aura-x11: Add custom web cursor support.

aura-x11: Add custom web cursor support.

BUG=102562
TEST=manually (using sites that use custom cursor, e.g.: http://beradrian.users.sourceforge.net/articles/cursor.html)

Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=130456

[sadrul, 2012-02-24 04:14:43]

ben@, derat@: Please review.

darin@: OWNERS for webkit/

[darin (slow to review), 2012-02-24 06:32:35]

http://codereview.chromium.org/9463003/diff/5015/webkit/glue/webcursor.h
File webkit/glue/webcursor.h (right):

http://codereview.chromium.org/9463003/diff/5015/webkit/glue/webcursor.h#newc...
webkit/glue/webcursor.h:148: typedef struct _XcursorImage XcursorImage;
nit: indentation

nit: usually forward declarations go at the top of the header.  Or, did you mean
to define a class-scoped typedef?

http://codereview.chromium.org/9463003/diff/5015/webkit/glue/webcursor_aurax1...
File webkit/glue/webcursor_aurax11.cc (right):

http://codereview.chromium.org/9463003/diff/5015/webkit/glue/webcursor_aurax1...
webkit/glue/webcursor_aurax11.cc:31: uint8* pixels = (uint8*)image_->pixels;
nit: use C++ style cast

http://codereview.chromium.org/9463003/diff/5015/webkit/glue/webcursor_aurax1...
webkit/glue/webcursor_aurax11.cc:32: // SkBitmaps are premultiplied, we need to
unpremultiply them.
is there really no common routine somewhere to unpremultiply a SkBitmap? 
really?  not in skia somewhere or elsewhere in chrome?  is there really no other
place in the code where we do this? :-)

http://codereview.chromium.org/9463003/diff/5015/webkit/glue/webcursor_aurax1...
webkit/glue/webcursor_aurax11.cc:69: return true;
nit: should these assert not reached?

[Daniel Erat, 2012-02-24 17:17:12]

http://codereview.chromium.org/9463003/diff/5015/ui/aura/root_window.h
File ui/aura/root_window.h (right):

http://codereview.chromium.org/9463003/diff/5015/ui/aura/root_window.h#newcode93
ui/aura/root_window.h:93: // explicit call to SetCustomCursor is made.
these semantics seem odd to me.  what about making
RootWindowHostLinux::SetCustomCursor() instead save the custom cursor to a
member variable and set |current_cursor_| to kCursorCustom, and making
RootWindowHostLinux::SetCursorInternal() maybe additionally take a
gfx::PlatformCursor for the kCursorCustom case?  that way, ShowCursor() and
HideCursor() will work similarly for both native and custom cursors.  is the
concern that the X cursor will be destroyed if we don't immediately assign it to
a window?  (i think the server does reference counting for cursors.)

[sadrul, 2012-02-24 17:38:19]

http://codereview.chromium.org/9463003/diff/5015/ui/aura/root_window.h
File ui/aura/root_window.h (right):

http://codereview.chromium.org/9463003/diff/5015/ui/aura/root_window.h#newcode93
ui/aura/root_window.h:93: // explicit call to SetCustomCursor is made.
On 2012/02/24 17:17:12, Daniel Erat wrote:
> these semantics seem odd to me.  what about making
> RootWindowHostLinux::SetCustomCursor() instead save the custom cursor to a
> member variable and set |current_cursor_| to kCursorCustom, and making
> RootWindowHostLinux::SetCursorInternal() maybe additionally take a
> gfx::PlatformCursor for the kCursorCustom case?  that way, ShowCursor() and
> HideCursor() will work similarly for both native and custom cursors.  is the
> concern that the X cursor will be destroyed if we don't immediately assign it
to
> a window?  (i think the server does reference counting for cursors.)

My concern is that by the time ShowCursor gets called, the PlatformCursor may
have been destroyed (e.g. the page navigated or was closed etc.), and is not
valid anymore. This can be fixed if we reach into RootWindow from
WebCursor::CleanupPlatformData and update the member variable for the
platform-cursor appropriately. Does that sounds better than this?

http://codereview.chromium.org/9463003/diff/5015/webkit/glue/webcursor.h
File webkit/glue/webcursor.h (right):

http://codereview.chromium.org/9463003/diff/5015/webkit/glue/webcursor.h#newc...
webkit/glue/webcursor.h:148: typedef struct _XcursorImage XcursorImage;
On 2012/02/24 06:32:35, darin wrote:
> nit: indentation
> 
> nit: usually forward declarations go at the top of the header.  Or, did you
mean
> to define a class-scoped typedef?

