Working with Legacy codebase

Transcript

1. Effective approaches when working with huge legacy codebase By Paul Taykalo, Stanfy Paul Taykalo, UMT 2016 1
2. Legacy code Paul Taykalo, UMT 2016 2
3. Plan 4 What is Legacy Code 4 Problems with Legacy Code 4 Dealing with Legacy Code 4 Features vs Refactoring balance 4 Measure our Success :) Paul Taykalo, UMT 2016 3
4. What is Legacy Code? Paul Taykalo, UMT 2016 4
5. an amount of money or property left to someone in a will :) Definition Legacy Code /lɛɡəsi/ Paul Taykalo, UMT 2016 5
6. So here we'll tell what legacy code is not :) Legacy code is not... Paul Taykalo, UMT 2016 6
7. Even the code is bad it doesn't make it a Legacy. Bad code can be covered with tests, and works fine. And sometimes more then enough. Don't mark it as a Legacy if you think that it's just look bad Bad Code is not always a Legacy Code Paul Taykalo, UMT 2016 7
8. It's always good to blame other people for making a legacy code. If you know people who are writing the code, and you can reach them - they writing kind'a ok code. In case if they left their job, for some reasons people automatically tend to think that code of people who left is legacy. by default. Other people's code is not a Legacy code Paul Taykalo, UMT 2016 8
9. From the other hand... Paul Taykalo, UMT 2016 9
10. This works backwards too Nice looking code can be a Legacy code Paul Taykalo, UMT 2016 10
11. Code without tests is bad code. It doesn't matter how well written it is 1 Michael C. Feathers Working with Legacy Code Paul Taykalo, UMT 2016 11
12. Why do we so care about it? May be Legacy Code is just another buzzword we complains about? Why do we care about Legacy Code? Paul Taykalo, UMT 2016 12
13. Legacy code is... Paul Taykalo, UMT 2016 13
14. Legacy code is... Complex Paul Taykalo, UMT 2016 14
15. It strikes you when you aren't expecting it. At Friday's evening, for example https://www.youtube.com/watch?v=DT2X5b0tyDI Legacy code is... Unpredictable Paul Taykalo, UMT 2016 15
16. Legacy code is... Unmaintainable Paul Taykalo, UMT 2016 16
17. Because all of above, it's really hard to support this kind of code in reasonable timte Legacy code is... Time consuming Paul Taykalo, UMT 2016 17
18. And beacuse of that Legacy code actually becomes a proble. It just stops feature delivery Legacy code ... Stops features delivery Paul Taykalo, UMT 2016 18
19. Ask yourself, do you have Legacy in the Project? Why do we write it? Paul Taykalo, UMT 2016 19
20. here are some generators which produce Unmaintainable code Unmaintainable code generators Paul Taykalo, UMT 2016 20
21. Sometimes it's really "Tecnical Debt", when customer comes in and asks to make feature really fast. But some lazy developers tend to use Unmaintainable code generators 4 So called "Technical Debt" Paul Taykalo, UMT 2016 21
22. No comment here Unmaintainable code generators 4 So called "Technical Debt" 4 Code without Tests Paul Taykalo, UMT 2016 22
23. YAA makes project way more complex, an way more harder to support Unmaintainable code generators 4 So called "Technical Debt" 4 Code without Tests 4 Yet Another Architecture Paul Taykalo, UMT 2016 23
24. It's the same for another cute awesome library. In ost cases it's better to avoid additional library. You just can't allow to add to your code even more complexity Unmaintainable code generators 4 So called "Technical Debt" 4 Code without Tests 4 Yet Another Architecture 4 Super Awesome Library X Paul Taykalo, UMT 2016 24
25. Funny one. Other people generates Legacy Code by default. It is known Unmaintainable code generators 4 So called "Technical Debt" 4 Code without Tests 4 Yet Another Architecture 4 Super Awesome Library X 4 Other people* Paul Taykalo, UMT 2016 25
26. How to avoid writing Legacy Code? Paul Taykalo, UMT 2016 26
27. So here ae some tecnics that will prevent us from writing legacy code. Dealing with Legacy Code Preventive measures Paul Taykalo, UMT 2016 27
28. Easy one Stop writing code! Paul Taykalo, UMT 2016 28
29. If you don't write any code - you won't be able to add more legacy to it. Also, it's obviously that reading few lines of code is way simpler than reading 200 Stop writing code! 4 More code - more legacy code 4 Less code - less errors 4 Less code is easier to read Paul Taykalo, UMT 2016 29
30. E=mc2 Errors = ( more code )2 Paul Taykalo, UMT 2016 30
