Re: no empty iXML prologs

Tom Hillman writes:

> Michael said earlier that he had an idea about how to formulate a
> grammar with optional, non-empty prolog(ue)s, but if we allow a choice
> of (optional) prolog types (e.g. version and namespace declarations),
> I'd be interested to see what that looks like...

> Perhaps
>        ixml: s, prolog?, rule+RS, s.
>      prolog: prologChild+RS.
> -prologChild: version; nsdecl.
> but this would then allow
> • multiple version declarations
> • version declarations not at the start of the prolog


> Alternatively:

>  ixml: s, prolog?, rule+RS, s.
>  prolog: version, RS, nsdecl*RS; (version, RS)?, nsdecl+RS.

> would be OK-ish for now, but scales horribly should there be another
> candidate member for the prolog.

I think the second refernce to 'version' is redundant; unless I'm
mistaken the same language is accepted by the slightly simpler rule

    prolog: version, RS, nsdecl*RS; nsdecl+RS.

I assume that the things that will feel like going in a prolog will all
be declarations of one kind or another.  The version declaration does
feel as if it should come first, but the other things I can think of as
candidates for a prolog are not constrained that way, so your idea of
'prologChild' would work:

    prolog: version, RS, decl*RS; decl+RS.

or making 'prolog' a little simpler by factoring out as much detail as
possible into other nonterminals:

    prolog:  version-info, declarations?; declarations.
    -version-info: version, RS.
    -declarations:  declaration+RS.
> ..., maybe allowing an empty prolog isn't terrible?

I think you're right, it's not terrible.


C. M. Sperberg-McQueen
Black Mesa Technologies LLC

Received on Tuesday, 12 April 2022 17:27:27 UTC