Moved the typedef at the top.

http://codereview.chromium.org/9463003/diff/5015/webkit/glue/webcursor_aurax1...
File webkit/glue/webcursor_aurax11.cc (right):

http://codereview.chromium.org/9463003/diff/5015/webkit/glue/webcursor_aurax1...
webkit/glue/webcursor_aurax11.cc:32: // SkBitmaps are premultiplied, we need to
unpremultiply them.
On 2012/02/24 06:32:35, darin wrote:
> is there really no common routine somewhere to unpremultiply a SkBitmap? 
> really?  not in skia somewhere or elsewhere in chrome?  is there really no
other
> place in the code where we do this? :-)

Indeed there is, for GTK+. In my effort to refactor this code now, I realized
that this code works well for GTK+, however it doesn't work well for these X
cursors (the color is not always right). I have changed this code to work
correctly and tested this more (e.g.
http://www.w3schools.com/cssref/playit.asp?filename=playcss_cursor).

http://codereview.chromium.org/9463003/diff/5015/webkit/glue/webcursor_aurax1...
webkit/glue/webcursor_aurax11.cc:69: return true;
On 2012/02/24 06:32:35, darin wrote:
> nit: should these assert not reached?

I am unsure if we want to NOTREACHED here. So I kept it as it was before this
change.

http://codereview.chromium.org/9463003/diff/8002/ui/aura/root_window_host_lin...
File ui/aura/root_window_host_linux.cc (right):

http://codereview.chromium.org/9463003/diff/8002/ui/aura/root_window_host_lin...
ui/aura/root_window_host_linux.cc:158: NOTREACHED();
Too heavy-handed?

[Daniel Erat, 2012-02-24 18:05:42]

http://codereview.chromium.org/9463003/diff/5015/ui/aura/root_window.h
File ui/aura/root_window.h (right):

http://codereview.chromium.org/9463003/diff/5015/ui/aura/root_window.h#newcode93
ui/aura/root_window.h:93: // explicit call to SetCustomCursor is made.
On 2012/02/24 17:38:19, sadrul wrote:
> On 2012/02/24 17:17:12, Daniel Erat wrote:
> > these semantics seem odd to me.  what about making
> > RootWindowHostLinux::SetCustomCursor() instead save the custom cursor to a
> > member variable and set |current_cursor_| to kCursorCustom, and making
> > RootWindowHostLinux::SetCursorInternal() maybe additionally take a
> > gfx::PlatformCursor for the kCursorCustom case?  that way, ShowCursor() and
> > HideCursor() will work similarly for both native and custom cursors.  is the
> > concern that the X cursor will be destroyed if we don't immediately assign
it
> to
> > a window?  (i think the server does reference counting for cursors.)
> 
> My concern is that by the time ShowCursor gets called, the PlatformCursor may
> have been destroyed (e.g. the page navigated or was closed etc.), and is not
> valid anymore. This can be fixed if we reach into RootWindow from
> WebCursor::CleanupPlatformData and update the member variable for the
> platform-cursor appropriately. Does that sounds better than this?

At least the X case, I think that the underlying cursor may remain usable as
long as it's assigned to a window, even after XFreeCursor() is called (at least,
I think that cursor IDs returned by XCreateFontCursor() work this way).  There's
still the downside that we'll lose the reference if the PlatformCursor is
cleaned up after a call to HideCursor(), I guess.

Would adding our own wrapper around the underlying X cursor and making it be
reference-counted be too gross?  (Possibly.)

http://codereview.chromium.org/9463003/diff/8002/ui/aura/root_window_host_lin...
File ui/aura/root_window_host_linux.cc (right):

http://codereview.chromium.org/9463003/diff/8002/ui/aura/root_window_host_lin...
ui/aura/root_window_host_linux.cc:158: NOTREACHED();
On 2012/02/24 17:38:19, sadrul wrote:
> Too heavy-handed?

No, this seems reasonable to me.

