Re: Any example of AtkDocument implementation?

[Bill Haneman <Bill Haneman Sun COM>]

Ashutosh asked about AtkText, and whether it would be better to
break a large document into per-paragraph AtkObjects (each of
which implements AtkText).

This is what we recommend (per-paragraph AtkObjects,
with AtkText support).  If these text objects are individually focussable,
then gnopernicus should "do the right thing" when focus changes.

I am actually dubious about whether obtaining the entire text content
of an object each time it's focussed is a good idea; that's a question for
the gnopernicus team.  But you are quite right in saying that reducing the
size of the text objects will help in this regard.

I answer in a little more detail inline:

> I'm trying to make an OCR application. I'm
> implementing AtkText for the same now. 
> I'm confused about a new point now. If the page has a
> lot of text, the get_text implementation is always
> receiving a request with start_offset = 0 and
> end_offset = -1 (from gnopernicus). In that case, I am
> returning the entire page's contents.
> So, every time the focus moves away and comes back to
> the page, the entire contents are sent back.
>
>  
This seems like a very inefficient operation, I agree.

> Currently, the entire page has one associated
> AtkObject. If I create an AtkObject for every
> paragraph/line/word and implement get_n_children and
> ref_child (setting their pointers in the class_init
> function), can the screen reader request for the text
> of one child at a time?
>  
Yes, the screen reader can and should do this.

> I mean, can I return only the appropriate AtkObject's
> text, based on where the user has clicked. If the user
> clicks inside a particular paragraph, only that
> AtkObject's text is returned. How can this be done? Do
> I need to implement AtkComponent?
>
>  
Every object on the screen should implement AtkComponent.
But I don't think in this case that AtkComponent will be heavily
used by gnopernicus.  For blind users, the access to various
paragraphs will be via keyboard navigation in almost all cases.

> Otherwise, what is the advantage of creating an
> AtkObject for each paragraph/line/word as against
> treating the entire page as one object?
>  
I think you've answered your own question :-)
Seriously, this becomes an even more significant issue for
scrolling text, which could become way too cumbersome without some
sort of subdivision of the document.

best regards,

Bill

> Regards,
> Ashutosh
>
>