using information fragments to answer the questions developers ask

Using Information Fragments to Answer the Questions

Developers Ask

Thomas Fritz and Gail C. Murphy

University of British Columbia

2

Who is working on what?

What is the evolution of the code?

What [have] people done lately?What classes has my team

been working on?

Who is working on the same classes as I am and for which work item?

Who has made changes to my classes?

What classes have been changed?

Which features and functions have been changing?

Has progress been made on blockers (blocking work items)

in your milestone?

What caused this build to break?

Who owns a test case?

Which API has changed?

Who owns this piece of code?

How is the team organized?

3




been working on?






in your milestone?






Test

Who

people team

Who Who

Who

Who

team

Team

4




been working on?






in your milestone?






Test

what

features

work item

blockers

What

What

Team

Work Items

5




been working on?






in your milestone?






Test

changing

changed

changes

changed

evolution

Changes

progress

Team

Work Items

6




been working on?






in your milestone?






Test

API

classes

classes classes

codeCode

classes

code

ChangesTeam

Work Items

7

Code

ChangesTeam

Work Items

TeamChanges

Work Items

Code

8

TeamChanges

Code

Work Items

9

TeamChanges

Code

Work Items

Can we automate the integration of information

from multiple domains?

11

“What has my team been working on?”

Team Fragment

12


Work Item Fragment

13


Change Set Fragment

14


Source Code Fragment

15


16


17

Team Fragment

Work Item Fragment

Change Set Fragment

Source Code Fragment


18

Information Fragment Model

Defect 303

WI 315

Work Item Fragment

id

Team Fragment

David

Julie

Allen

Φ(,)

Φ (,)

Presentation 1

Presentation 2

Composition

19

Information Fragments

Subset of information

F =(V, E, lF )

WI 315: Can’t ...id: 315owner: David

duplicate of

D303: Popup can ...id: 303owner: Julie

V: uniquely identifiable items with domain, type and properties

E: relationships between nodes

l F : mapping V → {Fragments}

20

Composition

F' : union of input fragments (nodes/edges)

+ newly created edges based on properties

Julie

id: Julie

FWI

WI 315: Can’t ...id: 315owner: David

D303: Popup can ...id: 303owner: Julie

David

id: David

Allen

id: Allen

FT

owner

owner duplicate of

idF' =

21

Presentation

WI 315

D 303David

Julie

WI 316

Φ (t, wi)

Φ(wi, t)

owner

owner

creator

creator, owner

Allen

23




been working on?






in your milestone?






Information Fragment

Model

24

1. Can developers use the model to answer the questions?

1. A2. Can developers use the model effectively

without detailed understanding of the model ?

Multiple Cases Study with 18 professional developers

Evaluating our Model

25

Source Code

Change Sets

Teams

Work Items

(1) What classes have been changed? X X(2) Who has made changes to my classes? X X X(3) What has changed between two builds and who

has changed it?X X X

(4) What have people been working on? X X(5) What classes has my team been working on? X X X X(6) What’s the most popular class? (changed most) X X(7) Who owns this piece of code? (modified most) X X X(8) How do recently delivered changes affect changes

that I am working on?X X

Evaluation – Method

• 8 questions from original 78 to cover domains of interest

26

Evaluation – Method

• Questions made more specific to reduce interpretations and match available data

“What classes has my team been working on?”

Yesterday, on which classes (of the SCM code) have Alex and Allen on the SCM team been

working on and why? For each developer name one class and

the reason for the change.

• 10-15 minute tutorial

• 18 developers from 4 different teams

• 8 questions from original 78 to cover domains of interest

27

Evaluation – Prototype

28

Evaluation – Results (Success)

(1) (2) (3) (4) (5) (6) (7) (8)0

6

12

18

Questions

# S

ub

ject

s

94%success

Failure

Hint Given

Success

29

(1) (2) (3) (4) (5) (6) (7) (8)0

2.5

5

7.5

10

Questions

Tim

e (i

n m

inu

tes)

Evaluation – Results (Time)

mean time2.3 mins

30

1. Can developers use the model to answer the questions?

2. Can developers use the model effectively without detailed understanding of the model?

• 94 % success rate• 2.3 mins mean time• 10 – 15 mins training• hint given in only 9 cases (6%)

Evaluation – Results

pp

31

Easy to use

Evaluation – User Comments

“I throw everything in and reorder it [...] and see if it seems reasonable.”

Answers multi-domain questions

“for this type of thing I never really thought of trying to solve these problems [...], I try and find

some obscure way [instead]?”

“it is answering questions that I don’t think we can answer right now”

32

Understanding links takes time

Evaluation – User Comments

“lot of trial and error”, but “made sense in the end”

Participants liked the tool

“pretty nifty”

“pretty cool”

“cool”

“neat tool”“definitely useful”

“really really cool”

“well done”

33

Evaluation – Threats

Predefined information fragments

8 questions may not be representative

Unfamiliar data

34

Discussion

How to best select information fragments?– actual selection– predefined

Is it possible to automatically choose composition?

35

Related Work

Following information linksFeldspar [Chau et al. 2008], JQuery [Janzen et al. 2003]

Querying for informationRelational views [Linton 1984], CodeQuest [Hajiyev et al. 2006]

Automatically linking informationHipikat [Cubranic et al. 2005], Deep Intellisense [Holmes et al. 2008]

Composing different informationFerret [de Alwis et al. 2008]

36




been working on?






in your milestone?






Information Fragment

Model

using information fragments to answer the questions developers ask

Documents

team changes code work

piece of code

team fragmen

test case

integration of information

information fragments

multiple domains