Bug 100268 – MathML tests trigger font cache assertions in debug bots

• Status: RESOLVED
• Resolution: FIXED
• Component: Tools / Tests

[Levi Weintraub]

Same sort of backtrace on both:

STDERR: ASSERTION FAILED: m_purgePreventCount
STDERR: /Volumes/data/b/build/slave/webkit-mac-latest-dbg/build/src/third_party/WebKit/Source/WebCore/WebCore.gyp/../platform/graphics/FontCache.cpp(317) : PassRefPtr<WebCore::SimpleFontData> WebCore::FontCache::getCachedFontData(const WebCore::FontPlatformData *, WebCore::FontCache::ShouldRetain)
STDERR: 1   0x12b11cb9 WebCore::FontCache::getCachedFontData(WebCore::FontPlatformData const*, WebCore::FontCache::ShouldRetain)
STDERR: 2   0x12c3c6d9 WebCore::FontCache::getFontDataForCharacters(WebCore::Font const&, unsigned short const*, int)
STDERR: 3   0x12b2f3de WebCore::Font::glyphDataAndPageForCharacter(int, bool, WebCore::FontDataVariant) const
STDERR: 4   0x12b323b3 WebCore::Font::glyphDataForCharacter(int, bool, WebCore::FontDataVariant) const
STDERR: 5   0x13f087a6 WebCore::RenderMathMLOperator::glyphHeightForCharacter(unsigned short)
STDERR: 6   0x13f08b53 WebCore::RenderMathMLOperator::updateFromElement()
STDERR: 7   0x13f08620 WebCore::RenderMathMLOperator::styleDidChange(WebCore::StyleDifference, WebCore::RenderStyle const*)
STDERR: 8   0x13e14f48 WebCore::RenderObject::setStyle(WTF::PassRefPtr<WebCore::RenderStyle>)
STDERR: 9   0x13e1456c WebCore::RenderObject::setAnimatableStyle(WTF::PassRefPtr<WebCore::RenderStyle>)
STDERR: 10  0x124d33e7 WebCore::Node::setRenderStyle(WTF::PassRefPtr<WebCore::RenderStyle>)
STDERR: 11  0x1245e685 WebCore::Element::recalcStyle(WebCore::Node::StyleChange)
STDERR: 12  0x1245eb08 WebCore::Element::recalcStyle(WebCore::Node::StyleChange)
STDERR: 13  0x1245eb08 WebCore::Element::recalcStyle(WebCore::Node::StyleChange)
STDERR: 14  0x1245eb08 WebCore::Element::recalcStyle(WebCore::Node::StyleChange)
STDERR: 15  0x1245eb08 WebCore::Element::recalcStyle(WebCore::Node::StyleChange)
STDERR: 16  0x1245eb08 WebCore::Element::recalcStyle(WebCore::Node::StyleChange)
STDERR: 17  0x12396a1c WebCore::Document::recalcStyle(WebCore::Node::StyleChange)
STDERR: 18  0x12392e67 WebCore::Document::styleResolverChanged(WebCore::StyleResolverUpdateFlag)
STDERR: 19  0x13b36cf9 WebCore::XMLDocumentParser::end()
STDERR: 20  0x13b36e3d WebCore::XMLDocumentParser::finish()
STDERR: 21  0x1386a4bd WebCore::DocumentWriter::end()
STDERR: 22  0x13844c0c WebCore::DocumentLoader::finishedLoading()
STDERR: 23  0x138ab56b WebCore::MainResourceLoader::didFinishLoading(double)
STDERR: 24  0x138cfe4f WebCore::ResourceLoader::didFinishLoading(WebCore::ResourceHandle*, double)
STDERR: 25  0x12ce35aa WebCore::ResourceHandleInternal::didFinishLoading(WebKit::WebURLLoader*, double)
STDERR: 26  0x1c64ddfe webkit_glue::WebURLLoaderImpl::Context::OnCompletedRequest(int, bool, std::string const&, base::TimeTicks const&)
STDERR: 27  0x1108a3ad (anonymous namespace)::RequestProxy::NotifyCompletedRequest(int, std::string const&, base::TimeTicks const&)
STDERR: 28  0x1108abc7 base::internal::RunnableAdapter<void ((anonymous namespace)::RequestProxy::*)(int, std::string const&, base::TimeTicks const&)>::Run((anonymous namespace)::RequestProxy*, int const&, std::string const&, base::TimeTicks const&)
STDERR: 29  0x1108aac6 base::internal::InvokeHelper<false, void, base::internal::RunnableAdapter<void ((anonymous namespace)::RequestProxy::*)(int, std::string const&, base::TimeTicks const&)>, void ()((anonymous namespace)::RequestProxy* const&, int const&, std::string const&, base::TimeTicks const&)>::MakeItSo(base::internal::RunnableAdapter<void ((anonymous namespace)::RequestProxy::*)(int, std::string const&, base::TimeTicks const&)>, (anonymous namespace)::RequestProxy* const&, int const&, std::string const&, base::TimeTicks const&)
STDERR: 30  0x1108aa05 base::internal::Invoker<4, base::internal::BindState<base::internal::RunnableAdapter<void ((anonymous namespace)::RequestProxy::*)(int, std::string const&, base::TimeTicks const&)>, void ()((anonymous namespace)::RequestProxy*, int, std::string const&, base::TimeTicks const&), void ()((anonymous namespace)::RequestProxy*, int, std::string, base::TimeTicks)>, void ()((anonymous namespace)::RequestProxy*, int, std::string const&, base::TimeTicks const&)>::Run(base::internal::BindStateBase*)

[Dave Barton]

Levi, can you tell which revision caused this? When did it start? (Is it often enough to tell?)

[Philip Rogers]

(In reply to comment #1)
> Levi, can you tell which revision caused this? When did it start? (Is it often enough to tell?)

Dave,
Sometimes this is noisy, but it might help in tracking this down: http://test-results.appspot.com/dashboards/flakiness_dashboard.html#tests=mathml%2Fpresentation%2Fmo-stretch.html

[Dave Barton]

Thanks to pdr and leviw. I see the flaky dashboard in Chrome now. Is there a way I can look at the stacktraces like Levi posted in comment 0?

[Levi Weintraub]

(In reply to comment #3)
> Thanks to pdr and leviw. I see the flaky dashboard in Chrome now. Is there a way I can look at the stacktraces like Levi posted in comment 0?

Click on "Show results" when on that page :)

[Dave Barton]

Those are the two tests with stretched () [] and integral sign at *various sizes*. The assertion failures only occur on Snow Leopard (10.6), where there are no STIX fonts on the bots. There are other fonts that provide the needed extender glyphs though. Actually, I see now that style-changed.html and subsup.xhtml and mo.xhtml have crashed also, though less often. These have stretched parens also, though sometimes at just one size.

As Levi says, the backtraces are all similar. In the most recent failures, Levi's 17 Document::recalcStyle got called [via WebCore::Document::updateStyleIfNeeded()] inside Document::styleRecalcTimerFired in the 3 rarer crashes, and inside Document::finishedParsing() for mo-stretch.html, and as in comment 0 for row.xhtml.

I believe the fix for this is to add a FontCachePurgePreventer to RenderMathMLOperator::glyphHeightForCharacter. I will submit a patch for this, but really need a font or at least WebKit expert to review it. Thanks in advance!

[Dave Barton]

Would it be better to put the FontCachePurgePreventer in RenderMathMLOperator::glyphHeightForCharacter's caller? Is it possible we're going to trigger a lot of slow purges, and should worry about minimizing that?

[Philip Rogers]

(In reply to comment #6)
> Would it be better to put the FontCachePurgePreventer in RenderMathMLOperator::glyphHeightForCharacter's caller? Is it possible we're going to trigger a lot of slow purges, and should worry about minimizing that?

+schenney
Schenney and Eseidel recently did some work on FontCachePurgePreventer.

[Eric Seidel (no email)]

schenney is your man.

[Levi Weintraub]

Changing the bug title as 2 more tests have exhibited this flake now: http://test-results.appspot.com/dashboards/flakiness_dashboard.html#tests=mathml%2Fpresentation%2Fmroot-pref-width.html%2Cmathml%2Fpresentation%2Fstyle-changed.html

[Dave Barton]

Created attachment 170485 [details]
Patch

[Eric Seidel (no email)]

Comment on attachment 170485 [details]
Patch

LGTM.

[Dave Barton]

Thanks, Eric. Apparently schenney is in an airplane. [Seems kind of extreme just to avoid reviewing one of my patches. :) ]

[WebKit Review Bot]

Comment on attachment 170485 [details]
Patch

Clearing flags on attachment: 170485

Committed r132424: <http://trac.webkit.org/changeset/132424>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.