Paper: “Impact of Software Engineering Research on the

Practice of Software Configuration Management

Authors: Estublier, Leblang, Hoek, Conradi, Clemm, Tichy, Wiborg-

WeberCitation: ACM TOSEM Oct 2005

The Impact Project • Provide scientific scholarly answers to:

– What impact has academic and industry research really had on the practice of software engineering?

– What future impacts should be expected?– What future directions will software research take?

• How?– ACM Sigsoft project (international)– NSF and Sigsoft funding– EU, Japanese, private funding- Deliverables: journal articles, conference panels

2000 - 2003

Initial Subject Areas

• Reviews/Walkthroughs – Dieter Rombach/Dewayne Perry

• Configuration Management – Jacky Estublier• Testing and Analysis – Lori Clarke/David Rosenblum• Middleware – Wolfgang Emmerich• Process/workflow/lifecycle models – Volker Gruhn• Modern Programming Languages – Mary Lou

Soffa/Barbara Ryder• Requirements Engineering – Anthony Finkelstein/Axel

van Lamsweerde• Reverse Engineering – Hausi Muller• Cost/Economic Models

How do they define Impact?

The research must have been:

1. Published – publicly available, AND

2. Incorporated in actual SCM product that are (or were) on the market, commercially or free.

Other impacts not considered:- people (graduates)- workshops and conferences

Software Configuration Management is…

The discipline of managing change in large, complex software systems.

Goals: manage and control corrections, extensions, and adaptations throughout lifetime of software system

- Systematic and traceable software development process

- Managing files and directories

In the beginning…

• 1950s • Aerospace industry• Colored punch cards

• 1960s• Integrated within OS

• 1970s• Separate discipline

Evolution of context of SCM Systems

Construction•Building•Snapshots•Regeneration•Optimization

(create exec)

Construction•Building•Snapshots•Regeneration•Optimization

(create exec)

Auditing•History•Traceability•Logging

(archive/rollback)

Auditing•History•Traceability•Logging

(archive/rollback)

Components•Versions•Configurations•Baselines•Project contexts

(keep track)

Components•Versions•Configurations•Baselines•Project contexts

(keep track)

SCM Spectrum of Functionality

Susan Dart, SCM-3, 1991

Accounting•Statistics•Status•Reports

(gather stats)

Accounting•Statistics•Status•Reports

(gather stats)

Process•Lifecycle support•Task mgmt.•Communication•Documentation

(choose tasks)

Process•Lifecycle support•Task mgmt.•Communication•Documentation

(choose tasks)

Controlling•Access control•Change requests•Bug tracking•Partitioning

(track change)

Controlling•Access control•Change requests•Bug tracking•Partitioning

(track change)

Team•Workspaces•Merging•Families

(conflicts)

Team•Workspaces•Merging•Families

(conflicts)

Structure•System model•Interfaces•Consistency•Selection

(how related)

Structure•System model•Interfaces•Consistency•Selection

(how related)

And,…remain universally applicable – PL and App independent

Partition of SCM Approaches

Product

- Versioning

- System Models and selection: support aggregate artifacts – configuration concept

Tool

- Workspace control: distributed users?, integration of change

- Building: executable

Process

- Support for general development processes to manipulate artifacts

Approaches to Versioning

• Capture artifacts as configuration items

• Track relations among items in version graph

Edges:- revision-of – seq develop- variant-of - || development- merge variants

1.0

1.1

2.0

1.2

2.1

1.2.1.0

1.2.1.1

When disk space was scarce:

• Delta storage (eg, SCCS, RCS): baseline+deltas• Data compression• Combination of delta and compression• More accuracy in deltas:

– Context-oriented– Operation-oriented– Semantics-oriented– Syntax-oriented

• But for generality -> classic line-based merging

Advanced Versioning – Change sets

• How it works:– Each change stored as a delta independently

from other changes– Allow more flexibility– Can combine changes as desired

• Not used in practice:– Deltas overlap/conflict – some combos do not

work– For binary objects – cannot combine some deltas

• Too unwieldy for large projects with large change

Alternative: Change PackagesTask, Activity, package,

Subproject,…

1.0

1.1

2.0

1.2

2.1

1.2.1.0

1.2.1.1

1.0

2.0

2.2

2.1

2.3

1.0

1.1

1.3

1.2

2.0

2.0.1.0

1.0

1.1

1.2

- track changes at logical level

Aggregating and Accessing Multiple Artifacts

• Data Models• Early 1970s – SCCS and RCS = file system• Since then – on top of commercial database

systems• Research systems:

