Make stdout and stderr async.

Make stdout and stderr async.

This required some minor changes to the Windows eventhandler, as we now spawn a new thread for writes (just like we do for reads).

BUG=https://code.google.com/p/dart/issues/detail?id=7218
R=sgjesse@google.com

Committed: https://code.google.com/p/dart/source/detail?r=27811

[Søren Gjesse, 2013-09-24 11:24:22]

lgtm

https://codereview.chromium.org/23609060/diff/1/runtime/bin/eventhandler_win.cc
File runtime/bin/eventhandler_win.cc (right):

https://codereview.chromium.org/23609060/diff/1/runtime/bin/eventhandler_win....
runtime/bin/eventhandler_win.cc:597: int Handle::Write(const void* buffer, int
num_bytes) {
This concept is pretty nasty. Please put in a comment saying that this is only
used for stdout and stderr, and the RawSocket for these are not exposed to user
code.

https://codereview.chromium.org/23609060/diff/1/runtime/bin/eventhandler_win....
runtime/bin/eventhandler_win.cc:618: _beginthreadex(NULL, 32 * 1024,
WriteFileThread, this, 0, &tid);
Please open an issue on not starting a new thread for each write to
stdout/stderr.

https://codereview.chromium.org/23609060/diff/1/runtime/bin/socket_win.cc
File runtime/bin/socket_win.cc (right):

https://codereview.chromium.org/23609060/diff/1/runtime/bin/socket_win.cc#new...
runtime/bin/socket_win.cc:204:
file_handle->EnsureInitialized(EventHandler::delegate());
How about getting rid of EnsureInitialized completely?

We can add a static on EventhandlerImplementation with the instance. There is
only one and it needs to be created before anything happens. I think the
event_handler_ member can be removed from the Handle class and we can just use
EventHandlerImplementation::instance_.

You can do that in a separate cl.

https://codereview.chromium.org/23609060/diff/1/sdk/lib/_internal/pub/lib/src...
File sdk/lib/_internal/pub/lib/src/exit_codes.dart (right):

https://codereview.chromium.org/23609060/diff/1/sdk/lib/_internal/pub/lib/src...
sdk/lib/_internal/pub/lib/src/exit_codes.dart:13: import 'dart:async';
Accidental edit?

[Anders Johnsen, 2013-09-24 11:41:56]

https://codereview.chromium.org/23609060/diff/1/runtime/bin/eventhandler_win.cc
File runtime/bin/eventhandler_win.cc (right):

https://codereview.chromium.org/23609060/diff/1/runtime/bin/eventhandler_win....
runtime/bin/eventhandler_win.cc:597: int Handle::Write(const void* buffer, int
num_bytes) {
On 2013/09/24 11:24:22, Søren Gjesse wrote:
> This concept is pretty nasty. Please put in a comment saying that this is only
> used for stdout and stderr, and the RawSocket for these are not exposed to
user
> code.
> 

Done.

https://codereview.chromium.org/23609060/diff/1/runtime/bin/eventhandler_win....
runtime/bin/eventhandler_win.cc:618: _beginthreadex(NULL, 32 * 1024,
WriteFileThread, this, 0, &tid);
On 2013/09/24 11:24:22, Søren Gjesse wrote:
> Please open an issue on not starting a new thread for each write to
> stdout/stderr.

Done.

https://codereview.chromium.org/23609060/diff/1/runtime/bin/socket_win.cc
File runtime/bin/socket_win.cc (right):

https://codereview.chromium.org/23609060/diff/1/runtime/bin/socket_win.cc#new...
runtime/bin/socket_win.cc:204:
file_handle->EnsureInitialized(EventHandler::delegate());
On 2013/09/24 11:24:22, Søren Gjesse wrote:
> How about getting rid of EnsureInitialized completely?
> 
> We can add a static on EventhandlerImplementation with the instance. There is
> only one and it needs to be created before anything happens. I think the
> event_handler_ member can be removed from the Handle class and we can just use
> EventHandlerImplementation::instance_.
> 
> You can do that in a separate cl.

Will do, it would be a nice cleanup.

https://codereview.chromium.org/23609060/diff/1/sdk/lib/_internal/pub/lib/src...
File sdk/lib/_internal/pub/lib/src/exit_codes.dart (right):

https://codereview.chromium.org/23609060/diff/1/sdk/lib/_internal/pub/lib/src...
sdk/lib/_internal/pub/lib/src/exit_codes.dart:13: import 'dart:async';
On 2013/09/24 11:24:22, Søren Gjesse wrote:
> Accidental edit?

Done.

[Anders Johnsen, 2013-09-24 11:48:37]

Committed patchset #2 manually as r27811 (presubmit successful).