[TOS] Great Piece from Brett Victor on Learnable Programming

[Mark Hoemmen]

On Thu, Sep 27, 2012 at 9:31 PM, Stephen Jacobs <sxjics at rit.edu> wrote:
> Originally brought to my attention via the Sugar Labs/OLPC IAEP mailing list
>
> http://worrydream.com/LearnableProgramming/

Argh, you beat me to it ;-P  It was a great article.

i was a bit bothered by the "copy and paste" code composability
example.  Maximal composability is good, but sometimes you really do
have state or context.   "fill" is a bad example because it's a silly
interface.  Languages have different ways of dealing with state, such
as making it temporary within a scope or "undo-able" (via dynamic
binding).  Getting rid of it altogether (by passing it in as a
parameter) can help, but sometimes it increases cognitive load
(because now there's this extra argument that i have to pass into the
next N function calls).

A thing that i like about visualizing program output is that it ties
back to the early days of computing, or even before.  People like
Gauss (a human "computer" as well as a mathematician, which was
unusual in those days) and Wilkinson (worked with Turing; a father of
numerical linear algebra) always used to watch the intermediate output
of an unfolding computation.  They had a sense for when it was going
wrong, and for the algorithmic choices that would more easily tolerate
"temporary wrong-ness" due to rounding error or even human error
(Gauss favored "self-correcting" iterations).  i work on algorithms
that can tolerate data corruption or incorrect arithmetic; it would
have helped me a lot had i had a stronger sense for "right-looking"
intermediate output of algorithms.

mfh