These are chat archives for elemental/chat

22nd
Nov 2016
Jack Poulson
@poulson
Nov 22 2016 03:30
I've confirmed that elemental/Elemental#198 is a legitimate issue (that somehow did not effect Ubuntu GCC builds)
I will push out 0.87.1 ASAP
Ryan H. Lewis
@rhl-
Nov 22 2016 07:24
@poulson given that there is now two updates to 0.87 perhaps we should mark them as release candidates
To give users an opportunity to report issues and then we can address them
Jack Poulson
@poulson
Nov 22 2016 07:25
let's do that next release
there was an RC for 0.86 but no official release and I wanted to avoid that happening again
Ryan H. Lewis
@rhl-
Nov 22 2016 07:26
:)
I'm surprised that Travis builds are down to 8 minutes
Jack Poulson
@poulson
Nov 22 2016 07:28
no more ScaLAPACK
Ryan H. Lewis
@rhl-
Nov 22 2016 07:28
Even with ccache my desktop is ~30
Is that really the problem ? But my desktop shouldn't have that either
@poulson can you take a quick look: https://bugzilla.redhat.com/show_bug.cgi?id=1389971
im not sure if im going to try and get ppc64 builds working
Ryan H. Lewis
@rhl-
Nov 22 2016 07:35
is this accurate:
  • Code under external/suite_sparse is under LGPLv2+ license.
    Other code/cmake files is MIT, Boost.
    Please, update License tag and pack debian/copyright by using
    %license tag.
Jack Poulson
@poulson
Nov 22 2016 07:35
see the debian/license file
err, debian/copyright
Ryan H. Lewis
@rhl-
Nov 22 2016 07:38
the PPC64 issues are here:
if you are interested
looks like some segfaults
Jack Poulson
@poulson
Nov 22 2016 07:38
yes, I have been looking
it looks like it's QuadDouble issues
likely a QD issue
Ryan H. Lewis
@rhl-
Nov 22 2016 07:39
an upstream problem? or an elemental problem?
Jack Poulson
@poulson
Nov 22 2016 07:39
probably upstream
the hackery is in QD itself
part of the oddness could be make test running several jobs at once due to the multithreading
actually, it looks like the failures are in libstdc++ calls
in particular, inserting floats into a stream buf
it's possible it's another toolchain issue
but the first priority should be running those failing tests in a debugger
and through valgrind
Ryan H. Lewis
@rhl-
Nov 22 2016 07:46
on a ppc64 machine ?
Ryan H. Lewis
@rhl-
Nov 22 2016 07:47
  • MUST: If the package does not successfully compile, build or work on an architecture, then those architectures should be listed in the spec in ExcludeArch. Each architecture listed in ExcludeArch MUST have a bug filed in bugzilla, describing the reason that the package does not compile/build/work on that architecture. The bug number MUST be placed in a comment, next to the corresponding ExcludeArch line. [8]
I only have macs
valgrind isn't running on any of them :)
Jack Poulson
@poulson
Nov 22 2016 07:48
do they provide hooks for running a debugger or valgrind on the test machines?
Ryan H. Lewis
@rhl-
Nov 22 2016 07:48
nope
Jack Poulson
@poulson
Nov 22 2016 07:48
one could hack the make test to prefix valgrind...
Ryan H. Lewis
@rhl-
Nov 22 2016 07:49
its been a long time, but, is valgrind interactive? we could submit a scratch build for ppc just for this purpose I suppose..
Jack Poulson
@poulson
Nov 22 2016 07:49
valgrind is not interactive
it might actually be worth us having an extension to make test that runs valgrind
e.g., make valgrind-test
PETSc devs have the laudable convention of always demanding valgrind-clean builds
Ryan H. Lewis
@rhl-
Nov 22 2016 07:54
yeah, thats not unreasonable.
Jack Poulson
@poulson
Nov 22 2016 16:25
speaking of the devil, the Satish and Hong (from PETSc) have reported some unused variables in El/number_theory/lattice/NearestPlane.hpp that I am going to remove
I'm also adding in the missing C and Python interfaces for reading a DistSparseMatrix in MATRIX_MARKET format
Ryan H. Lewis
@rhl-
Nov 22 2016 17:27
Is there a mailing list that I am not on?
Jack Poulson
@poulson
Nov 22 2016 18:32
Apparently you never signed up. dev@libelemental.org and friends
Ryan H. Lewis
@rhl-
Nov 22 2016 22:14
I'll have to get on it