Fix DrawQuad copy

Fix DrawQuad copy

Creating a copy of the quads by allocating a bag of bytes is problematic because
the delete doesn't match the new (new [] vs scalar delete), which is wrong in
general but in particular causes asserts in tcmalloc and ASAN.
Instead, do the proper C++ version, relying on implicit copy constructors.

BUG=None


Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=165248

[jamesr, 2012-10-31 01:30:14]

surprisingly simple!

lgtm

[aelias_OOO_until_Jul13, 2012-10-31 02:50:02]

https://codereview.chromium.org/11344050/diff/1/cc/draw_quad.h
File cc/draw_quad.h (left):

https://codereview.chromium.org/11344050/diff/1/cc/draw_quad.h#oldcode60
cc/draw_quad.h:60: unsigned size() const;
I don't think we want to delete this.  The InputEvent-style ubercomp IPC pickler
will need to know the object size from a base class.  I added this method
originally for that reason.

[jamesr, 2012-10-31 02:55:25]

On 2012/10/31 02:50:02, aelias wrote:
> https://codereview.chromium.org/11344050/diff/1/cc/draw_quad.h
> File cc/draw_quad.h (left):
> 
> https://codereview.chromium.org/11344050/diff/1/cc/draw_quad.h#oldcode60
> cc/draw_quad.h:60: unsigned size() const;
> I don't think we want to delete this.  The InputEvent-style ubercomp IPC
pickler
> will need to know the object size from a base class.  I added this method
> originally for that reason.

Why not just write a field-by-field pickler (using the IPC macros) like most of
chromium does?

[aelias_OOO_until_Jul13, 2012-10-31 03:05:32]

The pickler is going to walk through a vector of pointers to the base class, so
it doesn't know which field-by-field pickler it should use. It turns out the
only class hierarchy currently being serialized across IPC is WebInputEvent, and
that uses this size trick.

[jamesr, 2012-10-31 03:09:18]

On 2012/10/31 03:05:32, aelias wrote:
> The pickler is going to walk through a vector of pointers to the base class,
so
> it doesn't know which field-by-field pickler it should use. It turns out the
> only class hierarchy currently being serialized across IPC is WebInputEvent,
and
> that uses this size trick.

Not true - look at the serialization of base::Value implemented in
ipc/ipc_message_utils.cc.  It's a switch statement on type and is pretty simple.

For WebInputEvent the type itself is defined in WebKit which can't directly use
the IPC macros, so the size trick makes sense.  I don't think it is applicable
here where the type is in the chromium repo.

[aelias_OOO_until_Jul13, 2012-10-31 03:10:57]

OK, sounds good.  Then we can also delete the #pragma pack()s in the draw quad
headers.

[aelias_OOO_until_Jul13, 2012-10-31 03:12:30]

lgtm

[piman, 2012-10-31 05:42:12]

Right, I have a pickler that walks the fields, it's not that much code.

[danakj, 2012-10-31 16:04:57]

wow, nice template work! lgtm2

[commit-bot: I haz the power, 2012-10-31 20:04:20]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/piman@chromium.org/11344050/10001

[commit-bot: I haz the power, 2012-10-31 22:48:15]

Change committed as 165248