As promised earlier, here
are some more notes from this year's
European Lisp Symposium in Milan - this time about Scott McKay's keynote presentation on what he has learned in his years as a Lisp and Dylan programmer at companies like Symbolics, Harlequin, HotDispatch and
ITA Software.
Scott joked he might well be the
Zelig or
Forrest Gump of the Lisp community, after having been around for a long time and
making appearances in a number of (unlikely?) places. In amusing anecdotes,
he explained some of the key learnings he took away during his career, and what
those learnings might mean for the future of Lisp and the Lisp community.
Some notes (from memory, hence most certainly inaccurate):
- "Any bozo can write code" - this is how David Moon dismissed Scott's attempt
to back up one of his designs with code which demonstrated how the design
was meant to work.
- "Total rewrites are good" - Scott was the designer
of CLIM, which underwent several
major revisions until it finally arrived at a state he was reasonably happy with.
- "If you cannot describe something in a spec, that's a clue!" - amen, brother!
- "The Lisp community has a bad habit of building everything themselves"
- "Immutability is good" (even for single-threaded code)
- "Ruby on Rails gets it right"; only half-jokingly, he challenged the community to
build something like "Lisp on Rails". Later during the symposium, I learned that
we already have Lisp on Lines
("LoL" - I'm not kidding you here
).
- "Java + Eclipse + Maven + XXX + ... is insane!" - and later "J2EE got it spectacularly wrong"
- He reminded us that the Lisp Machine actually had C and Fortran compilers,
and that it was no small feat making sure that compiled C and Fortran programs
ran on the system without corrupting everybody else's memory. (I'd be curious
to learn more about this.)
- Lisp code which was developed during Scott's time at HotDispatch
was later converted to Java - they ended up in roughly 10x the code size.
- The QRes system at ITA has 650000 lines of code, and approx. 50 hackers
are working on it. Among other things, they have an ORM layer, transactions,
and a persistence framework which is "a lot less complicated than Hibernate".
- Both PLOT and
Alloy were mentioned as sources of inspiration.
Scott then went on to develop a list of features which a future version of Lisp, dubbed
Uncommon Lisp, should have. That list was pretty long; notable features which I remember were:
- Should run on a virtual machine
- Good FFI support very important
- Support for immutability
- Concurrency and parallelism support
- Optional type checking, statically typed interfaces (
define-strict-function)
- "Code as data" not negotiable
Not surprisingly,
Clojure was mentioned quite often,
both during the keynote and in the subsequent Q&A session. I'm still
not quite sure what Scott's position on Clojure really is. To me, most of the points
he made seemed to actually back up design decisions in Clojure: For instance,
Clojure runs on a VM, reuses the libraries and tools of the underlying platform,
connects easily to other languages, makes a decided effort
to support concurrency well, and while it breaks with Common Lisp syntax, it is still
in the Common Lisp spirit of being a pragmatic implementation of the fundamental Lisp
ideas. On the other hand, Scott also criticised some Clojure details (name resolution
comes to mind), and seemed uncertain whether to full-heartedly recommend
everyone to join the Clojure camp right away.
I
think what Scott tried to get across is that a revolutionary approach
is both possible and worthwhile for the Lisp community. Revolution,
of course, means breaking with the past, and it became obvious during Friday's panel
discussion on the future of Common Lisp that not everybody at the symposium felt
comfortable with the thought.
PS:
Michele Simionato
discusses the keynote presentation from a Schemer's point of view.
Revision: r1.5 - 06 Jul 2009 - 11:03 - ClausBrod