[Prev][Next][Index][Thread]

Re: XML catalog draft



At 02:53 PM 02/11/97 CST, Paul Grosso wrote:

>But suppose I have a local copy of the HTML 2.0 DTD (and who wouldn't),
>so my local system default catalog has:
>
>  PUBLIC "-//IETF//DTD HTML 2.0//EN" "/home/dtds/html.dtd"
>
>If you prefer SYSTEM, then there is no way to find the local copy in
>favor of the spyglass one (unless the latter fails).
>
>Suppose I want to avoid the unnecessary http download.  Suppose it's
>are very big file (not just the HTML 2.0 DTD).  Suppose a few years
>from now people start charging for accessing files.  Suppose I retrieved
...
>Public ids aren't just a fallback for broken URLs, they are a very
>important way to manage indirection of resource name resolution.

I think Paul's exactly right here. PUBLIC identifiers are supposed to
abstract away from *one* particular user's environment; that is precisely
what allows the two crucial things:

   * indirected resolution that works for everybody

   * local override that may happen to be different for the overrider.
     this may occur because they create a slightly different local version
     (say, adding archform attributes for ICADD or whatever), or because you
     sent it to an Orwellian country where MiniTrue must rewrite it, or, most
     likely of all, simply becaue their directory has a different name than
     the one on the original author's disk.

Those are only a few of the reasons I think PUBLIC should be tried first.
But I could live with a system that let you choose (that's where we ended up
in SGML Open, after all); that does have the downside that two conforming
implementation  that are installed in an identical environment and given
identical catalogs and other data, can conformingly produce quite different
results. That seems bad to me.