Bidirectional Programming Nate Foster Cornell University CS 2110 - PowerPoint PPT Presentation

Bidirectional Programming Nate Foster Cornell University CS 2110 24 November 2015 Languages                                                                                                  

Most programming languages, like C...

Java,

Python,

and C++ are general purpose.

I’m interested in designing languages that are specifjcally designed for particular tasks

Domain-specifjc languages 3 • Clean semantics • Natural syntax • Better tools

The View Update Problem In databases, this is known as the view update problem. [Bancilhon, Spryatos ’81] 6 View Database Query A B C x 1 true y 2 true z 3 false A B C x 1 false y 100 true Trigger

The View Update Problem In Practice It also arises in data converters and synchronizers... [Foster, Greenwald, Pierce, Schmitt JCSS ’07] — Harmony 6 Common target Replica in format A format Replica in format B Synchronized replica in Synchronized replica in format B format A

The View Update Problem In Practice ...in picklers and unpicklers... [Fisher, Gruber ’05] — PADS 6 In-memory representation Binary file application update Updated binary file

The View Update Problem In Practice ...in model-driven software development... [Stevens ’07] — bidirectional model transformations 6 Java code UML model Point x : int y : int translate(int x,int y) refactor Point x : int y : int translate(int x, int y) moveTo(int x, int y) Updated Java code Refactored model

Problem How do we write these bidirectional transformations? 7

Problem: Why is it hard? We want updates to the view to be translated “exactly”... 7

Problem: Why is it hard? We want updates to the view to be translated “exactly”... 7

Problem: Why is it hard? ...but some updates have many corresponding source up- dates... 7 ?

Problem: Why is it hard? ...while others have none ! 7 ?

Possible Approaches Bad: write the two transformations as separate functions. 8 • tedious to program • diffjcult to get right • a nightmare to maintain

Possible Approaches Good: derive both transformations from the same pro- gram. design 8 • Clean semantics: behavioral laws guide language • Natural syntax: parsimonious and compositional • Better tools: type system guarantees well-behavedness

“Bidirectional languages are an effective and elegant means of describing updatable views”

[Foster, Greenwald, Moore, Pierce, Schmitt TOPLAS ’07] Lenses ’‘Never look back unless you are planning to go that way” —H D Thoreau

Terminology 11 get

Terminology 11 put

Terminology 11 lens

Bidirectional vs. Bijective If get is non-injective, put needs access to the original source. Of course, the purely bijective case is also interesting. 12 non-injective function

The Bijective Case For bijective transformations... ...the desired behavior is obvious. 13 T S

The General Case But for bidirectional transformations... ... we need to identify conditions that allow us to 14 T S • recognize and reject bad (unreasonable) programs • understand and predict behavior

An Unreasonable Example 15 project out string component foo 0 foo

An Unreasonable Example 15 foo 0 foo bar

15 An Unreasonable Example foo 0 foo blech 5 bar return a constant

An Unreasonable Example 15 foo 0 foo blech 5 bar ≠ blech

The PutGet law Principle: Updates should be “translated exactly” — i.e., to a target. Formally: v 16 source for which get yields exactly the updated get ( put v s ) =

17 A Debatable Example project out and duplicate string component foo 0 foo foo

A Debatable Example 17 foo 0 foo foo bar foo

A Debatable Example 17 foo 0 foo foo bar 0 bar foo propagate "newest" string

A Debatable Example 17 foo 0 foo foo bar 0 bar foo ≠ bar bar

Another Unreasonable Example 18 project out string component foo 5 foo

Another Unreasonable Example 18 foo 5 foo bar

18 Another Unreasonable Example foo 5 foo bar 0 bar propagate always set numeric updated string field to 0

18 Another Unreasonable Example foo 5 foo = foo

18 Another Unreasonable Example foo 5 foo = ≠ foo 0 foo

The GetPut law Principle: If the view does not change, neither should the source. Formally: s 19 put ( get s ) s =

Another Debatable Example 20 project out string component foo 0 foo

Another Debatable Example 20 foo 0 foo bar

20 Another Debatable Example foo 0 foo bar 1 bar increment numeric component if string component has changed