– Adele, 1985: active, oop, versioned model– Object=any entity– Attribute=primitive, compound, predefined– Relations=model associations like derivation,

dependency, composition

• More advance commercial system: Aide-De-Camp 1989

Aggregating and Accessing Multiple Artifacts

• System Models – late 1970s onward– MIL (Module interconnection language) to

describe system structure• Model interfaces – provided, required functions• Behaviors - Pre and post conditions• Hierarchical construction of modules system architecture; UML

– Integrate into SCM – users can manage real organization of software

– Major Problem – keeping evolution of model and implementation versions in synch

Aggregating and Accessing Multiple Artifacts

• Selection– How do I get a set of artifacts in my workspace

without requesting them individually?– Default: All latest version in workspace. Fetch

the rest individually.– Other approaches:

• Hierarchical workspaces 1989. local, parent,…• General queries 1984.

– (status = approved) AND owner = Jacky) OR (date > 6.20.83)

• Leverage change-sets 2000. – Baseline 2.5 + bug-fix2.83 + bugfix.2 + feature-12

• Rule-based 88, 94. – First, my checked-out versions– Otherwise, the latest versions on my branch– Otherwise, the latest versions on the main branch

A Typical Development Scenario

CMrepository

Pete’s workspace

CBA

Ellen’s workspace

E CD

Workspace Control

3 functions of the workspace:

1. Sandbox – freely edit. May be locks.2. Building – expand compressed files,

keep compiled/derived objects3. Isolation – allow developer to make

changes, compile, test, debug without interference

Workspace Control

Classic SCCS and RCS Systems –no workspace management

CVS – first scripts on top of RCSNeed: avoid source file copies in 100s of

workspacesSun/Forte Teamware – manage projects of

subprojectsVirtual workspaces – only copies of files

editingClearCase – avoid recompiling sources on

builds

Building

• Make, 1979. – dependencies, date-based rebuild, fast.

• Improvement - Rebuild only if any source versions now in workspace are not exactly the same as in last build. – BOMs – bill of materials for each target object

built• Language-based smart rebuild: semantic

changes and dependencies• Winking-in (ClearCase) – language

independent, reuse binaries across workspaces

Process Support

Software Process = sequence of activities during creation and evolution

- Change control:- Change request (requirement change)- Trouble report (malfunction issue)

Landmark Contributions with Great Impact

Academic Industry

1972 SCCE

1976 Diff

1977 Make

1980 Variants, RCS

1980 Change sets (Aide-de-Camp)

1982 Merging, and/or graph

1984 Selection

1985 System model

1988 1st SCM workshop, process support, workspaces

1990 Virtual file system, multisite

1996 Activity-oriented SCM

Successful Transitions

• SCCE, Make, RCS – immediate, long lasting impact

• Change sets – slow, nonpractical, but standard feature – change packages.

• Process support – advanced support for modeling and enforcing process

• Differencing/merging – binary deltas, not semantic-based

• Distributed/remote development – client-server protocol, web-based interfaces

Failed Transitions

• Semantic-based recompilation – language dependent

• Advanced systems models – more power than needed

• Generic platform - research has focused on managing source code only. Too much needed for extra artifacts.

Summary

- High impact Research – useful, ease of use by developer, generality

- Low impact Research – level of complexity too high, not easy to master idea as a feature

What is Next for SCM?

• How to fit SCM with rest of development process/tools?

– Manage other artifacts beyond source code

– Maybe not be language independent

Recognizing a Valuable Resource: Mining Software

Repositories• Configuration management repositories

are traditionally a “depot”– occasional roll-back– occasional search for relevant information

• But what if we used the information captured by configuration management repositories to our advantage– understanding software developers– helping software developers