Re: [Boost-users] [C++ Now! 2012] Call for Submissions

Subject: Re: [Boost-users] [C++ Now! 2012] Call for Submissions
From: Boris Schaeling (boris_at_[hidden])
Date: 2011-11-10 14:52:15

On Thu, 10 Nov 2011 16:12:34 +0100, Nat Linden <nat_at_[hidden]> wrote:

> [...]Hmm, okay. That isn't specifically interesting to me. Full  
> disclosure:
> I want Boost.Process to provide a portable way to run a child process,
> both synchronously and asynchronously. (I personally would be content
> with a conceptual model resembling that of Python's subprocess API.)
> That is, to me, one of the primary benefits Boost can contribute in
> this space. Those who need platform-specific process-control features
> can (and do) use the platform-specific API. That support exists
> already. What's missing is a portable wrapper layer.

If I try to explain the problem we had in the past in a few sentences:  
Boost.Process has to be somewhere between the two extremes std::system()  
and a platform-specific API. No matter what Boost.Process is closer to -  
people will complain. The solution seems to be to focus more on  
extensibility. That's what Jeff Flinn proposed with what he calls executor  
concept. I'm not sure where to point you to though if you want to read  
more about it (there were quite a lot of discussions in the devel mailing  
list in spring).

> However -- that said -- if folding in asio::windows::object_handle
> allows others to come to consensus and SHIP THIS LIBRARY then I'm all
> for it.

The Boost.Process draft 0.4 was criticized for how it supported waiting  
for processes asynchronously. The object_handle is now a much better  
solution which I expect will be shipped with Boost.Asio (depends on Chris;  
he knows about the extension though and reviewed an early implementation a  
few months ago). This problem should then be gone. :)

> [...]Whatever work you can put into evolving Boost.Process between  
> January
> and May will be wonderful. But that's not even what I'm requesting.
> What I'd really like from you this May isn't so much a presentation,
> but rather chairing a working session to nail down any
> unresolved/controversial issues. I would be thrilled if, at the close
> of this inaugural C++ Now! conference, all that remained unfinished
> with Boost.Process was the implementation. I'd want us to emerge with
> a solid API ready for the Boost review process, as soon as the
> implementation is done.

I think that's we did at BoostCon 2011. With Jeff's executor concept we  
have an idea lots of people agreed with to give it another try. We wanted  
to work on the code after BoostCon 2011 but didn't do much (at least not  
at <https://github.com/JeffFlinn/boost-process>; I managed to create the  
object_handle at least). So I think it's definitely a good idea to do a  
session on Boost.Process again. But I also think we (or I :) should work  
on Boost.Process code before May to be able to present something new. :)

Boris