[r6rs-discuss] all that being said (re libraries)

From: Thomas Lord <lord>
Date: Sun, 24 Jun 2007 11:32:51 -0700

 Let's pretend that everyone agrees "no to version numbers".
Then the next question would be "what, then?"

Programmers still write *some* kind of name in
their imports and still give *some* kind of name
to their libraries. Running a program still requires
some mapping between those two classes of name.

How, in principle, should that work?

Some thoughts:

URIs are probably the best and most orderly standard
name syntax for distributed allocation of names,
hierarchical namespaces, and names that are atomic
from one perspective but can have rich internal
structure from another perspective. For those
reasons, URIs seem to me a good choice for
library names.

To really unambiguously specify a program one must
nail down which precise library revisions it will use.
No made-up name for a library will reliably do that
so it would be nice to *enable* (not require) the possibility
of libraries that are, essentially, named by their source
texts.

In particular, and following the example of XML,
Schemers should standardize (a) a canonical form
of any source text (e.g., removing whitespace
differences, comments, etc.) and (b) some standard
cryptographic-quality checksums of canonical forms.
Thus, one *alias* for a particular revision of a library
could be, for example, the SHA1 checksum of its
canonical source.

-t
Received on Sun Jun 24 2007 - 14:32:51 UTC