You have to deliver ambitious new features but your codebase is a huge mess of legacy technologies, with no test? It is very tempting to throw it all away and rewrite everything from scratch, but is it wise when you consider the associated cost, risk and delayed time-to-market? Through an experience report, we'll show a "Strangler Application" strategy where only carefully selected areas of legacy code are rewritten. Agile development techniques like BDD or TDD remain necessary, with some adjustments. We'll also describe step by step the overall thinking process you can use to deal with large legacy code bases efficiently. First presented at Agile France 2013, and in countless Brown Bag Lunches since, a best-seller! Video (In French) here: http://www.infoq.com/fr/presentations/code-legacy

1. Legacy Code
Evolve or
Rewrite ?
A retrospective
Cyrille Martraire - @cyriux -

2. http://1.bp.blogspot.com/_SOzNIrpkT7M/TIhYjDKKz-I/AAAAAAAABjo/NCPSaVmZn0w/
s1600/100_2824.jpg

3. Passionate
developer
PARIS
Since 1999
!
@cyriux

4. Paris Software
Craftsmanship Community
http://www.meetup.com/paris-software-craftsmanship/

5. TDD
BDD
DDD
Legacy

7. LEGACY

8. Legacy:
Evolve or rewrite?

9. Actual Example

10. Once upon a time…

11. There was a
frustrated ambition

12. We Want:

13. Large Ambition

14. Asset Management

16. wanna do big things
stuck with a legacy

17. So what do we do?

18. REWRITE

19. STOP!

20. We all want to
rewrite

22. hard to release
regressions

23. Legacy

24. Legacy
No Test
& Technical Debt too

25. http://skillsmatter.com/podcast/design-architecture/talk-from-eric-evans

26. Plenty of Legacy code
working since 1993
Redoing = Waste + Risk

27. Poorly designed code
Working well
Never changed
Is OK

28. KEEP iT

29. KEEP GUI, DB,
APP Server,
middleware

30. We Want
NEW FEATURES! CHANGE
SOMETHING

31. Haz Change?

32. JUST
DO IT!

33. QUICK

34. Did we break
something?

35. No Test
http://martinezglass.com

36. Takes a lot of time
No Test
+ Debug!

37. Just Do It:
Not that quick
Again and again

38. Could we do better?

39. EVOLVE

40. Surround with tests
FIRST

41. Surround with tests!

43. Testing legacy code is
hard & slow

44. Testing is an
investment

45. Driver = Cost of
testing

46. A matter of velocity

47. Legacy testing effort
Vs.
All new tested code?
Is it worth it?

48. 1:4
(Ok, I totally made it up)
ratio

49. REDO

50. Differentiating to
the business?

51. The CEO will tell you

52. Not differentiating?

53. BUY

54. &ADAPT
your
process

55. Differentiating!

56. REWRITE

57. "What if we also rewrote
this crappy part?"
• Why not, but not now
• Don't get lost redoing
everything forever

58. ...

59. JUST
ENOUGH

60. You can rewrite
You cannot
rewrite all

61. What to rewrite?
Where to
stop?

63. What to rewrite?
Find
boundaries

64. Discover the app
!
!
Read the documents
Interview people
Use the application
Read the code
Fix bugs

65. Visualize
[primary flow & support features]
http://www.fimarkets.com/pagesen/back-middle-front-office.php

66. Visualize
[primary flow & support features]
http://www.fimarkets.com/pagesen/back-middle-front-office.php

67. Visualize
[primary flow & support features]
http://www.fimarkets.com/pagesen/back-middle-front-office.php

68. http://staff.bath.ac.uk/enssa/LPDev.htm
Asset
Inventory

69. Asset Capture
Transmission
Exhaust
http://martinfowler.com/bliki/AssetCapture.html
Engine

70. Asset Capture
Transmission
Exhaust
Focus on one single asset : the engine
http://martinfowler.com/bliki/AssetCapture.html
Engine

71. Main concepts
[in selling-something business]
Trading
Financial
Instruments

72. Main concepts
[in selling-something business]
Trading
Financial
Instruments
DON’t
TOUCH
REWRITE

73. DOMAIN FTW!

74. BOUNDARIES
& STRUCTURE

75. Agree on maxims
[communicate the plan]
• “Only one
work site at
a time:
instruments”
!
• “When in
Rome, do as
Romans do”

76. "The code says it all"

77. Legacy: Fossilized
Domain Expertise
•inspiration
•obfuscated
http://newspaper.li/fossil/

78. Rewrite:
!
because of CHANGES
(or it’s a bad idea)

79. needs

80. Rewrite:
!
must inject NEW knowledge

81. Rewrite:
!
need business ANALYSIS

82. WHOLE
TEAM

83. Like a new
project from
scratch!

84. So how do
we
integrate?

85. STRANGLE

86. Strangler
application
A new module will
progressively
strangle the
former one, for
one single asset
only
http://martinfowler.com/bliki/StranglerApplication.html
http://www.flickr.com/photos/louisfoecy/4114597043

87. Feature Toggles
!
Progressively
turn off legacy
functionalities
& turn on the
new ones
OLD NEW

88. if(c == category1)
{legacy.compute(...)}
else if(c == category2)
{strangler.compute(...)}
//...

89. Grow a
Bubble
Context

90. !
Listen to domain experts
High hygiene standards
Pair-programming
TDD, BDD, DDD

91. Invest in domain
knowledge
• Bi-weekly 30mn
training sessions

92. Dream model must pass
every scenario
Given a floating rate bond on EURIBOR 3M
And a nominal of 15M EUR
And an issue date of 2011/06/15
And an end date of 2012/06/14
And an SEMI_ANNUAL calculation period !
When the EURIBOR 3M evolves:
Bespoke custom
DREAM MODEL
| 2011/09/15 | 3.5% |
| 2012/03/15 | 4.0% | !
Then the cash-flows are:
| 2011/12/13 | 23000 |
| 2012/06/14 | 25500 |

93. Anti-corruption
layer

94. Adapter

95. API
SPI

96. Easier to integrate by
construction
bespoke lib

97. Old code: Freeze it

98. Rewrite
=
clear boundaries
+
new strangler application
+
minimal integration

99. Deliver
Value
Quickly

101. Questions? Did you try
similar things too?
Let’s discuss!
@cyriux

102. Follow me @cyriux
!
Slides: slideshare.net/cyriux
Blog: cyrille.martraire.com
!
In Paris? Join !

103. Merci