[sadrul, 2012-02-24 18:30:53]

http://codereview.chromium.org/9463003/diff/5015/ui/aura/root_window.h
File ui/aura/root_window.h (right):

http://codereview.chromium.org/9463003/diff/5015/ui/aura/root_window.h#newcode93
ui/aura/root_window.h:93: // explicit call to SetCustomCursor is made.
On 2012/02/24 18:05:47, Daniel Erat wrote:
> On 2012/02/24 17:38:19, sadrul wrote:
> > On 2012/02/24 17:17:12, Daniel Erat wrote:
> > > these semantics seem odd to me.  what about making
> > > RootWindowHostLinux::SetCustomCursor() instead save the custom cursor to a
> > > member variable and set |current_cursor_| to kCursorCustom, and making
> > > RootWindowHostLinux::SetCursorInternal() maybe additionally take a
> > > gfx::PlatformCursor for the kCursorCustom case?  that way, ShowCursor()
and
> > > HideCursor() will work similarly for both native and custom cursors.  is
the
> > > concern that the X cursor will be destroyed if we don't immediately assign
> it
> > to
> > > a window?  (i think the server does reference counting for cursors.)
> > 
> > My concern is that by the time ShowCursor gets called, the PlatformCursor
may
> > have been destroyed (e.g. the page navigated or was closed etc.), and is not
> > valid anymore. This can be fixed if we reach into RootWindow from
> > WebCursor::CleanupPlatformData and update the member variable for the
> > platform-cursor appropriately. Does that sounds better than this?
> 
> At least the X case, I think that the underlying cursor may remain usable as
> long as it's assigned to a window, even after XFreeCursor() is called (at
least,
> I think that cursor IDs returned by XCreateFontCursor() work this way). 
There's
> still the downside that we'll lose the reference if the PlatformCursor is
> cleaned up after a call to HideCursor(), I guess.
> 
> Would adding our own wrapper around the underlying X cursor and making it be
> reference-counted be too gross?  (Possibly.)

I considered doing that at one point, but that seemed like too much work for not
much gain :) Having said that, having a cache for custom X cursors probably
won't be all that much work (there is a cache for standard cursors in x11_util,
we can probably reuse that to some extent).

[sadrul, 2012-02-24 19:46:29]

PTAL

http://codereview.chromium.org/9463003/diff/5015/ui/aura/root_window.h
File ui/aura/root_window.h (right):

http://codereview.chromium.org/9463003/diff/5015/ui/aura/root_window.h#newcode93
ui/aura/root_window.h:93: // explicit call to SetCustomCursor is made.
On 2012/02/24 18:30:53, sadrul wrote:
> On 2012/02/24 18:05:47, Daniel Erat wrote:
> > On 2012/02/24 17:38:19, sadrul wrote:
> > > On 2012/02/24 17:17:12, Daniel Erat wrote:
> > > > these semantics seem odd to me.  what about making
> > > > RootWindowHostLinux::SetCustomCursor() instead save the custom cursor to
a
> > > > member variable and set |current_cursor_| to kCursorCustom, and making
> > > > RootWindowHostLinux::SetCursorInternal() maybe additionally take a
> > > > gfx::PlatformCursor for the kCursorCustom case?  that way, ShowCursor()
> and
> > > > HideCursor() will work similarly for both native and custom cursors.  is
> the
> > > > concern that the X cursor will be destroyed if we don't immediately
assign
> > it
> > > to
> > > > a window?  (i think the server does reference counting for cursors.)
> > > 
> > > My concern is that by the time ShowCursor gets called, the PlatformCursor
> may
> > > have been destroyed (e.g. the page navigated or was closed etc.), and is
not
> > > valid anymore. This can be fixed if we reach into RootWindow from
> > > WebCursor::CleanupPlatformData and update the member variable for the
> > > platform-cursor appropriately. Does that sounds better than this?
> > 
> > At least the X case, I think that the underlying cursor may remain usable as
> > long as it's assigned to a window, even after XFreeCursor() is called (at
> least,
> > I think that cursor IDs returned by XCreateFontCursor() work this way). 
> There's
> > still the downside that we'll lose the reference if the PlatformCursor is
> > cleaned up after a call to HideCursor(), I guess.
> > 
> > Would adding our own wrapper around the underlying X cursor and making it be
> > reference-counted be too gross?  (Possibly.)
> 
> I considered doing that at one point, but that seemed like too much work for
not
> much gain :) Having said that, having a cache for custom X cursors probably
> won't be all that much work (there is a cache for standard cursors in
x11_util,
> we can probably reuse that to some extent).

