Bug#812852: Vigra auf ppc64el -- continues a little bit

[Daniel Stender]

On 05.03.2016 21:34, Daniel Stender wrote:
> On 04.03.2016 15:56, Ullrich Koethe wrote:
>> * Remove all occurencies of threading::memory_order* in multi_array_chunked.hxx (or equivalently, replace them with threading::memory_order_seq_cst). This will force all atomic accesses to use sequentially consistent ordering, the most strict possibility.
> 
> That worked out ... much thanks!

- First of all, there was a little mistake, that was the wrong log file. Unfortunately, the
  problem with test_multiarray_chunked resp. multi_array_chunked.hxx on ppc64el/le remains also
  with changing `memory_order_release/acquire` to `memory_order_seq_cst`.

- To build with Boost.Thread like now on armel doesn't have any effect on this, either.

Moreinfo:

- Yes, that break is reproducible, it takes place at every current build on this arch.

- AddressSanitizer undefined & adress doesn't brought up something, and thread check isn't
  available on this arch ("cannot find -ltsan", at least out of the box?). It seems
  it runs but just doesn't brings the expected results ...

I'm further experimenting with VIGRA_SINGLE_THREADED but no breakthrough so far ...

Concerning Debian:

I'm thinking about failsaving this test and just let it build remaining being problematic
on ppc64el until it get solved (I don't know if this says something, but Debian Code Search
doesn't bring up a use of <vigra/multi_array_chunked.hxx> in any other package) - there are
people waiting for the transition to get completed [1] so that they could rebuild their packages,
unfortunately hanging because of [2].

Anyway, any hint welcome (not only by upstream), any ppc64el guru out there?

Thanks,
DS

[1] https://bugs.debian.org/815153 (transition: libvigraimpex)

[2] https://bugs.debian.org/813415 (libvigraimpex5v5: soname bump without package name change)

---
4096R/DF5182C8
http://www.danielstender.com/blog/