Bug 155431 – Leak: Accelerated ImageBufferCairo doesn't destroy the used textures

• Status: RESOLVED
• Resolution: FIXED
• Component: Canvas

[Miguel Gomez]

When using the accelerated path, ImageBufferCairo creates up to two textures for rendering: m_texture to hold the buffer content, and m_compositorTexture used to take the content to the ThreadedCompositor (when enabled).

These 2 textures are passed to cairo to create a cairo_surface using cairo_gl_surface_create_for_texture(), but cairo doesn't take their ownership, so currently they are not being destroyed when the ImageBuffer gets destroyed.

[Miguel Gomez]

Created attachment 273958 [details]
Patch

[Zan Dobersek]

Comment on attachment 273958 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=273958&action=review

> Source/WebCore/platform/graphics/cairo/ImageBufferCairo.cpp:88
> +    if (!m_texture)
> +        return;

I assume m_texture has to be present for m_compositorTexture to be created, but it would be better not to count on that, and just either delete m_texture or m_compositorTexture if they are non-null.

> Source/WebCore/platform/graphics/cairo/ImageBufferCairo.cpp:90
> +    GLContext::sharingContext()->makeContextCurrent();

This destructor should store the previous context and restore it when done, as we usually do.

At some point, we should also use a separate offscreen context for the accelerated canvas, and not the sharing context.

> Source/WebCore/platform/graphics/cairo/ImageBufferDataCairo.h:58
> +    ~ImageBufferData();

This should be virtual.

[Miguel Gomez]

(In reply to comment #2)
> Comment on attachment 273958 [details]
> Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=273958&action=review
> 
> > Source/WebCore/platform/graphics/cairo/ImageBufferCairo.cpp:88
> > +    if (!m_texture)
> > +        return;
> 
> I assume m_texture has to be present for m_compositorTexture to be created,
> but it would be better not to count on that, and just either delete
> m_texture or m_compositorTexture if they are non-null.

Yes, using an accelerated path m_texture must exist and m_compositorTexture may or may not. I'll simplify it.

> > Source/WebCore/platform/graphics/cairo/ImageBufferCairo.cpp:90
> > +    GLContext::sharingContext()->makeContextCurrent();
> 
> This destructor should store the previous context and restore it when done,
> as we usually do.

Ok.

> At some point, we should also use a separate offscreen context for the
> accelerated canvas, and not the sharing context.

Agree.

> > Source/WebCore/platform/graphics/cairo/ImageBufferDataCairo.h:58
> > +    ~ImageBufferData();
> 
> This should be virtual.

Ok.

[Build Bot]

Comment on attachment 273958 [details]
Patch

Attachment 273958 [details] did not pass mac-debug-ews (mac):
Output: http://webkit-queues.webkit.org/results/977298

New failing tests:
js/function-apply.html

[Build Bot]

Created attachment 273974 [details]
Archive of layout-test-results from ews114 for mac-yosemite

The attached test failures were seen while running run-webkit-tests on the mac-debug-ews.
Bot: ews114  Port: mac-yosemite  Platform: Mac OS X 10.10.5

[Miguel Gomez]

Created attachment 273983 [details]
Patch

[Zan Dobersek]

Comment on attachment 273983 [details]
Patch

LGTM. Will wait for the EFL EWS to build, and cq+ after that.

[WebKit Commit Bot]

Comment on attachment 273983 [details]
Patch

Clearing flags on attachment: 273983

Committed r198205: <http://trac.webkit.org/changeset/198205>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.