Went ahead and did the caching (not too much work!). :-)

[Daniel Erat, 2012-02-24 21:04:36]

Thanks, I like this approach.

http://codereview.chromium.org/9463003/diff/5021/ui/aura/root_window.h
File ui/aura/root_window.h (right):

http://codereview.chromium.org/9463003/diff/5021/ui/aura/root_window.h#newcode86
ui/aura/root_window.h:86: // used. If the cursor type is kCursorCustom, then the
caller has to call
how about:

// To set a custom cursor, use SetCustomCursor() instead.

, and then adding a CHECK(cursor != kCursorCustom()) in SetCursor()?

http://codereview.chromium.org/9463003/diff/5021/ui/aura/root_window_host_lin...
File ui/aura/root_window_host_linux.cc (right):

http://codereview.chromium.org/9463003/diff/5021/ui/aura/root_window_host_lin...
ui/aura/root_window_host_linux.cc:558: ui::UnrefCustomXCursor(custom_cursor_);
put this in the d'tor too

http://codereview.chromium.org/9463003/diff/5021/ui/aura/root_window_host_lin...
ui/aura/root_window_host_linux.cc:563: XDefineCursor(xdisplay_, xwindow_,
cursor);
i think that you need to make ShowCursor() handle the custom-cursor case too. 
what about something like the following?

- make SetCustomCursor() set custom_cursor_ to the passed-in cursor, set
current_cursor_ to kCursorCustom, and (if cursor_shown_ is true) call
SetCursorInternal(kCursorCustom)
- make SetCursorInternal() use custom_cursor_ (after CHECK()-ing that it's not
0) for the kCursorCustom case
- get rid of the cursor-unref-ing in SetCursorInternal(), and only unref in the
d'tor, SetCustomCursor(), and SetCursor()

http://codereview.chromium.org/9463003/diff/5021/ui/base/x/x11_util.cc
File ui/base/x/x11_util.cc (right):

http://codereview.chromium.org/9463003/diff/5021/ui/base/x/x11_util.cc#newcod...
ui/base/x/x11_util.cc:222: cache_.clear();
do you need to call XFreeCursor() on all of the keys here?

http://codereview.chromium.org/9463003/diff/5021/ui/base/x/x11_util.cc#newcod...
ui/base/x/x11_util.cc:230: XCustomCursor(XcursorImage* image)
nit: document that it takes ownership of |image|

http://codereview.chromium.org/9463003/diff/5021/ui/base/x/x11_util.h
File ui/base/x/x11_util.h (right):

http://codereview.chromium.org/9463003/diff/5021/ui/base/x/x11_util.h#newcode85
ui/base/x/x11_util.h:85: UI_EXPORT Cursor
CreateReffedCustomXCursor(XcursorImage* image);
will defer to others on "Reffed" vs. "Refed" vs. "Referenced"

[sadrul, 2012-02-24 22:48:49]

http://codereview.chromium.org/9463003/diff/5021/ui/aura/root_window.h
File ui/aura/root_window.h (right):

http://codereview.chromium.org/9463003/diff/5021/ui/aura/root_window.h#newcode86
ui/aura/root_window.h:86: // used. If the cursor type is kCursorCustom, then the
caller has to call
On 2012/02/24 21:04:36, Daniel Erat wrote:
> how about:
> 
> // To set a custom cursor, use SetCustomCursor() instead.
> 
> , and then adding a CHECK(cursor != kCursorCustom()) in SetCursor()?

Done.

http://codereview.chromium.org/9463003/diff/5021/ui/aura/root_window_host_lin...
File ui/aura/root_window_host_linux.cc (right):

http://codereview.chromium.org/9463003/diff/5021/ui/aura/root_window_host_lin...
ui/aura/root_window_host_linux.cc:558: ui::UnrefCustomXCursor(custom_cursor_);
On 2012/02/24 21:04:36, Daniel Erat wrote:
> put this in the d'tor too

Ah, good catch. Done.

http://codereview.chromium.org/9463003/diff/5021/ui/aura/root_window_host_lin...
ui/aura/root_window_host_linux.cc:563: XDefineCursor(xdisplay_, xwindow_,
cursor);
On 2012/02/24 21:04:36, Daniel Erat wrote:
> i think that you need to make ShowCursor() handle the custom-cursor case too. 
> what about something like the following?
> 
> - make SetCustomCursor() set custom_cursor_ to the passed-in cursor, set
> current_cursor_ to kCursorCustom, and (if cursor_shown_ is true) call
> SetCursorInternal(kCursorCustom)
> - make SetCursorInternal() use custom_cursor_ (after CHECK()-ing that it's not
> 0) for the kCursorCustom case
> - get rid of the cursor-unref-ing in SetCursorInternal(), and only unref in
the
> d'tor, SetCustomCursor(), and SetCursor()

Very nice. Done!

