator, which I can include in the package. We just
> > need to find a way to get good randomness. If someone could make a port
> > of egd to the Hurd, which is started at bootstrap and copies good entropy
> > to /dev/random, this would be all that is needed.
>
> Shouldn't the egd port simply act as a translator?
It could act as a translator which /dev/random can the poll from.
If you write a translator which provides good entropy on read(),
I will change my /dev/random implementation to poll from files.
Thanks,
Marcus
Reply to: