Add gpu targets to untrusted NaCl build.

Add gpu targets to untrusted build, and add Graphics3D code to untrusted build.
BUG=116317
TEST=builds,manual

With this change, the IPC-based PPAPI proxy builds and runs the tumbler example.
Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=150343

[bbudge, 2012-07-19 16:34:23]

This compiles and almost links (missing ImageData shared and proxy code, which
David is working on.)

http://codereview.chromium.org/10796038/diff/5001/gpu/command_buffer/client/r...
File gpu/command_buffer/client/ref_counted.h (right):

http://codereview.chromium.org/10796038/diff/5001/gpu/command_buffer/client/r...
gpu/command_buffer/client/ref_counted.h:8: #if defined(__native_client__) &&
!defined(OS_NACL)
The NaCl SRPC proxy can't depend on base, so it pulls in a custom smart ptr. We
don't want it. Finesse around it by adding this clause, which is false for the
SRPC build which doesn't #include build_config.h.

[greggman, 2012-07-19 16:51:59]

LGTM for the GPU parts but please build the OpenGL ES 2.0 conformance tests and
make sure they run since changes like this often break them. (I'll add them to
the gpu bots)

You need src_internal. Then instructions are in src/gpu/gles2_conform/README

once they start and you see a window with flashing patterns you can Ctrl-C. Some
tests won't pass and a few take too long.

[piman, 2012-07-19 16:52:26]

At a high level, LGTM.
I don't understand the intricacies of the nacl untrusted build stuff, make sure
to have the gyp changes reviewed by someone who does.

http://codereview.chromium.org/10796038/diff/5001/gpu/command_buffer/client/r...
File gpu/command_buffer/client/ref_counted.h (right):

http://codereview.chromium.org/10796038/diff/5001/gpu/command_buffer/client/r...
gpu/command_buffer/client/ref_counted.h:8: #if defined(__native_client__) &&
!defined(OS_NACL)
On 2012/07/19 16:34:23, bbudge1 wrote:
> The NaCl SRPC proxy can't depend on base, so it pulls in a custom smart ptr.
We
> don't want it. Finesse around it by adding this clause, which is false for the
> SRPC build which doesn't #include build_config.h.

Worth a comment...
I assume at some point the SRPC proxy will go away, and this can be simplified,
correct?

[bbudge, 2012-07-19 18:38:10]

Refreshed patch and addressed a comment.

http://codereview.chromium.org/10796038/diff/5001/gpu/command_buffer/client/r...
File gpu/command_buffer/client/ref_counted.h (right):

http://codereview.chromium.org/10796038/diff/5001/gpu/command_buffer/client/r...
gpu/command_buffer/client/ref_counted.h:8: #if defined(__native_client__) &&
!defined(OS_NACL)
On 2012/07/19 16:52:26, piman wrote:
> On 2012/07/19 16:34:23, bbudge1 wrote:
> > The NaCl SRPC proxy can't depend on base, so it pulls in a custom smart ptr.
> We
> > don't want it. Finesse around it by adding this clause, which is false for
the
> > SRPC build which doesn't #include build_config.h.
> 
> Worth a comment...
> I assume at some point the SRPC proxy will go away, and this can be
simplified,
> correct?

Done.

[nfullagar, 2012-07-20 20:41:36]

Besides the conformance tests, can you please make sure existing ppapi 3d tests
& examples still build, link and run (both chrome & nacl ppapi tests)  IIR the
chrome tests can be built both trusted and untrusted - see
base/src/chrome/test/ui/ppapi_uitest.cc:

  TEST_PPAPI_IN_PROCESS(Graphics3D)
  TEST_PPAPI_OUT_OF_PROCESS(Graphics3D)
  TEST_PPAPI_NACL_VIA_HTTP(Graphics3D)

[dmichael (off chromium), 2012-07-20 22:09:47]

http://codereview.chromium.org/10796038/diff/5003/ppapi/ppapi_shared.gypi
File ppapi/ppapi_shared.gypi (right):

http://codereview.chromium.org/10796038/diff/5003/ppapi/ppapi_shared.gypi#new...
ppapi/ppapi_shared.gypi:248: 'shared_impl/ppb_image_data_shared.cc',
As discussed on IM, I think you shouldn't exclude this...  it's the source of
the link error. One possible fix for building that file is this patch:
http://codereview.chromium.org/10806052/diff/1/ppapi/shared_impl/ppb_image_da...