One exception: I am not CHECK()ing in SetCursor for cursor != kCursorCustom.
This is because RootWindowEventFilter essentially does
root->SetCursor(target->GetCursor()). Doing this correctly in RWEF will involve
adding a corresponding GetCustomCursor on aura::Window, WindowDelegate, and
ref/unref-ing the custom cursor in RWEF (for a
|custom_cursor_to_set_on_unlock_|). I feel like that's bordering on too much
complication, and so decided against it.

http://codereview.chromium.org/9463003/diff/5021/ui/base/x/x11_util.cc
File ui/base/x/x11_util.cc (right):

http://codereview.chromium.org/9463003/diff/5021/ui/base/x/x11_util.cc#newcod...
ui/base/x/x11_util.cc:222: cache_.clear();
On 2012/02/24 21:04:36, Daniel Erat wrote:
> do you need to call XFreeCursor() on all of the keys here?

Good catch. This needs to happen. I have moved the Cursor ownership from
XCustomCursorCache to XCustomCursor, and the cursor is free'd from its
destructor.

http://codereview.chromium.org/9463003/diff/5021/ui/base/x/x11_util.cc#newcod...
ui/base/x/x11_util.cc:230: XCustomCursor(XcursorImage* image)
On 2012/02/24 21:04:36, Daniel Erat wrote:
> nit: document that it takes ownership of |image|

Done.

[Daniel Erat, 2012-02-24 23:19:30]

http://codereview.chromium.org/9463003/diff/5021/ui/aura/root_window_host_lin...
File ui/aura/root_window_host_linux.cc (right):

http://codereview.chromium.org/9463003/diff/5021/ui/aura/root_window_host_lin...
ui/aura/root_window_host_linux.cc:563: XDefineCursor(xdisplay_, xwindow_,
cursor);
On 2012/02/24 22:48:49, sadrul wrote:
> On 2012/02/24 21:04:36, Daniel Erat wrote:
> > i think that you need to make ShowCursor() handle the custom-cursor case
too. 
> > what about something like the following?
> > 
> > - make SetCustomCursor() set custom_cursor_ to the passed-in cursor, set
> > current_cursor_ to kCursorCustom, and (if cursor_shown_ is true) call
> > SetCursorInternal(kCursorCustom)
> > - make SetCursorInternal() use custom_cursor_ (after CHECK()-ing that it's
not
> > 0) for the kCursorCustom case
> > - get rid of the cursor-unref-ing in SetCursorInternal(), and only unref in
> the
> > d'tor, SetCustomCursor(), and SetCursor()
> 
> Very nice. Done!
> 
> One exception: I am not CHECK()ing in SetCursor for cursor != kCursorCustom.
> This is because RootWindowEventFilter essentially does
> root->SetCursor(target->GetCursor()). Doing this correctly in RWEF will
involve
> adding a corresponding GetCustomCursor on aura::Window, WindowDelegate, and
> ref/unref-ing the custom cursor in RWEF (for a
> |custom_cursor_to_set_on_unlock_|). I feel like that's bordering on too much
> complication, and so decided against it.

Hmm, I see.  Will the current unsetting/unref-ing of custom_cursor_ in
SetCursor() cause problems here, since the custom cursor will be gone when we
try to restore it later?  Perhaps the old custom cursor should only be thrown
away in SetCustomCursor().

http://codereview.chromium.org/9463003/diff/3037/ui/aura/root_window_host_lin...
File ui/aura/root_window_host_linux.cc (right):

http://codereview.chromium.org/9463003/diff/3037/ui/aura/root_window_host_lin...
ui/aura/root_window_host_linux.cc:566: if (custom_cursor_)
should this method CHECK(cursor)?  i'm pretty sure that passing 0 will generate
a BadCursor error

