Make mojo::RunLoop just use pthreads.

Make mojo::RunLoop just use pthreads.

In particular, it just uses pthreads TLS stuff directly along with
pthread_once(), which allows RunLoop::{SetUp,TearDown}() to be removed.

This will allow us to make the "standalone" Environment like the
"Chromium"/base-using Environment, i.e., not require instantiation. This
will make things like ApplicationRunner more sane (since it currently
has to mix two things: instantiating the Environment -- which you need
to do once per "process", i.e., Mojo application binary instantiation --
and instantiating a RunLoop -- which you need to do once per thread).

(Really, ApplicationRunner should just be a function, but that's another
story.)

Also note that I can't just use C++11 TLS (I tried:
1c4b90d52b54811cfeeb7d72fd63ee3c33006866), since it's not supported on
iOS (which made Flutter sad). pthreads is OK though (including, AFAICT,
pthread_once()).

R=vardhan@google.com

Committed: https://chromium.googlesource.com/external/mojo/+/c3b1910166bbcd211901700ead52abc3e47934ec

[vardhan, 2016-05-19 21:37:35]

lgtm

https://codereview.chromium.org/1996703002/diff/1/mojo/public/cpp/utility/lib...
File mojo/public/cpp/utility/lib/run_loop.cc (right):

https://codereview.chromium.org/1996703002/diff/1/mojo/public/cpp/utility/lib...
mojo/public/cpp/utility/lib/run_loop.cc:30: int error =
pthread_key_create(&g_current_run_loop_key, nullptr);
does it make sense to provide a destructor to pthread_key_create() that tears
down the current runloop?

i guess the current thread's RunLoop will always be destroyed before the thread
is destroyed (unless RunLoop is dynamically allocated or something and
mismanaged or something).

[viettrungluu, 2016-05-19 21:46:29]

On 2016/05/19 21:37:35, vardhan wrote:
> lgtm
> 
>
https://codereview.chromium.org/1996703002/diff/1/mojo/public/cpp/utility/lib...
> File mojo/public/cpp/utility/lib/run_loop.cc (right):
> 
>
https://codereview.chromium.org/1996703002/diff/1/mojo/public/cpp/utility/lib...
> mojo/public/cpp/utility/lib/run_loop.cc:30: int error =
> pthread_key_create(&g_current_run_loop_key, nullptr);
> does it make sense to provide a destructor to pthread_key_create() that tears
> down the current runloop?
> 
> i guess the current thread's RunLoop will always be destroyed before the
thread
> is destroyed (unless RunLoop is dynamically allocated or something and
> mismanaged or something).

Yeah. (I imagine that if someone foolishly did pthread_exit(), the RunLoop
wouldn't be destroyed either.) But there's no canonical way to tear down the run
loop from such a destructor, so there's not much we can do.

[viettrungluu, 2016-05-19 21:57:02]

Description was changed from

==========
Make mojo::RunLoop just use pthreads.

In particular, it just uses pthreads TLS stuff directly along with
pthread_once(), which allows RunLoop::{SetUp,TearDown}() to be removed.

This will allow us to make the "standalone" Environment like the
"Chromium"/base-using Environment, i.e., not require instantiation. This
will make things like ApplicationRunner more sane (since it currently
has to mix two things: instantiating the Environment -- which you need
to do once per "process", i.e., Mojo application binary instantiation --
and instantiating a RunLoop -- which you need to do once per thread).

(Really, ApplicationRunner should just be a function, but that's another
story.)

Also note that I can't just use C++11 TLS (I tried:
1c4b90d52b54811cfeeb7d72fd63ee3c33006866), since it's not supported on
iOS (which made Flutter sad). pthreads is OK though (including, AFAICT,
pthread_once()).

R=vardhan@google.com
==========

to

==========
Make mojo::RunLoop just use pthreads.

In particular, it just uses pthreads TLS stuff directly along with
pthread_once(), which allows RunLoop::{SetUp,TearDown}() to be removed.

This will allow us to make the "standalone" Environment like the
"Chromium"/base-using Environment, i.e., not require instantiation. This
will make things like ApplicationRunner more sane (since it currently
has to mix two things: instantiating the Environment -- which you need
to do once per "process", i.e., Mojo application binary instantiation --
and instantiating a RunLoop -- which you need to do once per thread).

(Really, ApplicationRunner should just be a function, but that's another
story.)

Also note that I can't just use C++11 TLS (I tried:
1c4b90d52b54811cfeeb7d72fd63ee3c33006866), since it's not supported on
iOS (which made Flutter sad). pthreads is OK though (including, AFAICT,
pthread_once()).

R=vardhan@google.com

Committed:
https://chromium.googlesource.com/external/mojo/+/c3b1910166bbcd211901700ead5...
==========

[viettrungluu, 2016-05-19 21:57:04]

Committed patchset #1 (id:1) manually as
c3b1910166bbcd211901700ead52abc3e47934ec (presubmit successful).