Re: [PATCH 1 of 4] xl: idl: Abolish keyed union types

[Ian Campbell <Ian.Campbell@eu.citrix.com>]

On Thu, 2011-01-06 at 17:56 +0000, Ian Jackson wrote: 
> Gianni Tedesco writes ("[Xen-devel] [PATCH 1 of 4] xl: idl: Abolish keyed union types"):
> > xl: idl: Abolish keyed union types
> > 
> > Since the IDL file has become useful for generating language
> > bindings it has become apparent that the KeyedUnion type has no
> > straightforward translation to scripting languages which have no
> > notion of unions.
> 
> Uhh?  Most scripting languages' aggregate types are practically
> nothing _but_ unions!
> 
> The correct translation of a keyed union would probably be an
> aggregate (eg, in Python, a dictionary) containing an entry for the
> key and entries for whatever version of the union it involved.

The union type ends up looking quite nice in the ocaml bindings too:

        type build_spec = BuildHVM of build_hvm_info | BuildPV of build_pv_info

which allows nice ocaml idiomatic stuff like
        match build with
        | BuildHVM hvm -> do_some_hvm_stuff hvm
        | BuildPV pv -> do_some_pv_stuff pv

> > Turns out this is only used in domain_build_info which is hardly a memory
> > criticial structure.
> 
> However I don't have a strong opinion about the desirability of using
> a tagged union for this particular structure.

Me neither.

That ocaml stuff currently hand coded (and no one is imminently working
on changing that to my knowledge) so I don't think the libxl API change
arising from this IDL change prevents retaining this interface but it
does suggest why having the notion of a key which selects between
alternative fields could be a semantically useful concept in the
interface.

If having KeyedStruct instead of KeyedUnion simplifies things that could
be an alternative. Or go the whole hog and define a Dict type (turns
into an enum + the relevant number of fields)? Or just nuke it and cross
the bridge when we come to it ;-).

Ian.