06:25 <matrixb01x> I had a question redarding making python bindings for HPX . where would one start to work from .  or just picking algorithm and do it one by one ?

07:02 <gonidelis[m]> matrixb01x: just pick an algo. i would suggest you find a tool that binds python to C++ code

07:09 <matrixb01x> i am thinking of using pybind11 . but i have a question  if we just have to access the algorithm we can make use of ctypes to access shared files .so

07:10 <gonidelis[m]> a student try to work on this project last year and he used pybind11 successfully so that's a good choice

07:11 <gonidelis[m]> what is the question?

07:11 <satacker[m]> This is also a nice tool https://github.com/Tolc-Software/tolc

07:11 <matrixb01x> why cant we just use the .so files fro ctypes library

07:12 <satacker[m]> We can ig

07:13 <satacker[m]> But those are not C based shared libraries

07:13 <satacker[m]> So I am not aware how you'll be able to write a wrapper around with ctypes

07:16 <satacker[m]> I think you'll have to provide a C interface beforehand

07:17 <satacker[m]> It'll be tedious to do so.

15:00 <Yorlik> hkaiser: I assume functions returning incomplete types can't be actions, right?

15:01 <hkaiser> well, you should be able to define the action, but you will not be able to use it (with async, etc.)

15:02 <K-ballo> what does that mean? functions can't return incomplete types

15:05 <Yorlik> K-Ballo: I am just forward declaring the type in a header for a struct, but the type is used only in the .cpp file

15:06 <Yorlik> This one: https://gist.github.com/McKillroy/e51ea7abfa1decf3081b57fecaad7ed4

15:08 <Yorlik> The problem is dragonspeak is header only and I quickly run into double symbols when linking.

15:10 <K-ballo> sounds like it's not a proper header only library then

15:10 <K-ballo> shared_ptr<T> is a complete type, independent of T

15:10 <K-ballo> and you should be able to declare the serialization for T in the header, define it in the source

15:11 <K-ballo> I don't know if that would be enough though

15:14 <Yorlik> K-Ballo: I'm not sure what you mean with a proper header only library here. Also I still feel quite wonky with all these HPX macros. Trying to use them as I do now, gives me static asserts.

15:15 <Yorlik> I'm totally not happy with that current construct tbh. E.g. having the server config in a function that stores it as a static variable ... looks a bit like hack to me.

15:16 <Yorlik> Maybe I should write some proper wrapper type I can expose and use.

15:17 <K-ballo> a header only library doesn't result in double symbols when linking

15:18 <K-ballo> if it does, it's broken

15:18 <K-ballo> header only libraries can be included from 2 or more cpp files

15:20 <Yorlik> I think I need to learn something here. Are there good guidelines to achieve that ?

15:28 <Yorlik> I guess one problem we have is, that our library is using a static library under the hood that is not header only. So maybe the entire idea is just wrong.

15:32 <Yorlik> I'm also spotting some issues: We have some globals in the library. I guess I need to get rid of those.

17:43 <HHN93> hkaiser in some places in the core library we use stl algorithms instead of hpx algorithms, Example : std::sort instead of hpx::sort

17:43 <HHN93> is there any reason for it?

17:43 <HHN93> and often these std algorithms run in sequential mode (no exec policy mentioned), any reason for that?

17:48 <HHN93> Also, for algorithms which can't be vectorized, can we ignore them?

17:55 <srinivasyadav18[> HHN93 : std algorithms usually used only in sequential cases, i.e they are used only at places where hpx::<algo> runs in sequential

17:57 <srinivasyadav18[> it might be just to not re-implement the basic version of algorithm again, so we can just use std version of algorithm, (but not in all cases). For instance, if any algorithm can be implemented using util::loop, we use the util::loop implementation even for the sequential overloads, because util::loop in unrolled, hence may give better performance.

19:03 <gonidelis[m]> Second release candidate for 1.9.0 is out: https://github.com/STEllAR-GROUP/hpx/releases/tag/v1.9.0-rc2. Have fun :D

19:08 <hkaiser> gonidelis[m]: many thanks!

19:09 <hkaiser> everybody - please use the rc2 to run your codes