[r6rs-discuss] [Formal] Version reference syntax is overly complex.

From: Thomas Lord <lord>
Date: Thu, 21 Jun 2007 08:22:22 -0700

William D Clinger wrote:
> Tom Lord is partly right about this. Most of
> the versioning mess should be the responsibility
> of the same layer that maps library names to
> files and other real-world objects. The editors
> have declared that layer to be outside the scope
> of the normative part of the report, so versions
> should probably be outside of the report's scope
> also. Versions don't belong in ordinary source
> files; get rid of them.
>
>
>

I'm glad that you reach a similar conclusion but you
misrepresent my view when you attribute this to me:
"Most of the versioning mess should be the responsibility
of the same layer that maps library names to files
and other real-world objects."

I don't happen to see it *quite* that way.

Taking lessons from successful design patterns in the
W3C world, and noting that rising importance of
open source Scheme libraries that are widely available,
I would:

1. Name libraries using URIs designed to be globally
    unique at least among members of any minimally
    cooperative community. These URI names are
    a property of the library source and have no necessary
    relationship to where the library is stored as
    a "real world" file.

2. Define the abstract execution environment of programs
    to include a set of distinct, importable modules, no
    two of which have the same (URI) name.

3. Allow portable programs ("linkage specifications") to
    specify the mapping from short-hand names in imports
    to URI library names.

In any actual implementation there is then a need for
an additional layer of translation -- from URIs to the
appropriate source files, object files, or what have you.
That additional mapping onto real files is what you are
talking about, but not what I'm talking about.

Another trick from the W3C world that could be helpful
here is to define a canonicalization procedure for portable
Scheme source. Ideally that procedure would yield the
same results if two source texts differ only things like
insignificant whitespace. Given such source canonicalizations,
we can compute crypto-quality checksums of canonicalized
sources. So, the set of libraries available in an environment
is named by a list of URIs, yes, but additionally, a set of
URI + checksum pairs will (probabilistically) uniquely name
the library *content* of the environment. We trust
people to generate their own URIs because we don't want to
impose a centralized naming authority on library names --
but we can use canonical checksums to "trust but verify".

-t
Received on Thu Jun 21 2007 - 11:22:22 UTC