Категории: Все - assurance - testing - design - reliability

по Paweł Badeński 14 лет назад

343

There is such thing as too much testing

Determining the appropriate level of testing for software development is crucial from various perspectives, including the code, application, and team. Testing should ideally catch bugs and ensure good design practices.

There is such thing as too much testing

There is such thing as too much testing

Lessons Learned

It is supreme importance to keep your test suite fast and reliable
Unreliable test suites

Devs stop wondering what caused the fail and 'fix' the build in bad ways

Team will stop caring about fails

Slow suites

Fix it

Remove redundant tests

Profile your suite and examine slow tests

Stop hitting external services or database

Run on a fast machine or in the cloud

Parallelize

Start to see the test as an obstacle

Spend a lot of time on twitter

Devs won't check in near end of day

Integration tests can replace unit tests if you're not careful
Selenium/watir tests are deceptively easy to get into but hard to maintain
Start small if the team is new to testing

What's the Right Level of Testing

From the application's point of view
What level of assurance does it need?

Someone have to fill the form second time

Someone looses life

Somebody looses money

From the team's point of view
What's the level of buy-in for managers
What's the level of developer buy-in for testing

gridlock, poor velocity

morale problems

From the code's point of view
Tests (integration tests especially) should be catching bugs
Testing is supposed to be about good design

The Tour

The Social Behemoth Part III
Above 90% coverage
SQuiD
Right for the project - had to be out of the door really quickly!
No integration testing
No view testing
The Social Behemoth Part II
60% coverage
The VOIP Signup App
ridicilously well-tested code
RoR
very high quality demanded
The Social Behemoth
0 - 50% coverage
only 50% of the team (consultants) wrote tests

they saw tests as impediment

us versus them philosophy

even managers could write some code
if you write a spike do not start building on top of that
The Small Social Start-up
Extensive Selenium testing

not so good if your doing it just to test if it works in the userspace

bulletproof app

The IP Space Manager
Government
Ruby/Rails project
The Huge Financing App
test suite

it probably made poeple in the company feel better

they didn't know why they did run them

80% of these tests failed!

8 computers 36 hours

test hostile world

EJBs 1.0 shit

21% coverage, 25% 4 years later
The Dealer Portal
WebSphere

a bit heavy

webunit tests

slow build (15-20 minutes)

a lot of them

small project (6 devs/5 months)