Re: [g-a-devel] AccessibleTerminal Interface ?

[Samuel Thibault <samuel thibault ens-lyon org>]

Hi,

Bill Haneman, le Mon 06 Jun 2005 21:29:47 +0100, a dit :
> Your comment is not a solution either.

My comment is not indeed, but my previous proposal seems a reasonable
one to my mind.

> This is NOT slow, at least until someone provides concrete benchmarks 
> showing why more performance is necessary.

I've just tested: with my 800MHz pentium III and the "read line by line"
algorithm, the loop takes 3-15ms when getting the bottom of the screen
(most usual case at the prompt). While with a line-oriented protocol
(thus avoiding round-trips), it usually takes 6µs...

For instance, the X11 team carefully designed their protocol to avoid
round trips as much as possible. I'd say at-spi should as well. Why
not using a design that both makes code much less tricky in both BrlTTY
and libvte, and is much more efficient ?

> It already works transparently over networks.

I don't doubt that :) But networks typically add some ms more, which
makes things yet worse. While without round-trips, the network latency
is only added once.

> No matter what we do, terminals will continue to have to provide the 
> AccessibleText API (even if they also provide another supplementary 
> API), since AccessibleText should be implemented by anything that 
> renders text onto the screen. 

This means hairy code for terminals programs. I don't think
xterm/aterm/eterm/... authors will write such hairy code.

And terminals _don't_ render text but rather lines of text. That is
really obvious when reading xterm/aterm/eterm/...'s code.  If you
prefer, that could be an 1-dimension AccessibleTable of AccessibleTexts,
but defining a neat AccessibleTerminal would be much preferable I guess.

Regards,
Samuel