Untangling Maven dependencies


The Maven Project Object Model (POM) of a Java project contains references to all the libraries that are used in that particular project. These libraries are called dependencies in Maven and Maven handles for you the dirty work of downloading and managing the packages of the dependencies and including them in your project. It is a formidable tool but it can confront you with tricky puzzles from time to time.

In many Java test automation projects you’ll find references to Selenium libraries because Selenium is a popular tool for GUI test automation. In the project that I worked on as a test automation engineer some Selenium libraries were used. My Maven dependency challenge began when I wanted to start using Headless Chrome for UI automation. Headless Chrome is a handy browser because it does not require the actual Chrome browser so it is possible to run UI tests without the actual UI. It takes away some of the problems of using normal browsers for GUI automation, such as sluggishness and the need to be precise about the browser window size.

The starting point

I join the test automation project when it is already in full swing. As such, there is an existing Maven POM file which contains references to some Selenium libraries. There is, for example, a reference to the Selenium Chrome Driver library. This is the client library that is used in the Java code. The dependency is included in the Maven POM as below.


As you can see the version of the library is 3.0.1 which is a version that was released in October 2016.

In order to run Headless Chrome, it appears that at least version 59 of the Chrome browser is needed. This version of Chrome was released in May 2017, so in order to avoid compatibility conflicts I think it best to upgrade the version of the Selenium Chrome driver library. In retrospect I cannot find evidence that Chrome version 59 depends on a later version of the Selenium Chrome driver library, but this does not occur to me at that moment.

Making a mess

As we saw above the Selenium Chrome driver library can be included as a separate dependency in the Maven POM. But it is not necessary to include this driver library and other driver libraries (such as those for FireFox and Internet Explorer) separately since they are also a compile dependency of the encompassing Selenium Java library. So by including the Selenium Java library as a dependency I would be able to get rid of all the separate Selenium dependencies, which makes the POM a whole lot simpler. I decide to try this approach and copy the dependency from the POM of another test automation project. Here is the code that I copied.


As you notice, the version of the library is 3.4.0. This is by no means the latest version, but at the time I forget to look up what the latest version is. I am really satisfied having reduced the number of Selenium dependencies from five to one and hit the ‘maven clean install’ command to build the code. To my amazement, the code does not compile!

Untangling the web of dependencies

The compiler tells me that there is a certain method that is used in the Java code that cannot be recognized anymore. The method is JsonObject.keySet(), which is used in some part of the code that processes JSON files. The only thing I changed in the project was the Selenium dependencies, so I am flabbergasted by the fact that suddenly the compiler is tripped up by a JSON error. The two are entirely unrelated. In the import statement of the class in which the keySet() method is used it appears that the Google Gson library is somehow imported. This means that the Google Gson library should be mentioned in the Maven POM file. I look at the POM file. There is no Google Gson dependency! I roll back the changes that I made to the POM file. Still no Google Gson dependency! Where does this dependency come from??

Then I remember that Maven does a lot of the heavy lifting for you. Libraries can have compile dependencies and Maven manages the compile dependencies for you. We saw that the Selenium Java library has all the Selenium drivers as compile dependencies and that creating a dependency for the Selenium Java library dismisses us from having to import these driver libraries separately. So the Google Gson library must be a compile dependency of a library that is actually mentioned in the POM file!

I use IntelliJ as an IDE and IntelliJ has a useful array of plugins. One the plugins is the Maven Helper. This plugin allows you to dig deeper into the web of dependencies that is set up in the POM. It has a handy little search box that allows you to search through the dependencies and their compile time dependencies. Using this tool I find that the Google Gson library is actually a compile time dependency of the FireFox driver library. It is a bonus—so to say—that come with the Firefox driver package. See the screen shot below for the search using the Maven Helper plugin.

Furthermore, I notice that the version of the FireFox driver library that I use needs version 2.8.2 of the Google Gson library. Then I roll forward to the changes that I made in the POM file and see that version 3.4.0 of the Selenium Java library that I introduced only has version 2.8.0 of the Google Gson library as a compile dependency. Could it be that the keySet() method that was not recognized after I made the changes to the POM was introduced in version 2.8.1 or version 2.8.2 of the Google Gson library? Luckily the Google Gson Change Log provides the answer: the keySet method was introduced in version 2.8.1. So I cannot use version 3.4.0 of the Selenium Java library, unless I create a separate dependency for the Google Gson library in the POM. But that would lead to a dependency conflict.

Fixing it

Then I look again at the version of the Selenium Java library. Is 3.4.0 really the latest version or is there a later version that includes version 2.8.2 of the Google Gson library as a compile dependency? A quick search in the Maven repository tells me that version 3.11.0 is actually the latest version and that this version has version 2.8.2 of the Google Gson library as a compile dependency. I change the version of the Selenium Java library in the POM and hit ‘maven clean install’ again. This time the code builds successfully.

