Programming systems

Not just for data science

Tomas Petricek, University of Kent
tomas@tomasp.net | @tomaspetricek | http://tomasp.net

Making programming better

Caricatures of conventional views


What is programming?
Producing code in a programming language!

How to improve it?
Write a research paper with a prototype!

What matters about it?
Safety, efficiency, usability!


Technical dimensions

(with Joel Jakubovic and Jonathan Edwards)

Programming Languages

Programming is writing code

Formal semantics, implementation, paradigms

We know how to study this!

Programming Systems

Interacting with stateful system

Visualizations, feedback, interaction

But how do we study this?

Language is
a small part

Together with editors, debuggers & other tools

Less interesting but overemphasized?

Technical dimensions

Science of programming systems


Academics & start-ups build interesting systems!

Disconnected and hard to compare

Can we find a common language?

Stand on the shoulders of giants?



Dimensions

Identify interesting characteristics

Cover "good old" and new systems alike

Allow qualitative comparison

Don't say what's good!

Interaction

  • Feedback Loops
  • Modes of interaction
  • Abstraction Construction

Notation

  • Notational Structure
  • Surface/Internal
  • Primary/Secondary
  • Expression Geography
  • Uniformity

Error Handling

  • Error Detection
  • Error Response

Conceptual Structure:

  • Integrity/Openness
  • Composability
  • Convenience
  • Commonality

Customizability

  • Staging
  • Externalizability
  • Additive Authoring
  • Self-Sustainability

(Others)

  • Degrees of Automation
  • Learnability & Sociability

Abstraction Construction

From Concrete

  • Generalize from examples
  • Expanding range in Excel
  • Pygmalion system

From Abstract

  • Define function first
  • Most programming languages
  • Coding done without values

Notational Uniformity

Post-modernist

  • Variety of different notations
  • More to learn, but better problem fit
  • Perl language, Web platform

Modernist

  • Small set of uniform primitives
  • Not everything fits the notation
  • Lisp and (partly) Smalltalk

Self-sustainability

Separate language level

  • Implementation vs. user level
  • Limited changeability from within
  • Java and other languages

Integrated systems design

  • Implemented & modifiable in itself
  • Often changeable at runtime
  • Smalltalk, Lisp Machines

Technical dimensions

Why and how to use them


Systems are more than languages

Trying to find a common language

New look at past programming systems?

Map for exploring new design ideas?



Data exploration

Querying data

Athletes by number of gold medals from Rio 2016

1: 
2: 
3: 
4: 
5: 
6: 

olympics = pd.read_csv("olympics.csv")
olympics[olympics["Games"] == "Rio (2016)"]
  .groupby("Athlete")
  .agg({"Gold": sum})
  .sort_values(by="Gold", ascending=False)
  .head(8)
Unknown file
Column name

Querying data

Language and data source features you need to know


Python dictionaries {"key": value}

Generalised indexers .[ condition ]

Operation names sort_values

Data column names "Athlete"



Notebooks

Financial Times waste analysis

Note the clever abstraction mechanism!

Notebooks

Financial Times waste analysis

Note the clever abstraction mechanism!

The Gamma

Programming through iterative prompting


Program just by using auto-complete!

Recognition over recall principle

Make correct and complete suggestions



Demo

Iterative prompting in The Gamma

The Gamma

Technical dimensions analysis


Uniformity of notation - everything is a member

Self-sustainability - all libraries external!

Conceptual structure - convenient at the top-level

Error detection and prevention using types



Histogram

Programs as sequences of interactions


Creating code equal to using user interface

Sequence of actions rather than expressions

Extract function as a first-class feature

Integrate programming and execution



Demo

Data exploration in Histogram

Histogram

Technical dimensions analysis


Notational structure - explicit structure editing

Abstraction construction - from concrete example

Modes of interaction - editing & running in one mode

Self-sustainability - libraries remain external



Complementary science

Complementary science

Doing history to contribute to contemporary science

Are there lost ideas worth recovering?

Works even better in programming!

Demo

Recovering Commodore 64 BASIC

Commodore 64

Technical dimensions analysis


Modes of interaction - editing & running in one mode

Learnability - use BASIC even to play games!

Sociability - code shareable in magazines

Conceptual convenience - high with escape hatches



Conclusions

Programming systems

Technical dimensions to think about systems

Iterative prompting for non-expert data exploration

Programming as interaction to design new systems!

Complementary science to get more ideas!



Tomas Petricek, University of Kent
tomas@tomasp.net | @tomaspetricek | http://tomasp.net