81454 - Active Accessibility: expose tag name and attribs

Status: RESOLVED FIXED

(SeaMonkey :: General, defect, P1)

Description

[Aaron Leventhal]

Active accessibility's IAccessible  interface doesn't allow screen reader 
vendors to get the actual tag name and attributes of a component.

Allow a QueryInterface from IAccessible to our own component, IAccessibleMoz, 
that has the tag name, attributes and possibly style info for a node.

Comment 1

[Aaron Leventhal]

Attachment: Patch v1.0

Comment 2

[Aaron Leventhal]

I also have the original AccessibleMoz.idl file.
It imports oleacc.idl, which we can't distribute. You have to get it from the 
Microsoft Platform SDK.

It generates AccessibleMoz.h and AccessibleMoz_i.c, both of which are used here 
and needed by vendors who use IAccessibleMoz.

It also builds the proxy stub .dll, which an installer needs to call 
regsvr32.exe on if this interface is to be used across process.

That creates a problem, because now either we have to prebuild the .dll and 
include that in the tree somehow, or we have to convice Microsoft to let us 
distribute oleacc.dll with their source.

Anyway where should I put this stuff? We need to store it somewhere.
How about widget/src/windows/accessiblemoz_idl

Comment 3

[Aaron Leventhal]

There have been big changes
The new public interface is called IMozNode. It basically mirrors the Mozilla 
content DOM, but accross process boundaries. 

It can be built as part of the mozilla project from .idl files.
I'll need a review from leaf or granrose or someone, otherwise I might break 
some weird Windows platform.

Comment 4

[Aaron Leventhal]

There are potential build system breakages in this patch.

Leaf, when can you take a look at it>

Comment 5

[Jon Granrose]

that may have to wait til after 0.9.1 as leaf has a few 0.9.1 bugs to finish and
then we'll be in the 0.9.1 maelstrom.

Comment 6

[Aaron Leventhal]

Attachment: Patch v. 1.2

Comment 7

[Aaron Leventhal]

Leaf says dveditz can review the makefile.win file in this patch.
I need this right away, to work with another software development company.

dveditz, are you available?

Comment 8

[Aaron Leventhal]

Attachment: Patch v.1.3 - includes evaughan's suggestions

Comment 9

[Eric Vaughan]

1) Don't use NS_WITH_SERVICE there is a different way. Look 
at nsTableCell::GetAccessible to see how its done. 

2) nsDOMTreeWalker::GetAccessible is called a lot. Its called on every node in 
the tree. It could be potentially expensive to get the service each time. We 
should see what the overhead is. I rather would have the DOMTreeWalker hold a 
cached nsCOMPtr to the service that is set on construction. To make this call 
more efficient.

get_firstChild, get_lastChild, get_previousSibling, get_nextSibling, don't use 
MakeMozNode but should.

Comment 10

[Aaron Leventhal]

Attachment: Patch v1.4

Comment 11

[Eric Vaughan]

r=evaughan

Comment 12

[Daniel (Leaf) Nunes]

i've sent a makefile.win for the new expose directory, when that makes it into
the patch, r=leaf

Comment 13

[Aaron Leventhal]

Attachment: Patch v1.5 - thanks Leaf for the improved makefile.win

Comment 14

[Chris Waterson]

Could you please summarize your changes for someone who isn't really all that
familiar with accessability? Also, please enumerate the new build- and run-time
dependencies that you're adding (have you tested this on Win95? Win98? NT4.0?).
If there are new dependencies, can we add a switch to _not_ build this stuff?
(I'm nervous about the ``need file X from the platform SDK'' comment, above.
That could make it hard for a lot of people to build this, no?)

Comment 15

[Daniel (Leaf) Nunes]

I managed to build this on win98, and i'm not sure i even have any supplemental
sdk's installed.

It'd be nice to verify this on a machine that is *known* for certain to not have
supplemental sdks, but i'm not sure where such a machine lives.

Comment 16

[Aaron Leventhal]

Chris,

It turns out we don't need that file from the platform SDK anymore.

What I've done is built an IMozNode interface that you can QI to from an 
IAccessible. Normally what happens is: a screen reader or some other type of 
accessibility aid sends us a WM_GETOBJECT window message, and we return an 
IAccessible, implemented by Accessible.cpp. This gives them basic information 
about some nodes in our content tree, such as role, name and state - sometimes 
it has to traverse the frame tree instead. It's kind of a hybrid 
between the two - not all nodes are exposed, and not all data are 
exposed. You don't get attributes or tag names or styles. You get 
whatever Microsoft thought you should get for a widget - basically 
predifined enums that the raw data has been funneled into. 

Accessible.cpp just proxies to nsAccessible.cpp to get all of that  - which is 
in the Accessibility.dll. That doesn't get loaded until this whole process gets 
set in motion.

All of that stuff works already and is in the tree. This change is in here, 
because for some vendors, IAccessible doesn't have enough - I needed to create 
an MS COM interface so that they could grab exposed information about the actual 
DOM nodes in our content tree. That way they can get attributes, tag names and 
style data. 

This interface is called IMozNode which is implemented by MozNode.cpp in the 
class MozNode. Now Accessible just inherits from MozNode. MozNode lets you walk 
the entire DOM tree, not just the selected pieces that IAccessible allowed.

Comment 17

[Chris Waterson]

Ok, makes sense. Thanks. sr=waterson

Comment 18

[Peter Trudelle]

What is holding up checkin of this patch?

Comment 19

[Aaron Leventhal]

It's part of our branch, being sr'd by waterson

Comment 20

[Aaron Leventhal]

In trunk and branch.