1395059 - [e10s] Fetching table cell other than first with table interfaces returns broken object

Status: RESOLVED WORKSFORME

(Core :: Disability Access APIs, defect, P1)

Description

[James Teh [:Jamie]]

Version: 57.0a1 (2017-08-29) (32-bit)

STR (with NVDA):
1. Open this test case: data:text/html,<table><tr><th>a</th><th>b</th></tr><tr><td>c</td><td>d</td></tr></table>
2. Use the cursor keys to move to the line containing "c".
3. Press control+alt+rightArrow to move to the cell in the next column (d).
Expected: NVDA says "b  column 2  d".
Actual: NVDA says "edge of table".

Everything works for the first cell in the table (row 1 column 1). However, for any cell after that, things break. When called with either row >= 1 or column >= 1 (since they are 0-based), IAccessibleTable::accessibleAt and IAccessibleTable2::cellAt do return an IUnknown as expected, but QueryInterface to IAccessible2 (or IAccessibleTableCell) returns E_NOINTERFACE. As expected, they still return E_INVALIDARG if you pass row/column indexes that are out of bounds.

This is pretty serious, as it breaks table navigation for NVDA users.
Everything works as expected with e10s disabled.
This is not a recent regression; I just haven't tracked it down properly before now.

Comment 1

[Marco Zehe (:MarcoZ)]

I can confirm this. Marking as blocking E10S shipment. Aaron, is this an indication of more generally having a problem with positive indexes that are 0-based, or is this specific to the IAccessibleTable* interfaces?

Comment 2

[(No longer employed by Mozilla) Aaron Klotz]

Acknowledged. I'll diagnose this asap.

Comment 3

[(No longer employed by Mozilla) Aaron Klotz]

OK, I think this should go away when bug 1395329 lands. The test case is working for me with that patch applied.

Comment 4

[Marco Zehe (:MarcoZ)]

Jamie, can you confirm that this works as expected now in the latest Nightly, and close the bug as WFM if that's the case?

Comment 5

[James Teh [:Jamie]]

Confirmed worksforme in 57.0a1 (2017-09-03) (32-bit).