[brettw, 2012-07-23 19:42:50]

http://codereview.chromium.org/10796038/diff/15003/build/common.gypi
File build/common.gypi (left):

http://codereview.chromium.org/10796038/diff/15003/build/common.gypi#oldcode276
build/common.gypi:276: # Include the PPAPI IPC proxy for NaCl. This is a
work-in-progress; this
It's not clear to me why you deleted this.

[bbudge, 2012-07-23 19:53:39]

http://codereview.chromium.org/10796038/diff/5003/ppapi/ppapi_shared.gypi
File ppapi/ppapi_shared.gypi (right):

http://codereview.chromium.org/10796038/diff/5003/ppapi/ppapi_shared.gypi#new...
ppapi/ppapi_shared.gypi:248: 'shared_impl/ppb_image_data_shared.cc',
On 2012/07/20 22:09:47, dmichael wrote:
> As discussed on IM, I think you shouldn't exclude this...  it's the source of
> the link error. One possible fix for building that file is this patch:
>
http://codereview.chromium.org/10806052/diff/1/ppapi/shared_impl/ppb_image_da...

Done.

http://codereview.chromium.org/10796038/diff/15003/build/common.gypi
File build/common.gypi (left):

http://codereview.chromium.org/10796038/diff/15003/build/common.gypi#oldcode276
build/common.gypi:276: # Include the PPAPI IPC proxy for NaCl. This is a
work-in-progress; this
On 2012/07/23 19:42:51, brettw wrote:
> It's not clear to me why you deleted this.

To get gpu code to build, we need a compile time flag 'NACL_IPC_PPAPI_PROXY', to
avoid pulling in NaCl's custom ref_counted (used to avoid dependency on base in
the SRPC proxy). I couldn't successfully define such a flag using this variable
as defined here so I moved it to untrusted.gypi in the NaCl repo. This should be
OK since we shouldn't need to alter the trusted build. In fact, we may want to
remove this entirely and build both proxies side by side, using the Chrome flag
alone to determine which proxy starts up.

[brettw, 2012-07-23 19:56:08]

Okay, LGTM

[bbudge, 2012-07-25 01:16:42]

On 2012/07/23 19:56:08, brettw wrote:
> Okay, LGTM

Waiting for NaCl revision to update on trybots.

[dmichael (off chromium), 2012-07-25 01:45:11]

If the deps roll is checked in, you can try with -r HEAD
(Or whatever revision has the DEPS roll)
On Jul 24, 2012 7:16 PM, <bbudge@chromium.org> wrote:

> On 2012/07/23 19:56:08, brettw wrote:
>
>> Okay, LGTM
>>
>
> Waiting for NaCl revision to update on trybots.
>
>
http://codereview.chromium.**org/10796038/<http://codereview.chromium.org/107...
>

[bbudge, 2012-07-30 23:40:39]

On 2012/07/25 20:48:34, bbudge1 wrote:

Latest upload is FYI. Will remove 'printfs' when we get it all working.

[bbudge, 2012-08-02 22:11:00]

Add gpu targets to untrusted build, and add Graphics3D code to untrusted build.
BUG=116317
TEST=builds,manual

With this change, the IPC-based PPAPI proxy builds and runs the tumbler example.

[bbudge, 2012-08-02 22:13:21]

This patch incorporates some of dmichael's change to NaClIPCAdapter, to pass
handles. I will rebaseline. Waiting for bradnelson's changes to the untrusted
build to roll into Chrome, so the Windows build will be fixed.

[dmichael (off chromium), 2012-08-03 03:31:41]

http://codereview.chromium.org/10796038/diff/23014/ppapi/proxy/ppapi_messages.h
File ppapi/proxy/ppapi_messages.h (right):

http://codereview.chromium.org/10796038/diff/23014/ppapi/proxy/ppapi_messages...
ppapi/proxy/ppapi_messages.h:372: ppapi::proxy::SerializedHandle /* handle */)
I think you don't want this part of my patch here...

If I land first, you can just resolve the conflict. But I'm still working on
Windows, and then need review. I do think I'll be able to get it building on
Windows tomorrow. I think I know an easy way to do it, and my Windows local
build *finally* seems to be working right.

Anyway, point is, you're likely to land first, and if so you'll need to leave
SerializedHandle changes out.