[sadrul, 2012-02-25 00:02:29]

http://codereview.chromium.org/9463003/diff/5021/ui/aura/root_window_host_lin...
File ui/aura/root_window_host_linux.cc (right):

http://codereview.chromium.org/9463003/diff/5021/ui/aura/root_window_host_lin...
ui/aura/root_window_host_linux.cc:563: XDefineCursor(xdisplay_, xwindow_,
cursor);
On 2012/02/24 23:19:31, Daniel Erat wrote:
> On 2012/02/24 22:48:49, sadrul wrote:
> > On 2012/02/24 21:04:36, Daniel Erat wrote:
> > > i think that you need to make ShowCursor() handle the custom-cursor case
> too. 
> > > what about something like the following?
> > > 
> > > - make SetCustomCursor() set custom_cursor_ to the passed-in cursor, set
> > > current_cursor_ to kCursorCustom, and (if cursor_shown_ is true) call
> > > SetCursorInternal(kCursorCustom)
> > > - make SetCursorInternal() use custom_cursor_ (after CHECK()-ing that it's
> not
> > > 0) for the kCursorCustom case
> > > - get rid of the cursor-unref-ing in SetCursorInternal(), and only unref
in
> > the
> > > d'tor, SetCustomCursor(), and SetCursor()
> > 
> > Very nice. Done!
> > 
> > One exception: I am not CHECK()ing in SetCursor for cursor != kCursorCustom.
> > This is because RootWindowEventFilter essentially does
> > root->SetCursor(target->GetCursor()). Doing this correctly in RWEF will
> involve
> > adding a corresponding GetCustomCursor on aura::Window, WindowDelegate, and
> > ref/unref-ing the custom cursor in RWEF (for a
> > |custom_cursor_to_set_on_unlock_|). I feel like that's bordering on too much
> > complication, and so decided against it.
> 
> Hmm, I see.  Will the current unsetting/unref-ing of custom_cursor_ in
> SetCursor() cause problems here, since the custom cursor will be gone when we
> try to restore it later?  Perhaps the old custom cursor should only be thrown
> away in SetCustomCursor().

I have changed the code in SetCursor to unset custom_cursor_ only if it wasn't
called with kCursorCustom.

http://codereview.chromium.org/9463003/diff/3037/ui/aura/root_window_host_lin...
File ui/aura/root_window_host_linux.cc (right):

http://codereview.chromium.org/9463003/diff/3037/ui/aura/root_window_host_lin...
ui/aura/root_window_host_linux.cc:566: if (custom_cursor_)
On 2012/02/24 23:19:31, Daniel Erat wrote:
> should this method CHECK(cursor)?  i'm pretty sure that passing 0 will
generate
> a BadCursor error

I just removed the if check here. The XCustomCursorCache code will crash for a 0
(or any unknown cursor).

[Daniel Erat, 2012-02-25 00:38:56]

LGTM if this isn't an issue.

http://codereview.chromium.org/9463003/diff/5021/ui/aura/root_window_host_lin...
File ui/aura/root_window_host_linux.cc (right):

http://codereview.chromium.org/9463003/diff/5021/ui/aura/root_window_host_lin...
ui/aura/root_window_host_linux.cc:563: XDefineCursor(xdisplay_, xwindow_,
cursor);
On 2012/02/25 00:02:30, sadrul wrote:
> On 2012/02/24 23:19:31, Daniel Erat wrote:
> > On 2012/02/24 22:48:49, sadrul wrote:
> > > On 2012/02/24 21:04:36, Daniel Erat wrote:
> > > > i think that you need to make ShowCursor() handle the custom-cursor case
> > too. 
> > > > what about something like the following?
> > > > 
> > > > - make SetCustomCursor() set custom_cursor_ to the passed-in cursor, set
> > > > current_cursor_ to kCursorCustom, and (if cursor_shown_ is true) call
> > > > SetCursorInternal(kCursorCustom)
> > > > - make SetCursorInternal() use custom_cursor_ (after CHECK()-ing that
it's
> > not
> > > > 0) for the kCursorCustom case
> > > > - get rid of the cursor-unref-ing in SetCursorInternal(), and only unref
> in
> > > the
> > > > d'tor, SetCustomCursor(), and SetCursor()
> > > 
> > > Very nice. Done!
> > > 
> > > One exception: I am not CHECK()ing in SetCursor for cursor !=
kCursorCustom.
> > > This is because RootWindowEventFilter essentially does
> > > root->SetCursor(target->GetCursor()). Doing this correctly in RWEF will
> > involve
> > > adding a corresponding GetCustomCursor on aura::Window, WindowDelegate,
and
> > > ref/unref-ing the custom cursor in RWEF (for a
> > > |custom_cursor_to_set_on_unlock_|). I feel like that's bordering on too
much
> > > complication, and so decided against it.
> > 
> > Hmm, I see.  Will the current unsetting/unref-ing of custom_cursor_ in
> > SetCursor() cause problems here, since the custom cursor will be gone when
we
> > try to restore it later?  Perhaps the old custom cursor should only be
thrown
> > away in SetCustomCursor().
> 
> I have changed the code in SetCursor to unset custom_cursor_ only if it wasn't
> called with kCursorCustom.

