387347 - Text with "float" style inaccessible effective 2007070605 build

Status: RESOLVED FIXED

(Core :: Disability Access APIs, defect)

Description

[Joanmarie Diggs]

Attachment: test case

Steps to reproduce:

1. Open the attached test case
2. Use Accerciser to examine the children of the document frame

Prior to the 2007070605 build, you will find three children:  The first and third paragraphs have ROLE_PARAGRAPH; the second has ROLE_TEXT.  Beginning with the 2007070605 build, the second paragraph is completely absent from the hierarchy.

Comment 1

[Scott Haeger]

Attachment: test case 2

Another test case which shows that the problem may lie with the 'position' attribute.  This test case shows that removing 'position:absolute' will also destroy the hierarchy.  The original broken case can be found here http://www.mozilla.org/access/dhtml/class/tabpanel

Comment 2

[Ginn Chen]

Thanks for reporting this!

I met this problem with Mozilla project page a few days ago.

Comment 3

[Ginn Chen]

The regression is gone if I back out following piece. It's in bug 377783

accessible/src/base/nsAccessibilityService.cpp

@@ -1367,10 +1384,14 @@ NS_IMETHODIMP nsAccessibilityService::Ge
         }
       }
     }
 
     if (tryFrame) {
+      if (frame->GetRect().IsEmpty()) {
+        *aIsHidden = PR_TRUE;
+        return NS_OK;
+      }
       frame->GetAccessible(getter_AddRefs(newAcc)); // Try using frame to do it
     }
   }
 
   if (!newAcc) {

Comment 4

[Ginn Chen]

Attachment: patch

GetRect doesn't reliable for nsPlaceholderFrame 

Can we back out this hunk?

Comment 5

[Evan Yan]

If we just simply back out this, invisible <caption>s in html table will be wrongly created.

Comment 6

[Ginn Chen]

Then should we check empty Rect in nsTableCaptionFrame::GetAccessible?

Comment 7

[Ginn Chen]

I think we already checked that in nsTableCaptionFrame::GetAccessible.

Comment 8

[Aaron Leventhal]

Comment on attachment 271644 [details] [diff] [review]
patch

As Evan says, we should just do the check for captions now.

But, we need to do it so that *aIsHidden is set to TRUE, because that also prevents the children from getting accessibles created (the whole subtree is skipped when *aIsHidden = PR_TRUE -- it's an out param).

Comment 9

[Aaron Leventhal]

Comment on attachment 271644 [details] [diff] [review]
patch

Would this work?

if (frame->GetRect().IsEmpty() && !placeholderframe) {
  *aIsHidden = PR_TRUE; return NS_OK;
}

Comment 10

[Ginn Chen]

Attachment: patch v2

Comment 11

[Ginn Chen]

The patch didn't solve
411.com and
http://www.sfgate.com/cgi-bin/article.cgi?f=/chronicle/archive/2007/07/29/SP13R95O71.DTL

frame is nsBlockFrame

Comment 12

[Ginn Chen]

Attachment: patch for 411.com and sfgate

Let's do a blacklist rather than whitelist, ok?

Comment 13

[Aaron Leventhal]

Please add a comment why we do this for table caption frame, something like:

// Do not create accessible object subtrees for non-rendered table captions. This could not be done in nsTableCaptionFrame::GetAccessible() because the descendants of the table caption would still be created. By setting *aIsHidden = PR_TRUE we ensure that no descendant accessibles are created

Comment 14

[Aaron Leventhal]

Checked in with added comment.

Comment 15

[Aaron Leventhal]

I believe this fixed bug 390670 as well.