Wanting Types,
Demanding Mirrors

Brit Butler
2012-09-07

(use arrow keys or PgUp/PgDown to move slides)

• Are We There Yet?
• Simple Made Easy

Death by Complecting

complect -
To join by weaving or twining together; interweave.

My last talk left me thinking about the war between
Static and Dynamic languages.

Also that there have been multiple successful models for building software.

Two startling personal conclusions:

• The war is about Reflective Systems, not Types!
• More controversially, the Unix model is WRONG for language implementations.

But let's take a brief detour to look at programming today...

On 'Best Practices'

Several times this year, Github has done 100+ deploys/day.

What's a web service without continuous deployment?

Greenspunning Reflection

• Kernel Examples:
  • ksplice - Load security updates without rebooting.
  • SystemTap/DTrace - Dynamic Tracing. Took 4 years!
• Userland Examples:
  • xmonad - Great software. But plugins involved a thesis!

Other interesting papers...

• Types and Reflection
  see: related discussion
• Dynamic ML without Dynamic Types

Back to the Runtime Wars...

Pascal is for building pyramids - imposing, breathtaking, static structures built by armies pushing heavy blocks into place. Lisp is for building organisms...
- Alan Perlis

On Dreams

• The Static Dream
  • A compiler pronounces your program correct by construction.
  • Many classes of errors impossible.
  • Semantics assured provided no bugs in: Hardware, OS, Compiler
• The Dynamic Dream
  • The program is a living thing.
  • I can take it apart or observe it during execution.
  • I can decide to change it during execution.

On Runtimes and Tradeoffs

What it takes to do reflection *right*:

• Compiler/Debugger/Inspector always available.


• Implies runtime type info even for statically typed programs! No more Type Erasure!


• Implies increase in space costs.
  - Interesting analog to Immutable Data Structures here.


• Hard to sandbox highly reflective systems.


• Easy inspection a nightmare for proprietary software.

Sic Transit Gloria

• Lisp Machines!
  • Filesystem hacks
  • C interop
• Tektronix adds color...
• Serializing debuggers...
  • With Pharo...
  • And other 70s dreams...
• Isn't this what we dreamed of?

How did we get here?

A few possible causes...

• Backwards compatability/outside world


• Price! Custom hardware couldn't leverage economies of scale.


• Attempts to switch *AWAY* from textual mediums for code


• Too hard for average programmers

Today's Dynamism Kings

• The Browser - Javascript is more dynamic than Scheme!
• Emacs - The Last Lisp Machine. Duh.
• Surviving Dynlangs
  • Common Lisps
  • OpenDylan
  • Pharo Smalltalk
  • Erlang
• New Dynlangs
  • Factor
  • Io

Where to next?

• Natural Programming Languages!
  • We will need to integrate AI for this.
  • David Barbour seems to be on the right track.


• Self-explaining Computation!



• Higher-level languages on the bare metal!



• Or even just better environments.

Conclusions

• Dynamic systems died due to evolutionary pressures.



• Many of those pressures are likely gone. Let's try again!



• There are advantages to language-level support for dynamism.



• Statically and Dynamically Typed programs can benefit!

Q & A, or...

Slides Source