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

From: Thomas Lord <lord>
Date: Wed, 20 Jun 2007 18:00:38 -0700

AndrevanTonder wrote:
> I'm not sure I understand the issue you mention. With my simple suggestion, a
> client import of 6.2 will match the bug-fixed 6.2.1, and import the latter, as
> desired.
>


That concept is messed up.

It assumes that the question "should 6.2.1 be used instead
of 6.2?" has the same answer for all applications that
happened to allow both version numbers in their imports.

The problem is that the question can easily have
multiple answers. Perhaps 6.2.1 fixes a bug which is
critical to, say, the web server -- but the same bug
fix accidentally breaks the printer driver, for which
the bug fix wasn't even needed in the first place. What
then?

The decision "is library version X suitable and preferable
for the import in library Y?" should be a late bound
decision. That is, it should not be a static property of
the source texts of X and Y but, instead, should be
determined outside of X and Y, at run-time -- essentially
by a separate program.

Version numbers aren't needed at all, except for human
consumption (hints, nothing more). What is needed
is a way to configure linkages separately from programs.
Whatever that configuration mechanism should be, its
design should make it practical to manage the linkage
of many programs at once with just a few changes, and
also make it practical to manage the linkage of any
one program, in detail, regardless of what's happening to
other programs.


Installing a new version of a library should never change
the meaning of existing programs. Modifying the
linkage specification of a program or programs should
change their meaning. No need for version numbers at all.

-t
Received on Wed Jun 20 2007 - 21:00:38 UTC