00:29 <github> [hpx] K-ballo created dataflow-api (+1 new commit): https://git.io/vdb0a

00:29 <github> hpx/dataflow-api 238e62a Agustin K-ballo Berge: Cleaning up dataflow overload set

00:30 <K-ballo> ^ first steps, no functional changes (intended)

00:30 <K-ballo> there's also a ton of duplication in those files

00:30 <K-ballo> hkaiser: I'm going to deprecate the lcos/local/dataflow.hpp header, move everything but the lcos::local already deprecated overload to lcos/dataflow.hpp

01:31 <hkaiser> K-ballo: I think it's deprecated long enough, we can remove it now

01:31 <hkaiser> pls look into the main CMakeLists.txt, there is a note when it was deprecated

01:31 <K-ballo> V0.9.99

01:32 <hkaiser> yah, let's remove it

01:32 <hkaiser> at least disable it by default

06:55 <github> [hpx] taeguk opened pull request #2967: Make parallel::merge is stable. (Fix #2964.) (master...fixing_2964) https://git.io/vdbis

09:08 <github> [hpx] StellarBot pushed 1 new commit to gh-pages: https://git.io/vdbMn

09:08 <github> hpx/gh-pages 7dd8100 StellarBot: Updating docs

11:55 <github> [hpx] K-ballo force-pushed dataflow-api from 238e62a to 8ec263b: https://git.io/vdbHQ

11:55 <github> hpx/dataflow-api 8ec263b Agustin K-ballo Berge: Cleaning up dataflow overload set

15:49 <github> [hpx] K-ballo created syntax-only-header-tests (+1 new commit): https://git.io/vdbjd

15:49 <github> hpx/syntax-only-header-tests 69e3d5c Agustin K-ballo Berge: Add internal cmake flag for using -fsyntax-only when building header tests

16:11 <K-ballo> heller: you broke inspect checks on master?

16:14 <K-ballo> here https://github.com/STEllAR-GROUP/hpx/blob/master/tests/unit/resource/throttle.cpp#L92

16:16 <github> [hpx] K-ballo pushed 2 new commits to dataflow-api: https://git.io/vdNvG

16:16 <github> hpx/dataflow-api 0a0272d Agustin K-ballo Berge: Consolidate local and remote dataflow implementations

16:16 <github> hpx/dataflow-api 306ec04 Agustin K-ballo Berge: Simplify dataflow implementation

18:25 <heller> K-ballo: how was that not caught before?

18:27 <K-ballo> nobody was paying attention I suppose, it's been failing on circle for 4 days

18:28 <K-ballo> I only saw it because it flagged my own branch

18:41 <heller> Uh

18:44 <hkaiser> heller: I'm going to comment out the affinity error - it;s benign and makes everything fail without reason

18:46 <heller> Yes

18:47 <heller> hkaiser: shouldn't we rather set no affinity mask if bind=none was set?

18:54 <parsa> hkaiser: Klaus said he has no plans to make determinant() work without BLAS.

18:54 <parsa> what do you suggest? there isn't just one BLAS implementation that i can look for

19:08 <zao> Doesn't FindBLAS.cmake find anything conforming?

19:08 <zao> May it be MKL, OpenBLAS, GOTO, etc.

19:16 <parsa> do you know how to make it tell me where too look for libraries? on mac it spits out these: https://gist.github.com/parsa/83558e636ca184080d759583c30cc8d2 and nowhere in those directories i find a useable header file/library

19:20 <zao> BLASLIBRARIES is suitable for putting in a target_link_libraries command.

19:20 <zao> https://cmake.org/cmake/help/latest/module/FindBLAS.html

19:21 <zao> Sorry, BLAS_LIBRARIES

19:21 <zao> Not sure what to do about C/C++ headers.

19:23 <zao> It may be that you're supposed to have a cblas.h in path already or something.

19:24 <zao> Don't have any source to look at atm, sadly.

19:24 <parsa> i don't why that worked (i don't see any libraries in that directory) but it did. thank a bunch zao!