31. This is actually contrexample, when you need to write more code to make it more readable. The Idea here is to make code readable and understandable. Check two exampes above Less code exceptions mViewModels.value.map { $0.map { $0.map { $0.price }}} mViewModels.value.map { sections in sections.map { items in items.map { item in item.price } } } Paul Taykalo, UMT 2016 31
32. Start adding tests Paul Taykalo, UMT 2016 32
33. Hey, I'm captain obsious here, but you need to add them, seriously Start adding tests Paul Taykalo, UMT 2016 33
34. You should start writing tests from tomorrow. Start adding tests If you don't write tests you will never learn how to write them, its better to write bad tests then not to write any — https://t.co/KgfdqTbOSr Paul Taykalo, UMT 2016 34
35. Start adding tests Whenever you are tempted to type something into a print statement or a debugger expression, write it as a test instead. — Martin Fowler Paul Taykalo, UMT 2016 35
36. I have an opinion about this part. Depending on the project size So here, depending on the side of the project, if project is small, you would probably wont' receive any time saves, just beacuase o fproject size. So technically test setup and runs will take you way more precious time to do. Paul Taykalo, UMT 2016 36
37. Generated code. Really helpful to prevent mistakes. If it works - it works for every generated class. if it wont - it won't Do code generation Paul Taykalo, UMT 2016 37
38. Do code generation! 4 Stops you from manual work 4 Unifies project architecture 4 Prevents you from human-like mistakes Paul Taykalo, UMT 2016 38
39. If you're about to stick to the same architecture, it means, that you can predefine some parts to be in templates. This will decreas amount of time to rewrite things, and also Make templates for classes or big code parts Paul Taykalo, UMT 2016 39
40. Pull Requests. At least more than one person will know about what happening in the code base Let other people read your code Paul Taykalo, UMT 2016 40
41. Am I doing better today? Is it better then previous year? Dealing with Legacy Code Know your status Paul Taykalo, UMT 2016 41
42. Know your status Paul Taykalo, UMT 2016 42
43. More than that. Make your code easy to delete Dealing with Legacy Code Delete code Paul Taykalo, UMT 2016 43
44. EACH big project will have unused classes/methods etc It's superb to / Libraries Remove old unused code Paul Taykalo, UMT 2016 44
45. Just remove it. Easily. It's simple Remove old unused code func removeItem(item: MenuItem) { var sectionsToRemove: [CartOrderSection] = [] // mSections.value.forEach { section in // section.removeItem(item) // let containsNonPinnedItems = section.items.value.contains { $0.menuItem.pinned != true } // if !containsNonPinnedItems { // sectionsToRemove += [section] // } // } for section in sectionsToRemove { removeSection(section) } mSections.value = mSections.value } Paul Taykalo, UMT 2016 45
46. Not all IDEs will tell you about that. Be careful if you do dynamic classes Remove old unused code Paul Taykalo, UMT 2016 46
47. Feel free to mark classes/methods as deprecated, this will prevent these classes/methos usage across the prohect. You don't want really bad code to spread around more than it does now Deprecate unwanted code Paul Taykalo, UMT 2016 47
48. If you spent some time reading legacy code - and you cannot fix it now, but you unrestood what next possible steps should be Leave a comment. This will prevent you or other people for re- reading and rethinking the complex part of the code. Leave comments Leave warnings Paul Taykalo, UMT 2016 48
49. In general case dependencies always grows. We should recheck our deps from time to time Recheck and cleanup dependencies Paul Taykalo, UMT 2016 49
50. Instead of blindly rewriting old code in new "way", make sure that "new way" written code won't become Legacy in short We're forcing to have tests for the new functionality as well as when we're changing old parts of code those without tests Add tests on changes you make Don't use "Change and Pray" approach Paul Taykalo, UMT 2016 50
51. It's really hard to stop fixing everything, when there are a lot of things wrong, but you should be careful about big breaking changes One refactor at the time Paul Taykalo, UMT 2016 51
