[racket-dev] Package compatibility

On Sun, Jul 7, 2013 at 11:44 AM, Carl Eastlund <cce at ccs.neu.edu> wrote:
> It seems to me our new package system runs into trouble if two people write
> packages that use the same connection name.  Let's say we have two packages,
> alice-tree and bob-tree, both of which provide the collection
> data/red-black-tree, but both provide different interfaces (perhaps similar
> functionality with different naming conventions).  It seems to me this is
> not that unlikely to happen, but once it does, those two packages are
> completely incompatible.  They can never meaningfully be installed on the
> same system, and if they are, the behavior of any code depending on either
> becomes undefined.
>
> Have I got this right, or have I missed some detail?

You are correct.

> If this is right, how
> concerned are we about this feature?  The fact that packages are all spliced
> into a single flat collection namespace of course leads to the possibility
> of clashes like this, which can cause not just incompatibility between
> alice-tree and bob-tree, but a complete schism wherein all code depending on
> alice-tree at any level of indirection becomes completely disjoint from all
> code depending on bob-tree at any level of indirection.  I've seen this kind
> of library incompatibility before, in the ACL2 "books", and I wrote a
> dissertation to try to solve it.  I hope we're not repeating their mistake,
> but it seems like we have yet another flat namespace with no resolution
> mechanism.

Planet attempts to solve this problem technically by (a) having all
collections be prefixed by <author>/<package-name> and (b) mandating a
centralized server that enforces unique <author>s and unique
<package-name>s per author. Since Racket packages don't have a
mandated central server we can't enforce uniqueness like that. (i.e.
even if we mandated the prefix by <package-name>, there's no way to
enforce unique package names across the universe without a central
server.) Furthermore, bringing <package-name>s in to the code would
give us /internal linking/ which Racket packages are designed to avoid
for other reasons.

Since technical solutions to this problem are lacking, I/we/PLT
(depending on the level of agreement with this idea) are trying to
solve this problem socially by rewarding package authors (who list
their packages on the PLT run server) with the ring system. Ring 1
packages can't conflict with each other and my desire is for only ring
1 and 0 packages to be listed in the DrRacket GUI. Ring 0 packages
can't conflict with each other and must be "good" (have documentation,
tests, etc). The goodie they get is that their packages will be tested
with DrDr and we'll build their documentation and include it on the
main site.

Jay

> Carl Eastlund
>
> _________________________
>   Racket Developers list:
>   http://lists.racket-lang.org/dev
>



--
Jay McCarthy <jay at cs.byu.edu>
Assistant Professor / Brigham Young University
http://faculty.cs.byu.edu/~jay

"The glory of God is Intelligence" - D&C 93