New experimental Pepper device API....

New experimental Pepper device API.
- makes device contexts opaque to the plugin
- can get / set multiple attributes and flush with a single call (and underlying IPC message exchange)
- currently works in parallel with old API
- adapted pepper test plugin to use new API if use_new_npdevice_api=1

TEST=trybots, visual confirmation that pepper test plugin works with new API
BUG=none



Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=44840

[apatrick_chromium, 2010-03-30 23:47:41]

This still needs quite a bit of work. I would appreciate early feedback. I've
only done pepper 3D.

[darin (slow to review), 2010-03-31 21:58:01]

I just reviewed npapi_extensions.h:

http://codereview.chromium.org/1529005/diff/1/6
File third_party/npapi/bindings/npapi_extensions.h (right):

http://codereview.chromium.org/1529005/diff/1/6#newcode178
third_party/npapi/bindings/npapi_extensions.h:178: typedef
NPError(*NPDeviceGetNumConfigsPtr)(NPP instance,
nit: space after NPError

http://codereview.chromium.org/1529005/diff/1/6#newcode229
third_party/npapi/bindings/npapi_extensions.h:229: typedef NPError
(*NPDeviceRegisterCallbackPtr)(
i really like that you made this a separate function!  i'm curious how the setup
would work for audio.  how do we put the audio device in the mode where it runs
our callback on a background thread?

per our discussion offline, i agree it makes sense to initialize the background
audio thread if the callback_type indicates an audio callback.

http://codereview.chromium.org/1529005/diff/1/6#newcode241
third_party/npapi/bindings/npapi_extensions.h:241: /*  outputAttribList: The
output attribute / value pair array that was        */
the function signature doesn't have an outputAttribList parameter

http://codereview.chromium.org/1529005/diff/1/6#newcode328
third_party/npapi/bindings/npapi_extensions.h:328: NPCallback3D_Repaint         
        = 1
it'd be good to add some documentation for this callback

[apatrick_chromium, 2010-03-31 22:48:52]

All issues addressed. I made some simplifications as well.

[darin (slow to review), 2010-03-31 23:43:16]

API looks good to me so far.

http://codereview.chromium.org/1529005/diff/18001/15006
File third_party/npapi/bindings/npapi_extensions.h (right):

http://codereview.chromium.org/1529005/diff/18001/15006#newcode235
third_party/npapi/bindings/npapi_extensions.h:235: int32 callback_type,
nit: callbackType

http://codereview.chromium.org/1529005/diff/18001/15006#newcode237
third_party/npapi/bindings/npapi_extensions.h:237: void* callback_data);
nit: callbackData

[apatrick_chromium, 2010-04-06 22:18:56]

neb and/or brettw, can you take a look at this and see if you think it can be
adapted to audio? Ditto for 2D graphics. Thanks.

[brettw, 2010-04-07 00:13:26]

On Tue, Apr 6, 2010 at 3:18 PM,  <apatrick@chromium.org> wrote:
> neb and/or brettw, can you take a look at this and see if you think it can
> be
> adapted to audio? Ditto for 2D graphics. Thanks.

I think this approach should be fine for 2D, and I assume for audio as
well. It seems pretty straightforward.

Brett

[neb, 2010-04-13 17:52:20]

On 2010/04/06 22:18:56, apatrick_chromium wrote:
> neb and/or brettw, can you take a look at this and see if you think it can be
> adapted to audio? Ditto for 2D graphics. Thanks.

It would work very well for audio. Would definitely shorten the rtt for figuring
out which setup (mono/stereo/7.1 etc) to use.

[apatrick_chromium, 2010-04-15 00:05:26]

I have both the old device API and the new API working in parallel. 2D and audio
still need to be modified to use the new API.

Once these are working, the NaCl plumbing will need to be updated. At that point
we can get rid of the old API. It is my hope that this new design will actually
make the NaCl bindings simpler and make it easier to extend the API and add new
device types in the future.

PGL and the pepper test plugin now use the new API if enable_new_npdevice_api=1
is set in GYP_DEFINES.

I suggest I check this in so that we can start doing the above. I would like to
get rid of the ENABLE_NEW_NPDEVICE_API macro and the kludgy dual implementation
in webplugin_delegate_pepper.cc soon.

Can someone review? Thanks.

[brettw, 2010-04-16 18:17:45]

LGTM

[Nico, 2011-02-07 04:42:21]

http://codereview.chromium.org/1529005/diff/27001/chrome/renderer/webplugin_d...
File chrome/renderer/webplugin_delegate_pepper.h (right):

http://codereview.chromium.org/1529005/diff/27001/chrome/renderer/webplugin_d...
chrome/renderer/webplugin_delegate_pepper.h:158: void* callback_data);
Hi, this doesn't actually override the superclass method, because the parameter
type of the 2nd parameter is different. Will file a bug in a minute.