Bug 4274 – REGRESSION: canvas draws when window loses focus (flipped and in wrong position!)

• Component: Layout and Rendering
• Status: RESOLVED FIXED

[Todd Yandell]

In the latest version of WebKit the canvas element has a rendering error. It appears to be copying the 
rendered image on top of everything else, only it's drawn from another coordinate system (where [0, 0] is 
at the lower left hand corner, the default for Quartz), causing the image to be flipped over. The image only 
appears to get copied when the Safari window loses or gains focus. This doesn't happen in WebKit 412, 
and it doesn't appear to happen all of the time. It will work for a while, then stop. Restarting Safari causes 
the bug to show up again.

[Todd Yandell]

Created attachment 3214 [details]
A test case for this bug.

[Todd Yandell]

Created attachment 3215 [details]
Screenshot of the bug.

[Chris Petersen]

Confirming this issue only happens with TOT WebKit.

[Darin Adler]

When this bug does NOT happen, we get this on the console:

    CGContextDrawImage: invalid context

This makes me think that there's some code path that draws the image into the current context and the 
context is not set up.

[Darin Adler]

I figured it out. This regression was caused by the recent form element work.

[Darin Adler]

Created attachment 3763 [details]
patch to make sure canvas doesn't draw when QPainter has painting disabled

[Darin Adler]

Created attachment 3765 [details]
patch to make sure nobody tries to use the CGContext when painting disabled

[Darin Adler]

Comment on attachment 3765 [details]
patch to make sure nobody tries to use the CGContext when painting disabled

This patch should fix both this bug and bug 4755.

[John Sullivan]

Comment on attachment 3765 [details]
patch to make sure nobody tries to use the CGContext when painting disabled

I see, this one is more complete than the previous one. That's why the previous
one had the review ? flag removed.

[Geoffrey Garen]

I just hit this assertion failure a number of times with TOT: 

QPainter::currentContext(): ASSERT(!data->state.paintingDisabled);

EXC_BAD_ACCESS (0x0001)
KERN_INVALID_ADDRESS (0x0001) at 0xbbadbeef

Thread 0 Crashed:
0    QPainter::currentContext() + 96 (KWQPainter.mm:841)
1    QPainter::save() + 92 (KWQPainter.mm:181)
2    khtml::InlineTextBox::paintSelection(QPainter*, int, int, khtml::RenderStyle*, khtml::Font const*) + 
364 (render_text.cpp:524)
3    khtml::InlineTextBox::paint(khtml::RenderObject::PaintInfo&, int, int) + 1220 (render_text.cpp:330)
4    khtml::InlineFlowBox::paint(khtml::RenderObject::PaintInfo&, int, int) + 976 (render_line.cpp:785)
5    khtml::RootInlineBox::paint(khtml::RenderObject::PaintInfo&, int, int) + 60 (render_line.cpp:1197)
6    khtml::RenderFlow::paintLines(khtml::RenderObject::PaintInfo&, int, int) + 2004 (render_flow.cpp:
373)
7    khtml::RenderBlock::paintObject(khtml::RenderObject::PaintInfo&, int, int) + 436 (render_block.cpp:
1311)
8    khtml::RenderBlock::paint(khtml::RenderObject::PaintInfo&, int, int) + 696 (render_block.cpp:1240)
9    khtml::RenderBlock::paintChildren(khtml::RenderObject::PaintInfo&, int, int) + 808 
(render_block.cpp:1262)

Since this checkin last modified that assertion, I'm reopening this bug.

[Geoffrey Garen]

I'm reclosing this bug because the assertion failure is being tracked by 4985.