Why SCM & Bugtracking? • Even simple changes WILL have side-effects • You can see WHAT the change was in SCM • You need to know WHY the change was made hence the Bug # (and SCM-/Tracking integration) • Only then you can reliably fix the side-effect without breaking • previous bug fixes • other features November 2012 - EKON 16 Sebastian P.R. Gingter - @PhoenixHawk
Why continuous integration? • Fail fast as insurance • Fastest possible feedback on • THAT something broke • and from the logs also WHAT broke November 2012 - EKON 16 Sebastian P.R. Gingter - @PhoenixHawk
Why continuous integration? • Why is that good? • Developer is still in the topic • She‘s able to directly fix the issues without having to re-think the situation when a bug report comes in a few days (or weeks) later • Other devs know of the problem and don‘t check out broken versions that causes them problems • Enables automated testing November 2012 - EKON 16 Sebastian P.R. Gingter - @PhoenixHawk
„stable branches“ • Ensure that you are able • to release a • stable version • at every given moment in time November 2012 - EKON 16 Sebastian P.R. Gingter - @PhoenixHawk
Requirement for „stable branches“ • Only develop in branches, never in trunk/master • Features • Bug fixes • Reintegrate into release-branch only if 100% sure • that it‘s tested • you‘re confident in stability November 2012 - EKON 16 Sebastian P.R. Gingter - @PhoenixHawk
Code Reviews You‘re probably doing it wrong! • Code reviews are distracting • WTF WTF WTF • „Look at that! Can you tell me what he meant by this?“ • Reviewer needs too much time to get in the code • Unproductive and time consuming • Reviewer could mob developer • „You‘re not working with our guidelines“ • „But *Everyone* knows that {very special way of doing things here}“ November 2012 - EKON 16 Sebastian P.R. Gingter - @PhoenixHawk
IF you want to do them: • do „Over the shoulder“ reviews • Developer seeks reviewer in person • Developer explains to reviewer • Reviewer asks questions and points to possible problems • Developer takes notes • Developer fixes • Developer checks in November 2012 - EKON 16 Sebastian P.R. Gingter - @PhoenixHawk
Costs of pair programming • Overhead in total hours • Usual guess? • x2, or 100% • 50 % initial overhead while the pair syncs up • About 15 % overhead on the long run November 2012 - EKON 16 Sebastian P.R. Gingter - @PhoenixHawk
• Task: 8h • Together: x1.15 à 9.2h • Together: 9.2h / 2 à 4.6 h • Half a day + half an hour • Task: 8h • Alone: x1 à 8h • Alone: 8h / 1 à 8h • One full day Let‘s do the math Single Developer Pair November 2012 - EKON 16 Sebastian P.R. Gingter - @PhoenixHawk
Gains of pair programming • Faster time-to-market • About 15 % less defects • Since costs to fix a defect later is higher than instant fixing that alone saves more than pair programming costs • About 20 % less lines of code • Better architecture • Distributed knowledge • Higher motivation • Less distraction November 2012 - EKON 16 Sebastian P.R. Gingter - @PhoenixHawk
November 2012 - EKON 16 Sebastian P.R. Gingter - @PhoenixHawk But it works http://blog.jphpsf.com/2012/09/30/OMG-test-driven-development-actually-works http://www.typemock.com/blog/2009/03/05/the-cost-of-test-driven-development
November 2012 - EKON 16 Sebastian P.R. Gingter - @PhoenixHawk „Our experiences and distilled lessons learned, all point to the fact that TDD seems to be applicable in various domains and can significantly reduce the defect density of developed software without significant productivity reduction of the development team.“ Source: http://research.microsoft.com/en-us/groups/ese/nagappan_tdd.pdf
Let‘s look at some figures • First blog post: • Full TDD (test first) for everything • 50% less defects - after 2 years • No mention on costs, though • Second link (study from IBM & MS) • Costs: Time only. 15% to 35% more • Gains: Bug ratio went down to 61% (and one product down to 9%!) November 2012 - EKON 16 Sebastian P.R. Gingter - @PhoenixHawk
Source: „Benefit from unit testing in the real world“ November 2012 - EKON 16 Sebastian P.R. Gingter - @PhoenixHawk Source: http://blog.jphpsf.com/2012/09/30/OMG-test-driven-development-actually-works
Problems with TDD • Existing code is usually not testable • Code is only written testable if tests are there upfront • New features within existing code are hard to cover with TDD • Writing tests is not easy and fast to learn • especially when there‘s no TDD culture and mentors around November 2012 - EKON 16 Sebastian P.R. Gingter - @PhoenixHawk
First easy steps to tests • Write new code with TDD • Start with simple features to get used to it • If you access not-tested infrastructure, wrap it with something mockable that just passes through in production • Write tests for existing code only • when you need to do major refactorings • when you fix a bug and that code is already testable • or almost, with little work November 2012 - EKON 16 Sebastian P.R. Gingter - @PhoenixHawk
Some sources • „Clean Code“ by Robert C. Martin • Mentions: • http://alistair.cockburn.us/Costs+and+benefits+of+pair +programming • http://blog.jphpsf.com/2012/09/30/OMG-test-driven- development-actually-works • http://www.typemock.com/blog/2009/03/05/the-cost-of-test- driven-development November 2012 - EKON 16 Sebastian P.R. Gingter - @PhoenixHawk
phoenixhawk
nwiizo
sansantech
seosoft
shibuiwilliam
kddisnowvillage
kazuhitotakahashi
dahatake
chiroito
colopl
kgsi
yoshidashingo
yoshiitaka
jrom
eitanlees
jcasabona
addyosmani
trallard
bermonpainter
eileencodes
jonyablonski
keithpitt
ufuk
tanoku
shpigford