[Accessibility] Goal of AT-SPI inclusion of AT-SPI

[Olaf Schmidt]

[ Olaf Schmidt, Mi., 8. Nov. 2006 20:14 ]
> These are some thoughts on possible ways to include AT-SPI in LSB, and
> their problems and benefits

I forgot to include some important questions about the exact purpose of 
including AT-SPI in the LSB.

In Hawaii we agreed that AT-SPI should not be an ABI standard, but an IDL 
standard. People are free to write additional conformance tests (e.g. for 
D-Bus) that can be published later.

The LSB is an ABI standard, i.e. the ABI is normative (not an IDL).

If we want the LSB to include AT-SPI, then how are the normative IDL of AT-SPI 
and the normative ABI of the LSB related?

There are three ways to do this:

1. The FSG publishes both the AT-SPI IDL spec and the conformance tests 
separately from the LSB. The LSB references the AT-SPI spec by
a) requiring or
b) recommending
distributions to include a framework that implements AT-SPI (without defining 
an exact ABI).

2. The FSG publishes the AT-SPI IDL spec separately from the LSB. 
Additionally, an ABI derived from the IDL is made normative for the LSB in a 
module that is
a) mandatory or
b) optional.
Both standards share the same conformance test.

3. We convince the FSG to change the nature of the LSB and to allow parts that 
do not have a normative ABI.


ad 1.: This would allow us to have our own time schedules for new versions of 
the AT-SPI IDL and of conformance tests, giving us most flexibility regarding 
the D-Bus move. We could publish a conformance test for D-Bus immediately 
when it is ready (and maybe even before the implementation is passing it). 

ad 2.: This approach makes perfect sense once we have an ABI that everyone is 
happy with, but to go this route now would be a substantial change from the 
decision made in Hawaii. It is difficult to exactly evaluate the implications 
of a normative CORBA ABI for a D-Bus move, especially since the method for 
depreciating obsolete ABIs is still undecided in the LSB.

ad 2.a): It would be difficult to explain how our standardised ABI is not 
really intended to be normative in all cases, and what the standard exactly 
means.

ad 2.b): Publishing an "optional" module for AT-SPI would sound politically 
much more negative than recommending that LSB conformant distributions also 
to implement a separate standard.

ad 3.: It doesn't seem to make much sense to change the nature of the LSB, and 
I do not think that we could achieve it if we tried.


Olaf