Bug#137959: Bug #137959
Greetings! On second thought, I think my previous message was a bit
unclear.
OK, blas-1.1, which does integrity checks now as part of the build,
has built everywhere expect for hppa for some time. Coincidentally to
your note, I traced this down *last night* to a bug in the loader on
hppa which messes up the link ordering of dynamic objects. I
implemented a simple temporary work-around, and released -8, which now
compiles everywhere except arm, where g77 now dies with a catastrophic
internal failure.
The hppa bug listed in this report was worked around by *increasing*
the optimization -- the issue only shows up with -g on hppa.
So here is a summary of problems:
hppa g77 -g bug as listed in #137959 could have been there
forever as far as I know,
as the package was never
compiled with just -g before
hppa g77 -O tester failure known problem with the
(and dynamic loader, also not
higher) necessarily new.
arm g77 compiler failure new to -8, must be due to a
(default compiler upgrade on this arch
optimizations)
LaMont Jones <lamont@hp.com> writes:
> > Can you clarify whether this is a new failure, or one that has just gone
> > unnoticed previously?
>
> The blas package has previously built (although auric is down so I can't
> see _which_version...). I don't know if this code changed between then and
> now, or if this is a new bug in g77.
>
> > I'm downgrading it to "important" on the assumption that the bug has
> > existed in all older versions of the compiler as well. If it's actually
> > a regression, please feel free to set it back to "serious"; in that
> > case, it would help to know what the last version was that worked.
>
> > It would also be interesting to know whether gcc-snapshot is able to
> > compile the code successfully.
>
> I'll work on these later tonight when I get home.
>
> lamont
>
>
--
Camm Maguire camm@enhanced.com
==========================================================================
"The earth is but one country, and mankind its citizens." -- Baha'u'llah
--
To UNSUBSCRIBE, email to debian-gcc-request@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Reply to: