14:50 <jbjnr> K-ballo: is it possible to instantiate a specialization - only once - say you have a macro and it instantiates a struct member specialized on T, then someone uses the macro and all is fine, but if they use it twice with the same type, it ignore the second one.

14:52 <K-ballo> specializations are already only instantiated once per set of arguments (per TU), jbjnr what are you trying to do?

14:52 <jbjnr> a member of a struct is static, but must only be declared once

14:53 <jbjnr> template <>

14:53 <jbjnr> memory_region_allocator<unsigned char>::mempool_ptr = nullptr;

14:53 <jbjnr> memory_region_allocator<unsigned char>::mempool_type*

14:54 <jbjnr> if the user uses a macro blah_blah(unsigned char) twice, it doesn't try to redeclare the member

14:54 <K-ballo> the macro shouldn't be doing the declaration

14:54 <jbjnr> well, macros shouldn't ever do anything, but we use them for things like action declarations etc

14:55 <K-ballo> why would the macro be the one doing the declaration?

14:55 <K-ballo> memory_region_allocator should be the one doing the declaration

14:55 <K-ballo> template <typename T> mem....

14:56 <K-ballo> you still get one per instantiation, which is one per type

14:56 <K-ballo> if there's a reason for the macro to be doing the declaration, it is not captured by the example snippet

14:58 <jbjnr> I have one of these

14:58 <jbjnr> static mempool_type* mempool_ptr;

14:58 <jbjnr> but I should make it type independent then the problem goes away (allocator<T> uses one and do there needs to be one memool_ptr for each type instantiated, but if the user uses allocator<unknown type> we need to instantiate a mempool_ptr for it

15:00 <K-ballo> one instance per instantiated type, no macros:

15:00 <K-ballo> template <typename T> struct foo { static X x; };

15:00 <K-ballo> template <typename T> X foo<T>::x;

15:03 <jbjnr> the point is that I want to put all that inside a macro for something else and don't want it to declare template <typename T> X foo<T>::x; twice is the user puts it in twice. I will type erasse the pool ptr and the problem should go away. I think I should have done that originally, will make it much simpler to manage.

15:04 <K-ballo> you can't possibly put all tha tinside a macro, you'd be redeclaring foo itself

15:04 <jbjnr> I probably didn't explain the problem properly.

15:05 <K-ballo> template <typename T> X foo<T>::x is only ever declared once, and then you get a different foo<T>::x for each T

15:07 <jbjnr> The declaration isn'tt the problem, it's the instantiation of X foo<T>::x; for arbitrary T

15:08 <K-ballo> you don't want different foo<T>::x instances?

15:19 <jbjnr> it wants one for every T, but I want to make sure you can't ever do it twice for the same T, however, I realize after sayng it out loud that I should actually just have one pool ffor all T as I can cast the memory to any T anyway. Thanks for your suggestions.

15:20 <K-ballo> I don't follow why one would provide specializations to begin with, there's no need for them, the linker is capable of producing those itself from a base definition

15:23 <K-ballo> if the linker was asking for them, then there must have not been a base definition.. it was just a matter of providing that base definition

15:44 <K-ballo> reducing type dependencies when possible is always good, and less symbols too

18:03 <pedro_barbosa[m]> Hey guys, I'm getting the following error when compiling a program (GIST: https://gist.github.com/PADBarbosa/82a1ceb3cad16ffd5fbb6d12e5d670f4). Anyone knows how to solve it? Do I need to add a flag when building it? I used a simple CMake file (GIST:https://gist.github.com/PADBarbosa/07318f62da55180537fa7fb993c67a7c).

18:07 <ms[m]> I guess that question is for diehlpk_work ^

18:08 <ms[m]> pedro_barbosa: that second gist is possibly private?

18:10 <pedro_barbosa[m]> yeah sorry accidentally created a provate

18:10 <pedro_barbosa[m]> https://gist.github.com/PADBarbosa/db018246fe46475d2fb9538189803907

18:10 <pedro_barbosa[m]> * yeah sorry accidentally created a private

18:10 <pedro_barbosa[m]> * Hey guys, I'm getting the following error when compiling a program (GIST: https://gist.github.com/PADBarbosa/82a1ceb3cad16ffd5fbb6d12e5d670f4). Anyone knows how to solve it? Do I need to add a flag when building it? I used a simple CMake file (GIST:https://gist.github.com/PADBarbosa/db018246fe46475d2fb9538189803907).

18:12 <K-ballo> gists can't be private

18:13 <K-ballo> they can be "secret", which means unlisted, but anyone with the url can still get at them

18:13 <pedro_barbosa[m]> indeed, for a second thought it meant something else, sent a new link though, should work this time

19:09 <ms[m]> pedro_barbosa: you're not linking to hpxcl

19:10 <pedro_barbosa[m]> yeah I believe that's the problem but I tried adding a flag and it said it wasn't being used

19:11 <ms[m]> What flag? Hpxcl is not part of hpx

19:13 <pedro_barbosa[m]> something like "DHPXCL_ROOT=.." with the directory to the HPXCL-master folder

19:13 <pedro_barbosa[m]> I once had a similar problem with HPX and solved it by installing it on the default path, but not sure I can do that to solve this problem

19:17 <ms[m]> You also don't do find_package(hpxcl) so there's not much chance of it finding it (though I don't know what the status of hpxcl's cmake setup is, I saw there was some activity related to that which maybe makes it work if it's broken right now)

19:25 <pedro_barbosa[m]> I'm not entirely sure how to do it with HPXCL, it's my first time using it and was trying to compile a program outside the examples to see if I could do it

19:39 <ms[m]> ok, as I said I don't know much about hpxcl, diehlpk_work is probably most knowledgeable about it