[Mac] Create transparent cursor when given empty custom data.

[Mac] Create transparent cursor when given empty custom data.

Based on webcursor_win.cc's implementation of GetCursor().

BUG=73356
TEST=Watch crash server.

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=91452

[Scott Hess - ex-Googler, 2011-06-29 00:06:54]

Tested by replacing the TypeCustom stanza with:

      return CreateCustomCursor(std::vector<char>(), custom_size_, hotspot_);

Unfortunately, I've not been able to figure out a way to make custom_data_
empty.

[Scott Hess - ex-Googler, 2011-06-29 00:08:35]

In addition to the Windows impl of GetCursor() simply skipping SetDIBits() in
this case, things like http://crbug.com/35188 seem to imply that having an
invisible cursor is reasonable.

[Avi (use Gerrit), 2011-06-29 00:22:47]

okey dokey! LGTM

http://codereview.chromium.org/7273056/diff/1/webkit/glue/webcursor_mac.mm
File webkit/glue/webcursor_mac.mm (right):

http://codereview.chromium.org/7273056/diff/1/webkit/glue/webcursor_mac.mm#ne...
webkit/glue/webcursor_mac.mm:48: CGBitmapContextCreate(data,
You can do that, pass in a null data for a 0x0 image?

[Scott Hess - ex-Googler, 2011-06-29 01:06:13]

http://codereview.chromium.org/7273056/diff/1/webkit/glue/webcursor_mac.mm
File webkit/glue/webcursor_mac.mm (right):

http://codereview.chromium.org/7273056/diff/1/webkit/glue/webcursor_mac.mm#ne...
webkit/glue/webcursor_mac.mm:48: CGBitmapContextCreate(data,
On 2011/06/29 00:22:47, Avi wrote:
> You can do that, pass in a null data for a 0x0 image?

Docs say that passing NULL means that Quartz allocates the memory.

I tested with NULL - but the size was non-zero.  In that case it worked as
expected.  But I just went back and instrumented the code to pass a size of
{0,0}, and in that case CGBitmapContextCreate() returns NULL, and the exception
stuff from the old comment happens.

I tried revising so that |cursor_image| gets no representation in case of NULL
or 0x0, but no go.  Nothing bad happens, but the cursor doesn't change at all. 
The docs seem to imply that a zero-size image is not going to generally work
well, in any case.

I do not have enough insight into the overall system to be able to offhand say
if we can get an actual {0,0} size, or if the empty image data with a non-zero
size is expected.  The transparent cursor seems right, it would be easy enough
to pass NULL for data (system allocated transparent backing) and {1,1} for size
in that case.  Or I could back off and instrument RWHVM further to dump the size
into breakpad along with the type.

[I can't for the life of me figure out how to snag the URL for this crash, that
would rock.]

[Scott Hess - ex-Googler, 2011-06-29 01:30:50]

[Sigh.  Juggling and not thinking, I just have just put up a TBR review to
include the size info and left for the day.  Now it's too late.]

[Scott Hess - ex-Googler, 2011-07-01 21:03:35]

On 2011/06/29 01:30:50, shess wrote:
> [Sigh.  Juggling and not thinking, I just have just put up a TBR review to
> include the size info and left for the day.  Now it's too late.]

Got a report from the other CL, custom-cursor-size = {0,0}.  Odd.  I wonder what
it all means?

Anyhow, will modify to handle either empty image data or empty cursor size by
creating a single-pixel transparent cursor.

[Scott Hess - ex-Googler, 2011-07-01 22:20:38]

On 2011/07/01 21:03:35, shess wrote:
> On 2011/06/29 01:30:50, shess wrote:
> > [Sigh.  Juggling and not thinking, I just have just put up a TBR review to
> > include the size info and left for the day.  Now it's too late.]
> 
> Got a report from the other CL, custom-cursor-size = {0,0}.  Odd.  I wonder
what
> it all means?
> 
> Anyhow, will modify to handle either empty image data or empty cursor size by
> creating a single-pixel transparent cursor.

OK, done that.  Also backed out the earlier logging code.  I don't see where to
revert the LGTM :-).  Anyhow, look again?  No rush.

Have a good weekend!

[Avi (use Gerrit), 2011-07-01 23:15:59]

LGTM