06:54 <mdiers[m]> i have a problem with 1.5.0-rc3 on fedora 32. at runtime i get the error: PMIX ERROR: NO-MEM in file client/pmix_client_get.c. maybe a bug in openmpi. how can i change it again that no pmix is used? it's been so long ago ;-(

06:58 <ms[m]> jbjnr: you know anything about pmix? ^

07:30 <mdiers[m]> how can I switch from mpi to tcp communication via command line argument?

07:32 <ms[m]> mdiers: ah, that's a bit more complicated unfortunately (at least if you have mpi implicitly initialized in your environment)

07:33 <ms[m]> in principle --hpx:ini=hpx.parcel.tcp.enable=1 --hpx:ini=hpx.parcel.tcp.enable=0

07:33 <ms[m]> but that most likely currently doesn't work

07:33 <ms[m]> see https://github.com/STEllAR-GROUP/hpx/pull/4924 for a short explanation

07:33 <ms[m]> if we get a fix for that soon we'll have a patch release containing the fix

08:14 <mdiers[m]> I think it should be ```--hpx:ini=hpx.parcel.tcp.enable=1 --hpx:ini=hpx.parcel.mpi.enable=0```. But unfortunately it does not work as already described. I just suspected our slurm openmpi combination that something does not fit. With mpirun it works.

08:46 <rori> diehlpk_work: sure sorry for the trouble :/

08:49 <ms[m]> mdiers: yep, sorry, typo... yeah, sounds indeed like something in the environment, I wouldn't know what to do with that

10:17 <mdiers[m]> <mdiers[m] "i have a problem with 1.5.0-rc3 "> found it, there was a missing name resolution in the container. the error messages are still there but it works fine again.

10:51 <kordejong> I am using HPX 1.5.0 icw APEX+OTF2. When visualizing the trace with Vampir, actions show up, which is cool! But I cannot seem to get any custom tasks passed to `dataflow` or as a continuation to show up in the trace. I annotate tasks with `util::annotated_function(my_function, my_name)`. I must be doing something wrong. But what? Any hints maybe?

10:55 <kordejong> Never mind. Wrong ordering `unwrapping` and `annotated_function`. Funny how asking questions itself can lead to answers ;-)

10:59 <jbjnr> kordejong, you might need to use the annotate_function wrapper as the annotated_function RAII thingy doesn't work any more (I thought we got rid of it)

11:00 <jbjnr> note difference between annotate_function and annotated_function - you need the one that wraps the task which is more painful to use, but should work.

12:12 <kordejong> jbjnr: Aha, yes, `annotate_function` works also, thanks. Actually, I find it easier to use than `annotated_function`. I just instantiate it at the start of the tasks. Is that how it is supposed to be used? Are there any other considerations about using `annotate_function`, like a noticeable runtime cost?

12:32 <ms[m]> Kor de Jong: the way it's implemented right now means that doing an `annotate_function` has to yield the task to set the name

12:32 <ms[m]> depending on your use case it may have noticeable overhead

12:34 <ms[m]> very relevant for hpx/kokkos, but I haven't looked if there's a program yet

12:34 <kordejong> jbjnr: OK thanks, I will then conditionally enable it. It works great BTW.

13:27 <ms[m]> gdaiss, hkaiser kokkos meeting as usual? joining in a few minutes

13:28 <ms[m]> rori, jbjnr as well

13:33 <diehlpk_work> rori, No worries, we were able to fix things and the paper is released

13:34 <diehlpk_work> https://joss.theoj.org/papers/10.21105/joss.02352

13:34 <diehlpk_work> ms[m], The DOI is available. Can you might add the paper's DOI to the readme and the documentation>

13:37 <hkaiser> diehlpk_work: as said, I'll take care of the readme

13:38 <rori> diehlpk_work: thanks!

17:09 <K-ballo> do I still have a rostam account? what's my username?

17:10 <parsa> akheir: see ^

17:12 <akheir> K-ballo: I don't think so, send me an email I will create yours in a bit. aliz@cct.lsu.edu

17:14 <K-ballo> akheir: sent

17:19 <akheir> got it

17:40 <K-ballo> "Password change failed. Server message: Old password not accepted." lol

17:42 <K-ballo> akheir: actually, I get "old password not accepted" regardless of what I write

17:50 <akheir> K-ballo, the password change menu is a bit stupid, it ask for current password twice, let me reset your password

17:52 <akheir> K-ballo: see pm pls.

17:54 <K-ballo> confusing menu, makes sense now

18:44 <K-ballo> "TBB has some of the algorithms implemented, AFAIK not all of them. HPX implements all (except for three of them)"

18:45 <zatumil> hey, my installation directory /usr/local/include/hpx/include looks weird, can the second include directory stripped from path? I would like to #include <hpx/async.h>

18:50 <K-ballo> zatumil: looks right to me..

18:51 <K-ballo> hpx/async.hpp is not the same as hpx/include/async.hpp

19:04 <zatumil> oh I didnt know hpx/async.hpp exists too

19:06 <K-ballo> which one did you want?

19:06 <K-ballo> note there is no async.h anywhere

19:30 <zatumil> I think async_local/apply.hpp is fine for the hello world example, the #include <hpx/include/async.hpp> was confusing me, I would expect either #include <hpx/distributed/async.hpp> or #include <hpx/local/async.hpp>

19:31 <K-ballo> #include <hpx/include/async.hpp> sounds fine for an example

19:31 <zatumil> include looks redundant to me

19:32 <K-ballo> these are user facing headers who's job is to include a bunch of internal pieces

19:32 <zao> The subtree is for friendly user-facing includes, not getting you lost in the sea of implementation things in the full tree.

19:32 <K-ballo> you should be picking stuff from hpx/include, unless you really really know what you are doing

19:32 <zao> It's a convention of some header-heavy libraries, Spirit has a "home" IIRC.

19:34 <zatumil> maybe I can -I /usr/local/hpx/include and do #include <async.hpp> then

19:35 <zao> Please don't.

19:35 <K-ballo> ouch

19:36 <zao> Everything expects that the top-level dir(s) are included so that top-relative paths works.

19:40 <ms[m]> zatumil: for everything that you can find on this page, prefer the headers in the headings: https://hpx-docs.stellar-group.org/tags/1.5.0/html/api/public_api.html

19:40 <ms[m]> everything else requires a bit of guesswork, but the hpx/include/ headers give you access to all the rest

19:42 <ms[m]> that's if you're on 1.5.0...

19:46 <zatumil> the document is good, how to make docs/html from git?

20:12 <zao> zatumil: There's a build target for it (docs?) assuming you have the tools installed (doxygen, sphinx?)

20:23 <zatumil> yes with cmake -DHPX_WITH_DOCUMENTATION=ON

20:23 <zatumil> thank you