[BRLTTY] Broken colour names

[Mario Lang]

Samuel Thibault <samuel.thibault at ens-lyon.org> writes:

> Shérab, on Tue 17 Nov 2015 15:04:54 +0100, wrote:
>> Samuel Thibault (2015/11/17 14:06 +0100):
>> > The problem is that these userland-driven terminals don't provide the
>> > screen output. Ideally we'd manage to gather the people implementing
>> > them, and get to define a common interface to access the screen
>> > output.
>> 
>> Isn't AT-SPI good enough for that? Or is it too heavy for instance?
>
> It is heavy for performance (at *best* thousands of requests
> per second), and it is really not suited to getting terminal
> information. The AccessibleTerminal interface has never been actually
> implemented.
>
> Generally speaking, the way AT-SPI works seems overkill to me here. The
> way it works (insertion/removal notification) really does not suit
> terminal screen reading.
>
> A shared-memory solution looking like /dev/vcsa, but rethought for
> unicode etc. would work much better.

This is a very old open problem.  I remember we agreed long ago on doing
it similar to how Screen already does it (via the patch), but in a
"standardized" way, such that the *same* sort of SHM layout could be
used for several different userland terminal emulators.
It seemed simple enough, but for some reason never got done.

I agree that AT-SPI seems like overkill.
However, I wonder if we should at least consider D-Bus as an option for
doing IPC.
When we originally planned to do it via SHM, D-Bus was not around.
Is SHM still the best option?  I am guessing yes, especially because
we want the implementation on the side of the terminal emulator to be as
simple as possible, to raise the probability of mainstream acceptance.
However, D-Bus has access control, and maybe we could make use of that
to minimize potential security risks obviously exposed by such a patch.

Iif we find that we want to use D-Bus, we should consider reviving the
AccessibleTerminal interface, because AT-SPI, after all, is just sitting
on top of D-Bus.  The libraries are a convenience, but it is perfectly
possible to code a standalone AT-SPI client just by using a D-Bus
binding.

This is not a plan, just a question.

-- 
CYa,
  ⡍⠁⠗⠊⠕