So just to make sure that I understand correctly, is it now safe for
SetCursor(kCursorCustom) to be called even if custom_cursor_ is 0?  Or can this
never happen?  It seems like it can, if RootWindowEventFilter::UpdateCursor()
tries to set a custom cursor while the cursor is locked, but maybe this is
avoided for some other reason (I'm not familiar with RWEF).

[Ben Goodger (Google), 2012-02-25 00:44:40]

Owners LGTM for ui/aura, and ui/x since there's no OWNERS file there... note
however that there should be. I'll leave it to you X-heads to decide who should
be in it.

[sadrul, 2012-02-25 05:05:30]

On 2012/02/25 00:38:56, Daniel Erat wrote:
> LGTM if this isn't an issue.
> 
>
http://codereview.chromium.org/9463003/diff/5021/ui/aura/root_window_host_lin...
> File ui/aura/root_window_host_linux.cc (right):
> 
>
http://codereview.chromium.org/9463003/diff/5021/ui/aura/root_window_host_lin...
> ui/aura/root_window_host_linux.cc:563: XDefineCursor(xdisplay_, xwindow_,
> cursor);
> On 2012/02/25 00:02:30, sadrul wrote:
> > On 2012/02/24 23:19:31, Daniel Erat wrote:
> > > On 2012/02/24 22:48:49, sadrul wrote:
> > > > On 2012/02/24 21:04:36, Daniel Erat wrote:
> > > > > i think that you need to make ShowCursor() handle the custom-cursor
case
> > > too. 
> > > > > what about something like the following?
> > > > > 
> > > > > - make SetCustomCursor() set custom_cursor_ to the passed-in cursor,
set
> > > > > current_cursor_ to kCursorCustom, and (if cursor_shown_ is true) call
> > > > > SetCursorInternal(kCursorCustom)
> > > > > - make SetCursorInternal() use custom_cursor_ (after CHECK()-ing that
> it's
> > > not
> > > > > 0) for the kCursorCustom case
> > > > > - get rid of the cursor-unref-ing in SetCursorInternal(), and only
unref
> > in
> > > > the
> > > > > d'tor, SetCustomCursor(), and SetCursor()
> > > > 
> > > > Very nice. Done!
> > > > 
> > > > One exception: I am not CHECK()ing in SetCursor for cursor !=
> kCursorCustom.
> > > > This is because RootWindowEventFilter essentially does
> > > > root->SetCursor(target->GetCursor()). Doing this correctly in RWEF will
> > > involve
> > > > adding a corresponding GetCustomCursor on aura::Window, WindowDelegate,
> and
> > > > ref/unref-ing the custom cursor in RWEF (for a
> > > > |custom_cursor_to_set_on_unlock_|). I feel like that's bordering on too
> much
> > > > complication, and so decided against it.
> > > 
> > > Hmm, I see.  Will the current unsetting/unref-ing of custom_cursor_ in
> > > SetCursor() cause problems here, since the custom cursor will be gone when
> we
> > > try to restore it later?  Perhaps the old custom cursor should only be
> thrown
> > > away in SetCustomCursor().
> > 
> > I have changed the code in SetCursor to unset custom_cursor_ only if it
wasn't
> > called with kCursorCustom.
> 
> So just to make sure that I understand correctly, is it now safe for
> SetCursor(kCursorCustom) to be called even if custom_cursor_ is 0?  Or can
this
> never happen?  It seems like it can, if RootWindowEventFilter::UpdateCursor()
> tries to set a custom cursor while the cursor is locked, but maybe this is
> avoided for some other reason (I'm not familiar with RWEF).

You are right. This can be an issue.

