Research in Compilers and How it Relates to Software Engineering Part III: Relation to SE Tomofumi Yuki EJCP 2015 June 22, Nancy.

Research in Compilers and How it Relates to Software EngineeringPart III: Relation to SE

Tomofumi Yuki

EJCP 2015June 22, Nancy

What is Software Engineering? Can you describe SE? Engineering applied to software

What is engineering?

Goals of SE easy to understand easy to modify easy to debug with little cost in performance

2

EJCP 2015, June 22, Nancy

The Problem

Building software pyramids


3

If software developers built the pyramids …


4

Software Engineering

Programmers

Software Engineers

SE Researcher

s

no gotoclass hierarchygenericsdesign patterns

development cyclesUML specificationsgenerative methods

Application(Local)

Application(Global)

Develop/InventTechniques

structured prog.software coveragefault injectionproduct-linesfind cause of failure

5

compilertechniques


5

Quick and Dirty View

Compiler is all about abstraction Software Engineering is all about abstraction

Compiler = Software Engineering it is a stretch (of course)

Programming Languages probably a better intersection how to better write in a language how to design a language to ease

programming6


Example 1: Object-Orientation Initial motivation was modularity

More SE developments later on design patterns polymorphism

7

op1 a, b, cop2 a, d, eop3 f, c, e...

main() { ... foo(..); ... bar(..); ...}

class A { ... foo() ...}

class B {...}


Example 2: Deadlock Detection Many approaches

check if a problem can deadlock ensure its absence by language design ensure its absence by coding principles catch and resolve at run-time

Is it SE problem or PL problem? papers appear in both domains

8


Today’s Focus: MDE

Specific instance where compiler and SE meet

Model Driven Engineering

Key Connection

9


Model Instance = Program Instance Model = Grammar

Many Tools and Problems are Shared

MDE: A Solution

Goal of MDE: reducing accidental complexities bridging wide abstraction gaps

Key Element: support rigorous transformation of

abstractions to software implementations

MDE is concerned with developing software tools to support the work of software

developers

10

MDE helps make a software developer a software engineerEJCP 2015, June 22, Nancy

What is a Model?

A description of an aspect of a software-based system that may or may not exist. A model is created to serve one or more purposes.

11


Early focus of MDE

Models for documentation/communication Foundation: early work on CASE

Models for generating implementations OMG’s work on UML and MDA

Model Transformations Software development viewed as a

transformative process Key enablers of MDA

12


MDE Example 1

Discuss the desired software with model stay high-level easy to visualize model

Transform to another model fill some details check for properties

Generate code (skeleton) implement the details


13

Model

Model’

Code

Evolution of MDE

Breathing life into models Models as analyzable artifacts Foundation from formal specification

Models for managing runtime behavior [email protected]

Supporting domain-specific software development

Applying MDE technologies in other CS domains

14


MDE Example 2

There exists a software corresponding model

Develop another one similar but different

How to maximize reuse productivity reliability ...


15

Model ATool A

Model BTool B

Tools for MDE

Mapping concrete syntax parser generation graphical language

Constraint checking check if rules of the model is satisfied more advanced check (e.g., cycles)

Code generation from model


16

Optimizing Compilers

Goal: generate “efficient” code Execution time Energy consumption Code size

Wide range of optimizations Register allocation Dead code elimination Automatic parallelization Run-time optimizations

17


Optimizing Compiler Research Prototype implementations

“Proof of Concept” Evaluation

Compilers are complicated pieces of software Need for rapid development Development spans generations of

students Performance of compiler prototype not

critical18


Optimizing Compiler Examples High-level flow of two research

compilers:

19

1.Parse source language

2.Transform intermediate representations (IRs) for efficiency. May take domain specific knowledge as additional inputs.

3.Output code or binary


Research Compiler Challenges Maintainable and Sustainable Code

Developers may not have good SE background

Structural Validity of IR Is the IR consistent after

parsing/transformation? Complex Querying of IR

Find where to apply transformations Interfacing with External Tools

Avoid as much re-implementation as possible

20


More Challenges

Semantics Preserving Transformations Compilers should not alter semantics

Use of Domain Specific Knowledge How to incorporate domain specific

knowledge Code Generation

Generate code for different target architecture programming model

21


Bridging with MDE

View compiler IRs as models

2222

DSLs and

Tooling

Model Transformations and

Analyses

Code Generation


Challenges

Analyses and Manipulation of IRs

23

Structural ValidationStructural ValidationStructural ValidationStructural ValidationStructural ValidationStructural ValidationStructural ValidationStructural Validation

Complex Querying and Transformati

ons


ons


ons


ons


ons


ons


ons


ons

Complex Querying and Transformatio

ns

Complex Querying and Transformatio

ns


Challenges and MDE Solutions Analyses and Manipulation of IRs

24

Structural ValidationStructural ValidationStructural ValidationStructural ValidationStructural ValidationStructural ValidationStructural ValidationStructural Validation


ons


ons


ons


ons


ons


ons


ons


ons


ons


ons

Structural Properties on models

(conformity)OCL constraints

Structural Properties on models

(conformity)OCL constraints

OCL queriesM2M tools

Rewriting rules

OCL queriesM2M tools

Rewriting rules


Challenges

Domain specific knowledge is heavily utilized

25

Domain Specific Knowledge

Representation


Challenges and MDE Solutions Domain specific knowledge is heavily

utilized

26

Domain Specific Knowledge

Representation

MDE-based DSLsGenerative approaches

(editor, parser)


Challenges

Code generation and external tools

27

Domain Specific Knowledge as additional

inputs


inputsCode GenerationCode Generation

Use of External Tools(term rewriting, ML, LP,

CSP, …)

Code GenerationCode Generation


CSP, …)


CSP, …)


Challenges and MDE Solutions Code generation and external tools

28


inputs


inputsCode GenerationCode Generation


CSP, …)

Code GenerationCode Generation


CSP, …)


CSP, …)

M2T tools (Xpand/Xtend)M2T tools (Xpand/Xtend)

MetatoolsMetamodel instrumentation

Defining new generative tools

MetatoolsMetamodel instrumentation

Defining new generative tools


Limitations of MDE

MDE does not let you do something “new” It may take more time, but you can do it

MDE does not help if your idea isn’t right You may have your prototype in 2 weeks but, it does not help you get good

numbers

29


Some Remaining Challenges Model transformation verification and

testing Property preserving model transformations

Coping with a multitude of DSLs Avoid defining and tooling of a DSL from

scratch Inter-language reasoning

Scalability Compilers handle large models

Maturity of MDE tools30


Perspectives

MDE can learn from optimizing compilers in: Knowledge of the platform (PDM) Efficiency and scalability culture

Anticipate use of DSLs to grow Thus, every software engineer will have

to deal with compilation chains for DSLs

31


Main Message

Compiler may not be so far away from you!

3 years is short, but don’t just look at your tiny little domain many fields are closely connected of course, sometimes it’s good to focus

32


Acknowledgements

Some slides in this talk was taken from my a talk for our paper done by Robert France Benoit Combemale Steven Derrien Sanjay Rajopadhye Clement Guy Antoine Floch


33

Research in Compilers and How it Relates to Software Engineering Part III: Relation to SE Tomofumi Yuki EJCP 2015 June 22, Nancy.

Documents