Subject: Re: [boost] [Boost.Sync] no master branch?
From: Andrey Semashev (andrey.semashev_at_[hidden])
Date: 2014-01-16 12:44:22
On Thu, Jan 16, 2014 at 8:50 PM, Eric Niebler <eniebler_at_[hidden]> wrote:
> On 01/15/2014 02:06 PM, Andrey Semashev wrote:
>> ... decoupling from the time units is desirable because (a) in many
>> cases you don't need time units with synchronization yet you still
>> introduce dependency on Boost.Chrono and Boost.DateTime
>
> But in many cases you *do* care about time (e.g. wait with a time-out).
Yes, but why should these cases impose a mandatory dependency? I don't
want to pay for that when e.g. I just need a mutex to synchronize
access.
> How will you present a unified, consistent interface when you can't
> express some fundamental operations?
>
>> and (b) there is no way to use use e.g. std::chrono with
>> Boost.Thread without manually converting to Boost time units.
>
> This sounds like a failure of generic design to me. There should be a
> Duration concept, and the APIs should work with any concrete type that
> models that concept. That would also solve the dependency problem, since
> the actual concrete types don't need to be mentioned.
It's a little more complicated than that because Boost.DateTime and
Boost.Chrono implement different interfaces (as well as Boost.Thread
wrt different time units). But basically, that's what "decoupling"
means and what Boost.Sync does.