Re: [Gnu-arch-users] Features command for arch

[David Allouche]

Tom Lord wrote:
> David is slying suggesting that for every feature mapping F and every
> integer n we can define a mapping F_n whee
> 
>       F_n(name) := (n <= F(name))
> 
> To say that is to make an assertion about the natural "topology" of
> feature assertions.

Sorry I do not understand what you mean. And I do not think I have been
making any assertions about some "natural" properties.

"Feature versions" are just a convenient way to keep the name space
manageable. No metaphysics here.

> David: why this particular topology and not one with richer structure?
> (or poorer structure, as Jblack suggested?)

For no particular reason except that Robert proposed doing it that way
and that using name->integer mapping instead of a name->boolean mapping
     1. does not add any hair to the proposal,
     2. provide a simple way to prevent excessive growth of the
        namespace,
     3. can be ignored if deemed useless,
     4. can be used for other purposes if deemed useful.

Pointing out that this particular semantic of the name->integer mapping
allowed dual name->boolean mapping was just a way of saying "but if you
prefer, we can use boolean mappings, that can solve my problem just as
well".

>    Is there a "natural"
> topology that might be the trump-card of topologies for feature
> assertions?

Doing deep thinking in this area is your mandate, and I declare myself
not qualified to answer such questions.

I am just trying to solve annoying compatibility issues for wrappers.

>   If there is no "natural" topology then this is a design
> opportunity to accomidate conflicting constraints by making an
> otherwise arbitrary decision in a context-specific way --- and so in
> that case (of no "natural" topology): what is our design space?

Beg your pardon?

> (Likewise, re the math, we have a slightly foggy language barrier
> between us, afaict.)

Apparently.

-- 
                                                            -- ddaa