Boost logo

Boost :

Subject: Re: [boost] [GSoC] pipelines questions
From: Vicente J. Botet Escriba (vicente.botet_at_[hidden])
Date: 2014-03-14 04:06:26


Le 13/03/14 23:06, Thaler Benedek a écrit :
>> I'm interested in the Boost.Pipelines proposal of GSoC 2014.
> [snip]
>
> My proposal of proposal can be found online: [0]. Please take a look and
> notify me if you miss something or if you have any general comment.
>
Hi,

Your proposal is almost perfect. You have been able to implement a POC
and generate the documentation using the Boost environment tools. You
seem to master the problem domain.

Concerning the project proposal, I would like to see the following
points more elaborated:
* Why do you think that value semantic would be better?
* How would you specify the queue as a policy parameter?
* Do you plan to make the thread pool configurable?
* Do you plan to port it to C++98 compilers?
* What your performance tests will measure? Do you plan to compare
different implementations? Value-semantics versus shared semantics?
Different queues? Different thread pools?

Concerning the millstones, I would prefer a week based schedule and that
as far as possible each iteration includes something adding value with
the code the tests and the documentation. Maybe enumerating the use
cases by priority would help.
* 1-1 transformations
* 1-N transformations
* N-M transformations
* parallel segments
* make use of thread pools
* queue configuration

On each of these use cases maybe you could list the classes that you
will need to work on (roughly). Indication for each one which use case
will be a must/should/nice to have would help you to make a better plan.
Indicating also which performance test could be done at each step would
also be useful. Note that a plan is there to help you not constraint you.

Let me know if you want I help you on this.

Best,
Vicente


Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk