2015-06-18

Useful git aliases when working with Maven and Jira

Put these into your .gitconfig below the alias section:

# Return all jira issue keys from commit messages
issues = !sh -c 'git log --oneline $@ | egrep -o [A-Z]+-[0-9]+ | sort -V | uniq' -
# Fetch id of latest commit from maven release plugin
release-commit = !sh -c 'git log --all --grep=maven-release-plugin.*prepare.release --format=%H| head -1'
# Return all jira issue keys since last usage of maven release plugin
issues-since-release = !sh -c 'git issues $(git release-commit)..'


2012-06-04

Using Infinitest for real

Making your Test driven development faster in real projects

When you are practicing TDD like I do you may blindly know your keyboard shortcut to start your test runs after you've changed something. Or do you still use your mouse? :-)

Infinitest is a great tool to get rid of those shortcuts. Whenever a change occurs in a class it runs the tests that accompany it. Well, that's already awesome but fasten your seatbelt: Infinitest also runs tests of classes that are affected by your change.

How they do this magic I haven't elaborated but I can assure you of the effect: When using Infinitest you will notice net effects of your changes much faster.

But...

I always loved Infinitest for it's incredibly short feedback loop, yet disabled it in my current maven multi module project for seemingly obvious reasons: "No, I am too lazy don't have time to figure out how to properly set it up" (That of course I would not confess to myself.)

This meant I didn't figure out how to configure it so it wouldn't always mess up my own dev environment or, being quite more of an impact, the integration test environment.

As you might guess I have successfully set up Infinitest for this monster of a project.

The project

The Project is a maven multi module project with a dozen modules and specifically a module containing most of the integration tests. It is completely stored in a remote git repository. I have set up a local virtual machine containing a database for testing and a local memcached. The project is configured via a set of -D properties that specify which of these to use during integration tests, so that these can be run locally or on the dedicated test instances running on the integration server during the Jenkins CI build. The CI build runs on every push to the server repository.

And here comes the problem: Whenever a local integration test fails to run on the local database engine, but uses the integration instance instead, it may interfere with a running CI build. 

For a thing like Infinitest, which runs a couple of tests on each change, the test runs (including the integration tests) may happen quite often. So I have to make sure Infinitest propagates the properties of i.e. where to find the database to the test cases.

The easiest solution (read: solution of the lazy one) is of course: switch it off.

Beware: configuring Infinitest for such a project requires a bit of work if you don't have it enabled directly after you started. But it can be done and is well worth the effort.

There are a couple of constraints you have to consider when using Infinitest:

  1. When you are working in Eclipse you can not switch it on or off on a per project basis
  2. Settings are configured using files
  3. You have to have a set of files for each project
  4. Test cases may be filtered by
    1. Using regular expressions and/or
    2. Using TestNG groups
  5. Arguments are passed as -D properties in a files
See the Infinitest user guide for more information.

Preparations

The best way is: Have your integration tests in a single module. If you're using TestNG, use a group attribute in the test annotation. If you can't do that, move them into a separate package. If you can't do that and are not using TestNG, use a naming convention for your integration tests so you can easily filter them out. If you can't do that, you're screwed (well, I'm open for suggestions :-).

Setting it up

At first create a file called infinitest.args that contains all properties that have to be propagated to the test cases. Then create a file infinitest.filters that contains all filters for your integration tests (i.e. TestNG group or package regular expression or regular expression for the test case name. These files I'll call the master files for reference later on.

Put these two files into the top level maven project folder. Then make soft links in each module to these files. You can't do that? Well, you are on your own now ;-)

Hint: A very nice property of git is that it stores symbolic links in the repository. So you will share the links with the other project participants.

Don't put your master files into version control, or others will share the settings, which you probably don't want. But maybe you don't care whether they're using your database, do you? So you just put them into the .gitignore file.

Remember to not ignore the links in the module directories, of course.

So you should be fine right now. Just check the "Continuously test" option and enjoy the short feedback cycle from now on :-)

Hint: A slight modification which you may consider: For each project where you don't want to execute any tests just put an infinitest.filters with content .* instead of the symbolic link into the module directory.

So I hope this short guide helps you to get faster in test driven development and getting rid of manually executing tests via the mouse or a keyboard shortcut or whatever else you use. Please let me know if you have any corrections or suggestions.

My final notice: if you want to get really fast, just start using Infinitest whenever you start a new project. It's worth it!

Thanks for reading, anyway :-)

2012-03-29

Rules for test driven bug fixing

Just a quick note to myself for how to do bug fixing:

  1. Write an integration test that reproduces the bug
    This should create a test failure which will guard against regressions later on
  2. Write integration tests for the other cases
    You should write at least one test. This should prove that it works the other way so you'll feel safe you don't break behavior later on
  3. Track down the problem to unit level
    Debug to the lowest level where you can find the real cause of the problem
  4. Write a unit test that reproduces the bug
    This should create a test failure which will guard against regressions later on
  5. Write unit tests for the other cases
    You should write at least one test. There may be a chance that other corner cases have slipped as well, so examine closely. Keep in mind that unit tests are cheap!
  6. Now, fix the bug
    Change it so all unit tests are green
  7. Clean up
    Tidy up, don't leave a mess.
    Remember: Someone certainly will have to come back here. If that one is you, you will spend much more time, cursing the one who left that mess than if you do it right now!
  8. Run the unit tests again
    Verify your clean up didn't break anything
  9. Run the integration tests
    These should be green by now
  10. Integrate changes
  11. Create release

2012-03-13

Reasons For Developing Without Tests

I've come to write this short article because I had some discussion with a developer about why he does not test. This discussion made me remember that many teams either don't do testing, neither on unit nor on integration level, or do it very sparsely. So I've come to write down a list of reasons that I've encountered and some questions that anyone hiding behind those reasons should have answers for.

Before you go right on to flaming against automated tests, please bear this in mind:

A script or program
  • works in no more time than required
  • when repeated does exactly the same
A human
  • may be distracted
  • interprets, thus may not follow the test script exactly
  • takes shortcuts because he or she is bored of repetitive tasks

So here they are:

I don't have time for this

But you take the time for testing every change you make by hand each time, via the debugger or through the user interface? Or you don't and hope it will work? You want to continue doing a job manually that could be automated?
Don't you have some more important work to do?

I don't know the testing framework/tools

Learn them. The basic ramp up time for creating a unit test is very small. Every major IDE or build tool has support for testing.

Later on, when you start to integrate your tests into the build process, you will find out that Ant and maven both support JUnit out of the box.

I don't know if the boss will allow it

Ask him. Maybe he's glad that you asked and takes this as a sign of you taking responsibility for what you do.

I know the boss won't allow it

So you get paid to do a job that a script could also do, in less time and with fewer errors. Do you think he will like that when he discovers that?

I fear it will take too long

Try it in small steps. Consider that if you do nothing about it, it will take much longer in the long term.

I don't know where to start

Just start where you are, right now. The worst thing you can do is not do it, because it won't get better.

You will see, it's not as hard as you thought.

Are there any other reasons you know? Please provide them, I will see if I find an answer that convinces you.

Formatting curl xml result via xmllint

So I got fed up recently when reloading a url to check an xml result using chrome because I was using HTTP basic authentication also. Chrome was annoying me because each time i switched from no auth to auth I had to manually remove all browser data and restart chrome.

So I tried to use curl for that job. The only annoyance was that the xml is not formatted properly.

So xmllint for the rescue

curl 'yoururlhere' | xmllint --format -

and that's it.

2011-01-21

Warum ich TDD liebe

Ich liebe TDD. Nicht deswegen, weil es so ein toller Hype ist, oder weil mich die Theorie vollkommen überzeugt hat. Sondern deswegen, weil es mir schon oft den Hintern gerettet hat.

Gestern Abend hatte sich herausgestellt, das ich es mal wieder versaut hatte. Das kommt nicht mehr so häufig vor, wie früher, aber es kommt immer noch vor. Nicht, dass man mir deswegen einen Vorwurf gemacht hätte. Wir sind alle Menschen, und Menschen machen nun mal auch Fehler.

Das eigentliche Problem war, das ich die Änderung des Verhaltens nicht in allen betroffenen Problemfällen getestet hatte. Aber ich fange einfach mal anders an:

Es war Donnerstag spätnachmittags. Nachdem ich Erfolg bei der Implementierung der Erweiterung gemeldet hatte, war die neue Version auf dem Testssystem installiert worden. Nun kam es beim Testen zu Laufzeitfehlern, unter Anderem zu meinem Lieblingsfehler, der NullPointerException. (Thomas, Du erinnerst Dich? :-)

Mir fiel auf, dass ich einen Fall, der im Betrieb sehr häufig vorkommen würde, einfach vergessen hatte, zu testen. Genauer hatte ich keine Tests für die entsprechenden Konstellationen geschrieben, und deshalb waren die Probleme nicht aufgefallen. Ich hatte mich in der Sicherheit der Tests (insgesamt allein ca. 140 Integrationstests) gewogen und fiel aufgrund der Fehlermeldung aus allen Wolken.

Nun war es bereits Donnerstag spätnachmittags, kurz vor Feierabend, und das Feature sollte am Freitag demonstriert werden. Die ersten Perlen des Angstschweißes bildeten sich auf meiner Stirn. Natürlich nicht wirklich, dazu neige ich nicht, aber ich merkte, das ich ein ernsthaftes Problem hatte.

