[webkit-gtk] enable-accelerated-compositing flag does not work correctly

[Jianhua Shao]

On Thu, Apr 25, 2013 at 2:24 PM, Jianhua Shao <alex.sjh at gmail.com> wrote:

> On Thu, Apr 25, 2013 at 1:58 PM, Emmanuel Rodriguez <
> emmanuel.rodriguez at gmail.com> wrote:
>
>> On 25 April 2013 03:37, Jianhua Shao <alex.sjh at gmail.com> wrote:
>>
>>> On Thu, Apr 25, 2013 at 5:32 AM, Emmanuel Rodriguez <
>>> emmanuel.rodriguez at gmail.com> wrote:
>>>
>>>>
>>>>
>>>> Thanks for sharing your experience.
>>>>
>>> The version 1.6.3 does not meet my requirements.
>>> "enable-accelerated-compositing" is what I need in the newer version, from
>>> the dev document I found it is available from version 1.7.5.
>>> Seems I can give a try on version 1.7.x.
>>>
>> I just ment to try it and see if the transparencies where still enabled
>> in 1.6.3
>>
>>
>>>
>>> And have you read my another mail,
>>>
>>> cairo_surface_t* surface = webkit_web_view_get_snapshot(view);
>>> cairo_surface_write_to_png(surface, "/home/meego/dump.png");
>>>
>>> will result a transparent png file...
>>> Maybe I need to do some configuration on cario/gtk, but I am new on this
>>> subject, anyone can give me some hint?
>>>
>>
>> By the looks of the code of 1.6.3 and 1.10.2 it seems that
>>  webkit_web_view_get_snapshot() behaves like the old webkit_web_view_draw().
>>
>> In 1.6.3 the draw operation is done with:
>>
>> static void paintWebView(Frame* frame, gboolean transparent,
>> GraphicsContext& context, const IntRect& clipRect, const Vector<IntRect>&
>> rects)
>> {
>>     bool coalesce = true;
>>
>>     if (rects.size() > 0)
>>         coalesce = shouldCoalesce(clipRect, rects);
>>
>>     if (coalesce) {
>>         context.clip(clipRect);
>>         if (transparent)
>>             context.clearRect(clipRect);
>>         frame->view()->paint(&context, clipRect);
>>     } else {
>>         for (size_t i = 0; i < rects.size(); i++) {
>>             IntRect rect = rects[i];
>>             context.save();
>>             context.clip(rect);
>>             if (transparent)
>>                 context.clearRect(rect);
>>             frame->view()->paint(&context, rect);
>>             context.restore();
>>         }
>>     }
>>
>>     context.save();
>>     context.clip(clipRect);
>>     frame->page()->inspectorController()->drawHighlight(context);
>>     context.restore();
>> }
>>
>>
>> In 1.10.2 webkit_web_view_get_snapshot is done with:
>>
>> cairo_surface_t*
>> webkit_web_view_get_snapshot(WebKitWebView* webView)
>> {
>>     g_return_val_if_fail(WEBKIT_IS_WEB_VIEW(webView), 0);
>>
>>     Frame* frame = core(webView)->mainFrame();
>>     if (!frame || !frame->contentRenderer() || !frame->view())
>>         return 0;
>>
>>     frame->view()->updateLayoutAndStyleIfNeededRecursive();
>>     GtkAllocation allocation;
>>     gtk_widget_get_allocation(GTK_WIDGET(webView), &allocation);
>>     cairo_surface_t* surface =
>> cairo_image_surface_create(CAIRO_FORMAT_ARGB32, allocation.width,
>> allocation.height);
>>     RefPtr<cairo_t> cr = adoptRef(cairo_create(surface));
>>     GraphicsContext gc(cr.get());
>>
>>     IntRect rect = allocation;
>>     gc.applyDeviceScaleFactor(frame->page()->deviceScaleFactor());
>>     gc.save();
>>     gc.clip(rect);
>>     if (webView->priv->transparent)
>>         gc.clearRect(rect);
>>     frame->view()->paint(&gc, rect);
>>     gc.restore();
>>
>>     return surface;
>> }
>>
>>
>> Both functions have the same drawing logic:
>>
>>         context.clip(clipRect);
>>         if (transparent)
>>             context.clearRect(clipRect);
>>         frame->view()->paint(&context, clipRect);
>>         context.restore();
>>
>> Wit the exception that webkit_web_view_get_snapshot() forces the drawing
>> into an image surface except of accepting a cairo context.
>>
>> Yeah, I have modified "webkit_web_view_expose_event" (the
> webkit_web_view_draw thing for gtk2), and I have made it transparent now.
>
> Then I tried setting 'enable-accelerated-compositing' flag. the result
> page I got is not correct. The appearance is a little wired I think.
> Firstly, a correct page shows up, then immediately after that (I mean,
> maybe 0.1 second after that), the page turns gray in most part, and stay at
> gray.
>
> The issue exists even if I did not modify "webkit_web_view_expose_event".
>
> Now I am trying to figure out whether something wrong with
> "WEBKIT_WEB_VIEW(widget)->priv->backingStore->cairoSurface()".
>
>
Here I conclude what has been modified in webkit_web_view_expose_event.
1. create a GraphicsContext, and call gc.clearRect() on gtkwidget's cario
surface if transparent flag is set.
    This modification will give me a transparent web page.
2. I print the size of backingStore cario_surface, I found if
'enable-accelerated-compositing' flag is set,
    the size of the backingstore cario_surface sometimes would be 1x1.
Which means, the whole page will be cleared, and then it is written by just
1 pixel...
    I guess this is the reason why "Firstly, a correct page shows up, then
immediately after that (I mean, maybe 0.1 second after that), the page
turns gray in most part, and stay at gray."


> --
>> Emmanuel Rodriguez
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.webkit.org/pipermail/webkit-gtk/attachments/20130425/f4df86cc/attachment-0001.html>