168107 - [ATK Accessibility] better support for editable text

Status: RESOLVED FIXED

(Core :: Disability Access APIs, defect)

Description

[Kyle Yuan]

We need to distinguish *insert* and "delete" actions, get the start position of 
inserted/deleted text and the length.
This issue was remained by bug 164763.

Comment 1

[Kyle Yuan]

sfraser, is it possible that fire a dom event before *and* after the editor's 
text changing? Now we only fire a "input" event after typing.
If that's impossible, can I get the previous text, I mean the text before 
typing, then do a string compare? How can I get that text, from the editor's 
undo buffer?

Comment 2

[Simon Fraser [no longer active]]

I have no idea.

Comment 3

[Kyle Yuan]

Attachment: use nsIEditActionListener to track the text change

1) use mEditor and mTextFrame to cache the nsIEditor and nsITextControlFrame
instead of calling MakeSelection() every time;
2) nsHTMLTextFieldAccessible inherits from nsIEditActionListener, so we can
catch the text change event before its happening;
3) use a global variable to interchange the text change information between
nsHTMLTextFieldAccessible and nsRootAccessible.

seeking r=.

Comment 4

[Aaron Leventhal]

Comment on attachment 102774 [details] [diff] [review]
use nsIEditActionListener to track the text change

Destructors should always be virtual.
The global gTextData should be on nsTextFieldAccessible instead of using the
global namesspace, so it becomes nsTextFieldAccessible::gTextData.

What happens if an input even occurs and none of the editor action methods  get
called. It will use the old gTextData, which is now incorrect. Should we store
a boolean like gIsTextDataNew, and set it to false after it's used? Or, should
the nsHTMLTextFieldAccessible be responsible for firing it's own events?

Sorry about r-
I promise to help you get this is soon.

Comment 5

[Kyle Yuan]

Attachment: new patch

>Destructors should always be virtual.
Done!

>The global gTextData should be on nsTextFieldAccessible instead of using the
>global namesspace, so it becomes nsTextFieldAccessible::gTextData.
I moved gTextData into nsRootAccessible, because it will be used not only by
nsHTMLTextFieldAccessible but also by my future work for <editor>
accessibility.

>What happens if an input even occurs and none of the editor action methods 
get
>called. It will use the old gTextData, which is now incorrect. Should we store

>a boolean like gIsTextDataNew, and set it to false after it's used? Or, should

>the nsHTMLTextFieldAccessible be responsible for firing it's own events?
I prefer the second way, because the currect dom input event isn't suitable for
ATK. ATK needs the "insert-text" event to be fired *after* text gets inserted,
but "delete-text" event to be fired *before* text gets removed. I promise I'll
do that in another bug that includes other work for editable text,

Comment 6

[Aaron Leventhal]

Comment on attachment 115833 [details] [diff] [review]
new patch

r=aaronl

Thanks Kyle. Check to make sure the white space changes are correct for 
accessible/src/base/Makefile.in
It looks like a tab vs. spaces problem.

Comment 7

[Kyle Yuan]

Comment on attachment 115833 [details] [diff] [review]
new patch

Thanks Aaron, I'll correct the tab/space in makefile.in when I check it in.

Comment 8

[Henry Jia]

Comment on attachment 115833 [details] [diff] [review]
new patch

Please ask someone else who is more suitable to sr this patch.

Comment 9

[Kyle Yuan]

Comment on attachment 115833 [details] [diff] [review]
new patch

After a long discussion with Henry, he already understood what I want to do in
this patch :)

Comment 10

[Kyle Yuan]

mass changing QA, platform, target milestone ...

Comment 11

[Kyle Yuan]

also fixed by bug 195687.