Bug#836130: trilinos: FTBFS without __sync_val_compare_and_swap_8

[Aaron M. Ucko]

Graham Inggs  writes:

> I'm dropping this bug's severity to wishlist as Trilinos last built on
> 32-bit architectures in 2010 [1], before the Kokkos package was
> introduced.  These binaries are not in testing and therefore should
> not prevent Trilinos from migrating.

So I see.  It looks like the only outstanding regression is on alpha,
which is a non-release, 64-bit architecture, and moreover hit a cp
segfault(!) rather than anything that could plausibly be interpreted as
a Trilinos portability bug.

> We have opened a PR [3] with upstream, in the hopes that we can work
> together on a solution.

Great; thanks for all your work here.

> PS: If you are happy with the outcome of #815725, please close it.

Done earlier today.  In retrospect, that really should have been two
separate reports.

-- 
Aaron M. Ucko, KB1CJC (amu at alum.mit.edu, ucko at debian.org)
http://www.mit.edu/~amu/ | http://stuff.mit.edu/cgi/finger/?a...@monk.mit.edu

Bug#836130: trilinos: FTBFS without __sync_val_compare_and_swap_8

[Graham Inggs]

severity 836130 wishlist
retitle 836130 trilinos: FTBFS without __sync_val_compare_and_swap_8
thanks


Hi Aaron

On 30 August 2016 at 20:18, Aaron M. Ucko  wrote:
> Source: trilinos
> Version: 12.6.4-1
> Severity: serious
> Justification: fails to build from source (but built successfully in the past)

I'm dropping this bug's severity to wishlist as Trilinos last built on
32-bit architectures in 2010 [1], before the Kokkos package was
introduced.  These binaries are not in testing and therefore should
not prevent Trilinos from migrating.

> Thanks for looking into #835406!  trilinos now successfully builds on
> i386 and x32, and the builds for other 32-bit architectures don't fail
> as quickly as they used to, but they nevertheless remain broken (at
> least on mips, mipsel, and powerpc -- builds for other architectures
> are still in progress):

The build was also successful on armhf, but still fails on
architectures without '__sync_val_compare_and_swap_8'.  There's a GCC
bug [2] requesting an implementation, however it has been closed
WONTFIX.  Before you reassign this bug to gcc in Debian, there seems
to be a workaround mentioned in that bug which I intend investigating
when I have some time to spend on a porterbox, probably after the
openmpi transition.

We have opened a PR [3] with upstream, in the hopes that we can work
together on a solution.

Regards
Graham

PS: If you are happy with the outcome of #815725, please close it.


[1] https://buildd.debian.org/status/logs.php?pkg=trilinos&arch=powerpc
[2] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63368
[3] https://github.com/kokkos/kokkos/pull/410