Subject: Re: [boost] [Boost-users] Attn: We need GSoC 2014 mentors for Boost!
From: Vicente J. Botet Escriba (vicente.botet_at_[hidden])
Date: 2014-02-09 11:24:50
Le 09/02/14 16:47, Hartmut Kaiser a écrit :
>>>> If none suit you, would you have a GSoC project proposal to mind?
>>> I'd be happy to mentor any of the projects related to parallelism or
>>> concurrency.
>> Ok, I'll add you as a potential mentor to any project I think relates to
>> parallelism or concurrency. Be aware you may be surprised at what I think
>> you interested in :)
> Thanks Niall.
>
> Some comments:
>
> Boost.Functional / Invoke: This has already been fully implemented by
> Agustin Berge here:
> https://github.com/STEllAR-GROUP/hpx/blob/master/hpx/util/invoke.hpp. Feel
> free to use.
There are some implementations of invoke by there, even one in
Boost.Thread. My intention is that the project result in a new library
in Boost that can be used by the user and other libraries in Boost. I
want also that an invoker class be added and take in account a possible
invocation_trait in Boost.Thread. Thanks for the offer.
> Boost.Thread / Work-Stealing-Thread-Pool: This is something already
> implemented as part of HPX (https://github.com/STEllAR-GROUP/hpx), we have
> several work-stealing implementations, FWIW. Feel free to use.
Again thanks. The intent is to have this in Boost.
>
> Boost.Thread / Scheduler-Executor: Note that there is a newer document
> (D3904, sorry I have no direct link), which will be discussed next week in
> Issaquah and I expect for it to be changed significantly.
I'm aware of it. I have proposed on the std-proposal ML to turn the
dynamic polymorphism on a static polymorphism, but I have not received
any comments from the authors. IMO, the time scheduled operations are
orthogonal to the executor and so I propose to have an adaptor that adds
these operations to any model of Executor.
>
> Boost.Thread / Parallel algorithms: Should that include N3554:
> http://www.open-std.org/JTC1/SC22/WG21/docs/papers/2013/n3554.pdf?
It could, of course. At least the std::par functions, but I prefer to
replace the execution_policy by an Executor. Specializations for a
specific executor would always be possible.
I've added the reference on the wiki project.
Best,
Vicente