Ok great, this is what I had in mind. The two main issues:

1. proc.hpp includes os.hpp, so making os.hpp include proc.hpp introduces a
circular dependency, no?

2. As you mentioned, ProcessStatus is linux specific for the most part.
We'll either need to make it generic, or #ifdef it accordingly. What did
you mean by implement things like children differently?


On Thu, Apr 25, 2013 at 1:53 PM, Benjamin Hindman <
[email protected]> wrote:

> I think we want "everything" (pids, children, alive, etc.) in os:: with
> implementations that use the proc:: functions for Linux and do other things
> for Mac OS X. As in, os::pids can be implemented for Linux as a call to
> proc::pids, while the implementation for Mac OS X can be whatever else it
> needs to be. The trick here is that ProcessStatus is a pretty specific
> Linux abstraction, so either we need to make a generic one or we need to
> implement things like children differently.
>
>
> On Thu, Apr 25, 2013 at 1:41 PM, Benjamin Mahler <
> [email protected]> wrote:
>
>> Ok so linux/proc.hpp currently has:
>>
>> Try<std::set<pid_t> > pids();
>> Try<std::set<pid_t> > children(pid_t pid, bool recursive = true);
>> Try<ProcessStatus> status(pid_t pid);
>>
>> We want these to work for OSX as well for now.
>>
>> Say we move linux/proc.{cpp,hpp} into stout/proc.hpp, then we can move
>> alive into os and fail the compilation if anyone includes stout/proc.hpp
>> without __linux__ defined.
>>
>> We'll also want os::pids(), os::children(pid_t) and os::status(pid_t) for
>> non-linux systems. I originally wanted to have these call into
>> stout/proc.hpp for linux, but that introduces a circular dependency.
>>
>>
>> On Thu, Apr 25, 2013 at 11:51 AM, Benjamin Hindman <
>> [email protected]> wrote:
>>
>>> I like os:: a lot. I think the type signature (i.e., taking a pid_t) is
>>> sufficient for disambiguation.
>>>
>>>
>>>
>>>
>>>
>>> On Apr 25, 2013, at 11:43 AM, Benjamin Mahler <[email protected]>
>>> wrote:
>>>
>>> We also have a src/common/process_utils.hpp which contains only
>>> mesos::internal::utils::process::killtree() at the moment.
>>>
>>>
>>> On Thu, Apr 25, 2013 at 11:37 AM, Yan Xu <[email protected]> wrote:
>>>
>>>> I guess os:: is fine, but in a separate file?
>>>>
>>>> --
>>>> Jiang Yan Xu <[email protected]> @xujyan <http://twitter.com/xujyan>
>>>>
>>>>
>>>> On Thu, Apr 25, 2013 at 11:29 AM, Vinod Kone <[email protected]> wrote:
>>>>
>>>>> I don't like process:: because it conflicts with the libprocess
>>>>> namespace as you mentioned.
>>>>>
>>>>> I still like proc:: but clearly BenH doesn't like it. I'm ok with os::
>>>>> namespace.
>>>>>
>>>>>
>>>>> @vinodkone
>>>>>
>>>>>
>>>>> On Thu, Apr 25, 2013 at 11:19 AM, Benjamin Mahler <
>>>>> [email protected]> wrote:
>>>>>
>>>>>> Is there any consensus on how to place process utilities in stout? I
>>>>>> would expect this to be in a process:: namespace but of course that is
>>>>>> confusing because we use libprocess, which should perhaps have a
>>>>>> libprocess:: namespace instead..
>>>>>>
>>>>>> I'll be moving process utilities etc into stout, hopefully with the
>>>>>> same calls for linux and OSX but I'm not yet certain if that is 
>>>>>> possible. I
>>>>>> would like to place these in a process.hpp file inside a process::
>>>>>> namespace.
>>>>>>
>>>>>> I think these read very nicely:
>>>>>> process::alive(pid_t)
>>>>>> process::children(pid_t)
>>>>>> process::stat(pid_t)
>>>>>>
>>>>>> Thoughts?
>>>>>>
>>>>>>
>>>>>> On Tue, Apr 23, 2013 at 6:29 PM, Yan Xu <[email protected]> wrote:
>>>>>>
>>>>>>> This batch of commits changed the reaper to use "Future" as the
>>>>>>> notification mechanism.
>>>>>>>
>>>>>>> Sequence:
>>>>>>> https://reviews.apache.org/r/10744/
>>>>>>> https://reviews.apache.org/r/10745/
>>>>>>> https://reviews.apache.org/r/10746/
>>>>>>> https://reviews.apache.org/r/10747/
>>>>>>>
>>>>>>> Best,
>>>>>>> Yan
>>>>>>> --
>>>>>>> Jiang Yan Xu <[email protected]> @xujyan <http://twitter.com/xujyan>
>>>>>>>
>>>>>>
>>>>>>
>>>>>
>>>>
>>>
>>
>

Reply via email to