On 21 April 2011 17:17, pablomme <email address hidden> wrote:
> Could somebody please change the openmpi-bin package dependency on
> openmpi-checkpoint from "recommends" to "suggests", to stop the flood of
> duplicates? This is what most commonly pulls blcr-dkms, which is non
> functional since kernel 2.6.35 and its failing to build causes an error
> to show up on every attempt to rebuild the dkms modules, which users
> tend to eventually report. There's about 0.95 reports a day on average
> (116 duplicates between 18 November and 21 April).
>
> libmpich2-2 also pulls it in via its "depends" on libcr0 - I think this
> is a mistake, it should be a "recommends" at most, but preferably a
> "suggests", like for OpenMPI. Neither MPI library needs blcr to work.
Why not fix the reporting problem by patching the bug submission tool
to check for duplicates? It would be a useful feature more generally
in other cases too surely?
I have a patch for this, adding support for more recent kernels in
testing/development with the upstream authors at the moment, but there
are a number of non-trivial changes in code that may have usage cases
I'm not aware of or subtle bugs in corner cases so I'm awaiting an ACK
from them to proceed.
On 21 April 2011 17:17, pablomme <email address hidden> wrote:
> Could somebody please change the openmpi-bin package dependency on
> openmpi-checkpoint from "recommends" to "suggests", to stop the flood of
> duplicates? This is what most commonly pulls blcr-dkms, which is non
> functional since kernel 2.6.35 and its failing to build causes an error
> to show up on every attempt to rebuild the dkms modules, which users
> tend to eventually report. There's about 0.95 reports a day on average
> (116 duplicates between 18 November and 21 April).
>
> libmpich2-2 also pulls it in via its "depends" on libcr0 - I think this
> is a mistake, it should be a "recommends" at most, but preferably a
> "suggests", like for OpenMPI. Neither MPI library needs blcr to work.
Why not fix the reporting problem by patching the bug submission tool
to check for duplicates? It would be a useful feature more generally
in other cases too surely?
I have a patch for this, adding support for more recent kernels in
testing/development with the upstream authors at the moment, but there
are a number of non-trivial changes in code that may have usage cases
I'm not aware of or subtle bugs in corner cases so I'm awaiting an ACK
from them to proceed.
Alan