I also notice that even when this isn't happening, custom cursors do not always
updated correctly. So I have went ahead with a more intrusive surgery: define a
ref-counted aura::Cursor as the NativeCursor, which has an int (one of the
various kCursor___ values), and a PlatformCursor. This way, I believe all of the
issues are resolved, and I kind of think this is also cleaner. Sorry for the
large-ish change after LGTMs, but please take another look!

Note: I added a few operator overloads to keep the changelist small (and I
personally think the overloads in this case makes sense).

[Daniel Erat, 2012-02-27 17:53:55]

http://codereview.chromium.org/9463003/diff/6042/ui/aura/cursor.h
File ui/aura/cursor.h (right):

http://codereview.chromium.org/9463003/diff/6042/ui/aura/cursor.h#newcode127
ui/aura/cursor.h:127: bool operator==(int type) const { return native_type_ ==
type; }
This feels a bit weird to me.  I think that it'd be better to do explicit
comparisons with native_type() instead.

http://codereview.chromium.org/9463003/diff/6042/ui/aura/cursor.h#newcode142
ui/aura/cursor.h:142: PlatformCursor platform_;
nit: rename to platform_cursor_

[sadrul, 2012-03-27 21:34:18]

Hi! I have moved all of the aura-cursor related code into ui/base/cursor/. This
way, webkit does not have to depend on aura (or any new dependency), and we
avoid a cyclic dependency (since aura depends on webkit for the compositor).
Does this sound like a reasonable approach?

[Daniel Erat, 2012-03-27 22:15:05]

It seems a bit strange to me to have code in ui/base that's only applicable to
Aura so far, but I'll defer to Ben on this.

http://codereview.chromium.org/9463003/diff/22001/ui/aura/root_window.h
File ui/aura/root_window.h (right):

http://codereview.chromium.org/9463003/diff/22001/ui/aura/root_window.h#newco...
ui/aura/root_window.h:14: #include "ui/base/cursor/cursor.h"
nit: fix alphabetization

[sadrul, 2012-03-27 22:19:10]

On 2012/03/27 22:15:05, Daniel Erat wrote:
> It seems a bit strange to me to have code in ui/base that's only applicable to
> Aura so far, but I'll defer to Ben on this.

I thought about it too, but we do have X/GTK/win-only things in ui/base/, and
perhaps it would be acceptable to have some aura-only code in there too!

If we can do this without having aura-only code in ui/base/ and without
introducing aura<-->webkit circular dependency, then I will happily take that
route.

[sadrul, 2012-03-27 22:26:12]

http://codereview.chromium.org/9463003/diff/22001/ui/aura/root_window.h
File ui/aura/root_window.h (right):

http://codereview.chromium.org/9463003/diff/22001/ui/aura/root_window.h#newco...
ui/aura/root_window.h:14: #include "ui/base/cursor/cursor.h"
On 2012/03/27 22:15:05, Daniel Erat wrote:
> nit: fix alphabetization

Done (used tools/sort-headers.py, and it fixed a few other places)

[Rick Byers, 2012-03-28 14:58:59]

As requested I tested this CL out with a few of the example pages I was using
for custom web cursors.  It worked great for me (colors looked identical to
Linux too).  Thanks Sadrul!

[sadrul, 2012-04-03 15:52:49]

ping!

[darin (slow to review), 2012-04-03 17:55:03]

LGTM for src/webkit

[Ben Goodger (Google), 2012-04-03 18:08:18]

http://codereview.chromium.org/9463003/diff/22002/ui/base/cursor/cursor.h
File ui/base/cursor/cursor.h (right):

http://codereview.chromium.org/9463003/diff/22002/ui/base/cursor/cursor.h#new...
ui/base/cursor/cursor.h:105: RefCustomCursor();
please do all of this in cursor.cc

[sadrul, 2012-04-03 19:16:34]

(patchset13 is patchset12 merged with ToT. patchset14 has the requested change)

http://codereview.chromium.org/9463003/diff/22002/ui/base/cursor/cursor.h
File ui/base/cursor/cursor.h (right):

http://codereview.chromium.org/9463003/diff/22002/ui/base/cursor/cursor.h#new...
ui/base/cursor/cursor.h:105: RefCustomCursor();
On 2012/04/03 18:08:18, Ben Goodger (Google) wrote:
> please do all of this in cursor.cc

Done.

[Ben Goodger (Google), 2012-04-03 19:29:02]

lgtm