On 01/02/2010 21:43, Frank Mittelbach wrote:
>   >  Thoughts? At present, (3) is all that is on offer, but perhaps that is
>   >  all we need (do designers care about muskips at all?).
>
> your last question perhaps depends on whether or not there is going to be a
> lot of "templates" for math objects which I somehow doubt, mainly because most
> such objects will not need a "designer interface" supporting a number of
> possible layouts for some object. But I might be wrong.

I'd agree, at least on the current (scant) evidence.

> (1) is my favorite as it is clean and fits the rest of the types.
>
> As you say to implement that it would probably need an underlying support for
> muskips as code level types, but you can start by simply having this type
> accepting a tokenlist and you deal with it internally for the moment. That
> could then later be changed to use whatever the kernel then offers as type
> support.

My main motivation in sending the mail was to make people aware of the 
question. I'm happy at the moment with the current solution (use 
tokenlist and sort it out in the code). This is very much the same as 
other specialist input (most obviously float-point numbers, but anything 
which has some "other" structure is handled as a tokenlist at the 
template level).
-- 
Joseph Wright