Nun tat ich, was zunächst jeder tut, ich bekam Panik, und versuchte, auf Biegen und Brechen das Verhalten zu ändern. Aber immerhin hatte ich vorher noch etwas sehr Vernünftiges getan, wie sich später herausstellen sollte: Ich hatte Testfälle erstellt, die das bestehende Problem belegten.

Das genaue Problem war ein Fehler in einem Datencontainer, der, um die Daten später schneller wiederfinden zu können, auf dem Datentyp Short als Schlüssel basierte. Durch die Erweiterung konnte aber nicht mehr Short als Schlüssel benutzt werden, sondern es musste String-basiert sein.

Wie man sich vorstellen kann, ist das Ändern eines öffentlichen Interfaces, das diverse Zugriffsmethoden per Schlüssel besitzt, keine leichte Aufgabe, da ja auch andere Klassen als Client diese Klasse benutzen und es beim Durchführen der nötigen Änderungen dadurch zu diversen Compile-Fehlern kommt.

Wie gesagt, tat ich, was ich früher in solchen Fällen getan hatte. Aufgrund meiner aufkommenden Panik versuchte ich, mit dem brutalen Ansatz „Suchen-Ersetzen“ dem Problem der Interface-Änderung beizukommen, was sich natürlich aufgrund des Maßstabs der Änderungen als unmöglich herausstellte.

Nach zwei erfolglosen Versuchen und circa eineinhalb Stunden unnütz investierter Zeit gab ich kurz auf. Der Vorteil, wenn es spät ist, ist: man ist allein im Büro und kann in Ruhe nachdenken. Das tat ich, und beschloss einen neuen Schlachtplan:

  • Compile-Fehler außerhalb der zu ändernden Klasse und der Testklasse ignorieren
  • Nur kleine Iterationen
  • Sobald die Klasse und die Testklasse selbst wieder kompilierbar sind, sofort alle diese Tests ausführen, mit anderen Worten : Red – Green – Refactor

Nach einer halben Stunde hatte ich die Änderungen an dieser einen Klasse und ihrer Testklasse durchgeführt. Danach machte ich mich daran, zunächst die Compile-Fehler in den benutzenden Klassen zu beheben, und als alles wieder kompilierbar war, checkte ich ein.

Dann führte ich die Integrationstests aus und stellte erfreulicherweise fest, dass „nur“ circa 20 Test fehlschlugen.

Als ich dann letztendlich alle Fehler behoben hatte, und alle Tests wieder grün waren, waren insgesamt vier Stunden vergangen. Wobei ich zugeben muss, dass ungefähr die Hälfte der Zeit unnötig war, wenn ich Ruhe bewahrt hätte und wie immer vorgegangen wäre.

Nachdem ich eine Nacht über den Vorfall geschlafen habe, kann ich drei Schlüsse ziehen:

  1. Wenn man unter Stress gerät, neigt man dazu, in ursprüngliche (und in diesem Fall schlechte) Vorgehensweisen zurückzukehren.
  2. Ohne die vorhandenen Tests wäre ich nicht in der Lage gewesen, so vorzugehen, wie es letztendlich nötig war, um zum Erfolg zu gelangen.
  3. Den Vorteil von TDD sieht man erst wirklich ein, wenn man während der Entwicklung in ernsthafte Probleme gerät.

Übrigens: Wäre dies zwei Tage eher passiert, hätte ich eine hervorragende Geschichte für den Vortrag vom 19.01. gehabt, aber dies ist halt keine konstruierte Geschichte, sondern wirklich passiert. Dann hätte ich aber wahrscheinlich auch den Vortrag nicht gehalten.

2009-04-15

Modal JDialogs

Recently our customers experienced problems running our software after updating to JRE6U12. In detail on Windows XP the desktop application would hang directly after connecting to the database. Downgrading the RE to U11 fixed the problem so we first thought it was a bug in the JRE.

Since google investigation did not bring any results except this we were convinced it should be our fault. After several hours of checking swing calls outside of the evnt dispatch thread, checking for undisposed dialogs and hunting down modal dialogs we found the following two statements.
dialog.setVisible(true);
dialog.setModal(true);
Do you see the problem? Neither did we at first. And since the code had worked before we did not think the order of these two statements would be the problem.

In fact it does not make much sense to set a dialog to modal after making it visible, but still these two lines were the cause of the main window losing focus, not reacting to input events any more, and for the user being unable to reactivate the main window. To the user this looks like a freeze of the application.

So now, after several hours of debugging, we know that we must call setVisible(true) after setModal(true)!

Followers