[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Bug#519006: mips/ld: non-dynamic relocations refer to dynamic symbol

Aurelien Jarno <aurelien@aurel32.net> writes:

> Hi all,

> I have made some progress on this bug, though it has not progress in the
> expected direction. In other words I got some surprises.
> First of all it the bug has been introduced in binutils when introducing
> the MIPS PLT support [1] (commits around 2008-08-08). I am still
> convinced it is actually a binutils bug. Secondly, if the problem is
> actually not present in gcc-snapshot in gcc 4.5 we have *in Debian*, it
> is not due to the fix linked in the bugzilla entry. It is due to the
> fact we default to -mplt on these compilers. It is possible to
> workaround the bug on gcc 4.4 using -mplt (like it is possible to 
> workaround it by dropping the -g).

To sum up, there's no fix upstream for the bug but the
solution/workaround for now is -mplt, right ?

> Given all of that, the only forseen solution for this fix is to also
> default to -mplt on gcc 4.4. It also brings speed improvements, as 
> well as some more "standard" binaries with regard to other 
> architectures. On the other hand, I understand it is something quite 
> risky so close to a release. If we go for this solution, we may want

If the bug is "fixed" by this change and as a side effect, there are
improvements, there's no reason to not include it. I don't think it can
be worse than the current situation. Moreover (iirc) this bug is a
release blocker for mips so seeing it closed would be really nice.

> to trigger a rebuild of part of the archive as it is currently done
> on sparc.

There's a bunch of packages known to be bitten by this bug but there may
be others we're not aware of. imho a rebuild is mandatory. This would
also give us a good up-to-date status of the mips/mipsel ports for the

My 2 cents,

Reply to: