REGRESSION: Reproducible assert while loading this test file if css is already in the cache

[Grace Kloba]

Load LayoutTests/fast/dom/css-insert-import-rule.html first to get the resource in the cache.

Change address to be LayoutTests/fast/dom/css-insert-import-rule.txt. As there is no existing file, you should see an error page.

Now change address back to LayoutTests/fast/dom/css-insert-import-rule.html, you should see assertion like this,

WebCore/dom/Document.cpp:1878: failed assertion `m_pendingStylesheets > 0'

[David Kilzer (:ddkilzer)]

Confirmed with local debug build of WebKit r19972 with Safari 2.0.4 (419.3) on Mac OS X 10.4.8 (8L127).

This is a regression from shipping Safari 2.0.4 (419.3) on Mac OS X 10.4.8 (8L127).

Console output on debug build (CrashReporter does not launch!):

/path/to/WebKit/WebCore/dom/Document.cpp:1894: failed assertion `m_pendingStylesheets > 0'
Abort trap

[Mark Rowe (bdash)]

<rdar://problem/5045718>

(In reply to comment #1)
> Console output on debug build (CrashReporter does not launch!):

Evil lowercase assert()s!

Created attachment 13568 [details]
Sample fix

The problem is that the pending stylesheet counter and the stylesheet's loadCompleted() flag get out of sync. This patch fixes the test case, but I noticed several other calls to stylesheetLoaded() that need to be tested and possibly fixed...

Created attachment 13573 [details]
Keep loadCompleted() in sync with the pending stylesheet count

[Darin Adler]

Comment on attachment 13573 [details]
Keep loadCompleted() in sync with the pending stylesheet count

r=me

[David Kilzer (:ddkilzer)]

Comment on attachment 13573 [details]
Keep loadCompleted() in sync with the pending stylesheet count

Why is the m_sheet null check kept in one place, but not the other two?  In both methods were the null check is not kept, m_sheet is used previously in the method, but how do we know its value won't change between uses?

Not kept:

>Index: WebCore/dom/ProcessingInstruction.cpp
>-    // Tell the doc about the sheet.
>-    if (!isLoading() && m_sheet)
>-        document()->stylesheetLoaded();
>+    m_sheet->checkLoaded();
> }

Kept:

>Index: WebCore/dom/StyleElement.cpp
>-    if (!isLoading() && m_sheet)
>-        document->stylesheetLoaded();
>+    if (m_sheet)
>+        m_sheet->checkLoaded();
> }

Not kept:

>Index: WebCore/html/HTMLLinkElement.cpp
>-    // Tell the doc about the sheet.
>-    if (!isLoading() && m_sheet && !isDisabled() && !isAlternate())
>-        document()->stylesheetLoaded();
>+    m_sheet->checkLoaded();
> }

(In reply to comment #7)
> Why is the m_sheet null check kept in one place, but not the other two?

Because it's needed in that one place (in case the type attribute is wrong or the media doesn't match) and not in the other two.

> In
> both methods were the null check is not kept, m_sheet is used previously in the
> method, but how do we know its value won't change between uses?

In HTMLLinkElement it's obvious (setMedia() cannot get m_sheet changed). In ProcessingInstruction it's quite easy to verify that it cannot become 0.

[David Kilzer (:ddkilzer)]

Committed revision 20098.