19:25 <parsa> >don't see* why that worked

19:25 <zao> On macOS, you can link "frameworks" which are kind of an bundle of headers, libraries, and associated stuff.

19:25 <zao> You can cd into them in a shell, while Finder shows them opaque.

19:26 <parsa> i mean i did a find <> -iname '*blas*' on them, didn't see anything useful there

19:30 <zao> Guessing that any eventual header is elsewhere, and the library name varies among implementatinos.

19:30 <zao> In particular, Apple's seem to be "Accelerate".

19:30 <zao> While MKL would be mkl, etc.

20:49 <jbjnr> hkaiser: is the input to a continuation a pair?

20:49 <hkaiser> no

20:49 <hkaiser> just the future

20:50 <jbjnr> even when I remove all unwrapping and dataflow, I get a pair of the future and some other stuff "std::pair<hpx::threads::thread_state_enum, boost::intrusive_ptr<hpx::threads::thread_data> > (hpx::lcos::detail::continuation<hpx::lcos::future<double>, hpx::util::detail::bound<hpx::util::detail::one_shot_wrapper<int (*)(hpx::lcos::future<double>&&)> (hpx::util::detail::placeholder<1ul>...

20:50 <jbjnr> ...const&)>, int>::*)(boost::intrusive_ptr<hpx::lcos::detail::future_data<double> >&&)"

20:51 <hkaiser> that's the return type of the lowest level hpx thread

20:51 <jbjnr> but it's also the input type of the continuation

20:51 <hkaiser> shouldn't be

20:52 <hkaiser> jbjnr: do you have a smaller test case?

20:52 <jbjnr> lol

20:53 <jbjnr> I'll see what I can put together. Half asleep after a day out cycling though, so probly won't o it tonight.

20:53 <hkaiser> k

21:08 <github> [hpx] K-ballo opened pull request #2968: Cleaning up dataflow overload set (master...dataflow-api) https://git.io/vdNGj

21:15 <jbjnr> hkaiser: if an executor provides "async_execute(F && f, Ts &&... ts) " for async execution - should it provide a different interface for continuations - or is it the same one?

21:15 <hkaiser> should be the same

21:15 <jbjnr> and traits onw_way, true, two way, true?

21:15 <hkaiser> yah

21:15 <jbjnr> ok

21:15 <jbjnr> ta

21:55 <hkaiser> jbjnr: this compiles for me: https://gist.github.com/hkaiser/930f2dcbd99fb3dfc35cc4e50aafe21b

21:55 <hkaiser> how is your example different?

22:00 <github> [hpx] hkaiser pushed 1 new commit to master: https://git.io/vdNCL

22:00 <github> hpx/master a2d9428 Hartmut Kaiser: Disable error reporting for affinity problems on CircleCI (for the time being)

22:15 <K-ballo> what is HWLOC_MEMBIND_BYNODESET ?

22:16 <hkaiser> place memory close to a socket

22:16 <K-ballo> time to upgrade hwloc?

22:16 <hkaiser> is it failing?

22:17 <hkaiser> (I have not looked in a while :/)

22:17 <K-ballo> my version doesn't have it

22:17 <hkaiser> ahh

22:20 <K-ballo> I was on 1.11.2 that didn't had it, now on .8 that does

22:20 <hkaiser> ohh, THAT new?

22:23 <K-ballo> .3 introduced BYNODESET

22:24 <K-ballo> Jul 27, 2016

22:24 <hkaiser> k

23:03 <K-ballo> I'm getting a few linker errors from jbjnr's branch, could it be missing a few EXPORTS?

23:04 <K-ballo> or maybe having too many..

23:34 <github> [hpx] K-ballo created is_tuple_like (+1 new commit): https://git.io/vdNB1

23:34 <github> hpx/is_tuple_like ed46f93 Agustin K-ballo Berge: Improve is_tuple_like trait...

23:59 <K-ballo> hkaiser: many executors using bind unprotected

23:59 <K-ballo> the binpacking_distribution_policy component too, curiously

23:59 <hkaiser> K-ballo: good point, didn't think about this