52. There are some good tools for refactoring. Don't rely on the Xcode only. Some IDE's know how to do it correctly. Use tools for refactoring Paul Taykalo, UMT 2016 52
53. In case of big and long running project, there's a big desire to try some new better approaches to do something in new way. Try to avoid this as much as possible. i.e KVO vs manual observing vs Notifications. Stick to the same architecture within the project* Paul Taykalo, UMT 2016 53
54. This one is a bit cardinal solution, but there's always an option to put some parts of the project into the library. This will prevent logic leaking through All the layers of the project, and also decrease main project size. This could take a while, but it worth it. In worst case, we could try to make another application insteead of doing everthing in one app. Decrease project size Paul Taykalo, UMT 2016 54
55. Dealing with Legacy Code Understanding what went wrong Paul Taykalo, UMT 2016 55
56. Understanding what went wrong 4 Crashes 4 Bugs 4 Deadlocks 4 Data loss 4 Unexpected behaviours Paul Taykalo, UMT 2016 56
57. Captain obvious :)) Everyone doing it, right? Understanding what went wrong Crash logging Paul Taykalo, UMT 2016 57
58. Here are some tools for crash Understanding what went wrong Crash logging Tools 4 Crashlytics 4 HockeyApp 4 TestFlight 4 BugSee Paul Taykalo, UMT 2016 58
59. This can be done while trying to replicate things Understanding went wrong Debugging Paul Taykalo, UMT 2016 59
60. Xcode, LLDB, Chisel Understanding went wrong Debugging tools 4 Xcode 4 LLDB 4 Chisel 4 View Debugging / Description Paul Taykalo, UMT 2016 60
61. This can be done while trying to replicate things Understanding went wrong Remote Logging Paul Taykalo, UMT 2016 61
62. Here are some tools for remote logs Understanding went wrong Remote logging tools 4 Crashlytics 4 LogEntries Paul Taykalo, UMT 2016 62
63. Reminder Paul Taykalo, UMT 2016 63
64. Reminder Fighting Legacy Code is not our goal Paul Taykalo, UMT 2016 64
65. If you give enough time to the developers, they will rewrite code again and again and again until it will be "perfect" At least for some time Features delivery We aren't paid for fighting with Technical Debt Paul Taykalo, UMT 2016 65
66. But each feature 4 Adds more code 4 Adds complexity 4 Increases project size 4 Increases testing time Paul Taykalo, UMT 2016 66
67. This one is hard. Depending on how big your project is, you can either work with with new features and Features vs Tech Debt Paul Taykalo, UMT 2016 67
68. Different features types Paul Taykalo, UMT 2016 68
69. Different features types 4 Affects nothing (independent) 4 Affects one localized previous feature 4 Affects multiple previous features 4 Affects all application Paul Taykalo, UMT 2016 69
70. Always have time to clean up a project. Plan accordingly. From the other hand, focus on feature while doing it And don't spend any time on rewriting old functionality - Put TODOS, comments etc. 90/10 Principle* Paul Taykalo, UMT 2016 70
71. Measure Paul Taykalo, UMT 2016 71
72. Some techical stuff, easy to read, easy to check shoud be check automatically //TODO : Graph Image Measure 4 Tests count 4 Test coverage 4 Code Lines Count 4 Tests Lines Count Paul Taykalo, UMT 2016 72
73. Measure Paul Taykalo, UMT 2016 73
74. Not everything can be tracked automatically, but can be measured. Results need to be checked at retru // TODO: Other reports Measure 4 Crash-free users/sessions 4 Number of days since last incident 4 Amount of bugs create in the last X days 4 Regressions count Paul Taykalo, UMT 2016 74
75. Measure development speed Paul Taykalo, UMT 2016 75
76. Measure development speed 4 Features count per time 4 Time per feature to be delivered Paul Taykalo, UMT 2016 76
77. Recap Paul Taykalo, UMT 2016 77
78. Recap 4 Get the project status 4 Set up goals 4 Tack changes 4 Stop writing Legacy Code Paul Taykalo, UMT 2016 78
79. Stop writing Legacy Code Thank you Paul Taykalo, UMT 2016 79
80. Links 4 https://vimeo.com/138774243 4 https://tech.blacklane.com/2015/12/13/test- automation-for-ios/ 4 http://inessential.com/2016/06/03/ thecleanscreen_habit 4 https://engineering.instagram.com/instagram- android-four-years- Paul Taykalo, UMT 2016 80
81. Effective approaches when working with huge legacy codebase By Paul Taykalo, Stanfy Paul Taykalo, UMT 2016 81