Why we banned statics and then allowed again?

Blog

Why we banned statics and then allowed again?

Selenide Advent Calendar
Day 9

Good evening!

Here is the 9th day of Selenide Advent Calendar, and
I’ll tell you what excites the public most of all.

Why statics were banned in Selenide 5.0.0, but allowed again in 5.4.0?

Short answer: we banned them occasionally. But it was good.

Let me explain it in detail :)

How Selenide holds WebDrivers

Selenide stores webdriver instances in ThreadLocal.
It allows you to run tests in parallel: different threads get different webdrivers. The code looks like this:

class WebDriverRunner {
  private static final ThreadLocal<WebDriver> webdriver = new ThreadLocal<>();
  private static final ThreadLocal<SelenideProxyServer> proxy = new ThreadLocal<>();
}

Say, method $("a").click() works like this:

  WebDriverRunner.webdriver.get().findElement("a").click();

Briefly about SelenideDriver

But ThreadLocal implies one limitation: you cannot use two webdrivers in one thread
(as well as one webdriver in two threads - but aren’t there yet).

We planned to solve this problem in Selenide 5.0.0. We created a special class SelenideDriver, that allowed you to use two webdrivers in one test:

SelenideDriver browser1 = new SelenideDriver();
SelenideDriver browser2 = new SelenideDriver();

browser1.open("https://google.com");
browser2.open("http://yandex.ru");

browser1.$(h1).shouldHave(text("Google"));
browser2.$(h1).shouldHave(text("Yandeks"));

It forces us to do a major refactoring: we removed all usages of old good static method open(), $ and $$ inside Selenide itself. Every piece of Selenide code that needs a webdriver now gets a SelenideDriver parameter.
Yes, we had to pass this parameter everywhere…

Here comes the question

What should we do with old good static methods open(), $ and $$? They have to get SelenideDriver instance somewhere. Where to take it from?

Selenide 5.0.0: Statics got a punch

We added to class WebDriverRunner a third ThreadLocal:

  private static final ThreadLocal<SelenideDriver> selenideDriver = new ThreadLocal<>();

Briefly, SelenideDriver is a simple class with two fields WebDriver and SelenideProxyServer.
In general, it worked and solved the initial problem.

What I could not have envisioned at that moment was that so many folks defined their web elements in static fields:

public class MyPageObject {
  public static SelenideElement fname = $("#fname");
  public static SelenideElement lname = $("#lname");
}

And re-open the webdriver between tests.

One important detail:

We did one more improvement in Selenide 5.0.0.

Before 5.0.0, Selenide automatically opened a new browser if it wasn’t opened yet. At any moment.
Sometimes it caused perplexity because browser opened at unexpected moments (for example, when trying to log an error caused by a crashed browser).

Of course, it happened because of bad tests. Bug after all, we created Selenide to help people, right?

That’s why starting from version 5.0.0, Selenide didn’t occasionally open a browser anymore.
Instead, it said: “There is no an opened browser, I cannot do my job. You need to call method open() first.”

And what failed then?

The coincidence of these two factors led to the following problem:

  1. Test creates a static field static SelenideElement fname = $("#fname").
  2. This fname remembers a SelenideDriver it was created with.
  3. Test closes the browser in the end.
  4. A following test opens a new browser tries to use static field fname.
  5. And fname calls its own instance of SelenideDriver it was created with.
  6. And fails, because that `SelenideDriver has been closed.

More than a year - from September, 2018 to October, 2019 - I was trying to explain that static variables are evil.
I even created a special talk “Antistatic” (yes, this time in English:)).

But finally I gave up. Because it would be too big refactoring for many folks to rewrite their projects from static variables.

After all, we created Selenide to help people, right?

Selenide 5.4.0: Statics won

How we finally solved this problem?

The solution is quite simple. We replaced ThreadLocal<SelenideDriver> by a static variable (yep :))

private static final SelenideDriver = new ThreadLocalSelenideDriver();

Now this “static” SelenideDriver is a singleton. It always exists. It’s never closed. All static SelenideElement variables created with it will live forever. But it also doesn’t hold fields WebDriver and SelenideProxyServer. It fetches them from WebDriverRunner’s ThreadLocals every time.

P.S.

That’s why method WebDriverRunner.getSelenideDriver() disappeared in Selenide 5.4.0.

I was surprised that many people have already managed to use it. People, I do not understand you! How do you manage to use everything so wrong?
Well, it was my mistake to make this method public. But I never mentioned it in any of my posts nor documentation. I never recommended to use. How could someone decide to use it? How this magic happens?

What’s now

We returned the possibility to declare your SelenideElements in static fields.

But please do not abuse it.

I still don’t like it :)


Andrei Solntsev

selenide.org

blog comments powered by Disqus

The project is maintained by