Used tools

  • IntelliJ
  • IntelliJ Maven Helper plugin
  • Maven repository browser
  • Google Search

New Nationalities in the Testing Blogosphere


Recently I’ve been adding some new weblogs to my overview of software testing weblogs. The overview now lists 267 weblogs on software testing. I do not claim to have an overview of all software testing weblogs, but I think I have quite a large number. So for me, it was strange to notice that in the last few days I added four weblogs from countries that were not yet on my list. Proof of the fact that blogging about testing is truly an international sport.

These are the newly added weblogs, their authors and their countries…


Martial Testing by Andrés Curcio and Ignacio Esmite


AskTester by Thanh Huynh and others


One Software Tester by Jason B. Ogayon


Mr. Slavchev by Viktor Slavchev

A Software Tester’s Bookstore


I think it can be said that I am known for keeping lists. There is a simple reason, I like to collect resources related to testing. Mostly I do this in the fine spirit of Umberto Eco’s Anti-Library (see Nassim Nicholas Taleb’s The Black Swan for reference), to be aware of what’s there and what’s not. And, in the fine spirit of Rumsfeldian epistemology, to be aware of the fact that I’m unaware of what’s there and what’s not.

Anyway, for a while I used Google Books to track suggestions for and references to books that came across my path. I reached a total of 100 books (which is nothing, compared to the 50,000 Eco is said to have in his library) and thought it might be nice to categorize this list and show it on my website. I believe most of the books in the list contain at least bits (as in portions) of valuable information for modern-day testing.

I placed the books in an Amazon bookstore and I link to a flat list below. The categorization contains faults and could have been done in many other ways. The category ‘Management’ could, for practical purposes, be split up into ‘People management’, ‘Project management’, ‘Risk management’, ‘Leadership’ etc… These categories in their turn open up a wide landscape of sub categorizations. For now, I use a relatively flat categorization, to prevent anyone from getting truly lost. And, as categorizations go, they can never be false.

Yes, books and categories (such as coaching) are missing. As always, additions are welcome.

Here is a link to the store.

And a link to the list (PDF, opens in new window).


The oddness of two familiar words


Today, finally, the oddness of two words dawned on me. The words are ‘software tester’ and I used them, in a conversation, to describe my role in the project I am currently involved in.

The first reason why I felt strangely alienated from the description ‘software tester’ was that, for last couple of months, I have not tested software. Not a single piece of software has been been submitted to me with the question “Can you please test this?”. Instead, I’ve been reviewing (mostly functional) design. All I’ve seen is complex thoughts spelled out in writing.

Reviewing – from a testing perspective – is hard work. I have sat down with other reviewers and no one of these persons ever described reviewing as even relatively easy. It can be tedious and mind boggling at the same time. It involves reasoning from many perspectives. It involves recognizing the traps of your own reasoning and that of others. And it takes a lot of effort to find out which set of perspectives will lead to a manageable conclusion.

This brings me to the second word, which is ‘tester’. Today, through a couple of discussions, I realized that the tester – the independent assessor of quality – is as much bound to the context, as much the modifier of context as he is the assessor of the design. Though we would like to see ourselves as the evaluating party, it is this evaluation, this test, that is part of the design, part of the meaning of a system in its context. It reminded me that the design is bound to the tester just as it is bound to the designer. That means that we should be excruciatingly careful in selecting our approach to testing and that we should be looking at not only the result of the test but at the effect of the test on the design. The latter may turn out to be a much bigger issue than the former. Which is why I think ‘ tester’ is just a very odd word.

A benignly malicious list of books


In line with the good old context-driven tradition I’m presenting, in this blog post, a list of books. I decided to use the adjective ‘malicious’ to characterize this list. As far as I know there are no books in this list that have actually done harm. But I think the knowledge in these books may be used in a benignly malicious way. If software testing is about evaluating beliefs in software systems and if good software testing is relentless in exposing the value of those beliefs then we, as testers, had better be armed with the knowledge to dissect those beliefs, ruthlessly. And since we are not paid to expose only those things that make everybody happy, there is a certain malicious aspect in our craft.

So what I want is to be as ruthless as possible in my analysis. And in order to achieve that a certain amount of knowledge is neccessary. Also, to survive as a software tester, it may be neccessary to be five or twenty steps ahead when it comes to reasoning. This list is one of the ways to get closer to that goal.

I did not read all of these books. In fact, most of them are on my list of books that I want to read. Most of the titles are taken from philosophy and psychology. I also decided to leave out, as much as possible, books that are said to be life- or game-changing. In other words, I  want the straight stuff, from the source, not from the hype.

I think Nelson Goodman’s Languages of Art should be a very promising read. Furthermore Feyerabend’s Against Method seems to be one of the founding documents of context-driven testing. More on that book in some future blog post.


A satisfying testing experience on a rainy Sunday


The problem statement: an annoying pause

The issue was really quite simple: on all devices in a family home (three laptops, two tablets and one smart phone) the display of videos through the internet (whether through Youtube or another channel) was very slow. I was shown a video on the iPad. The beginning of the video was loaded, was paused after a few seconds because more of the video had to be loaded, the video ran on for a couple of seconds and then was paused again to load more of the video. This went on and on so viewing a twenty minute video was a very annoying, gruelling, experience. The lagging video display had been frustrating the family for quite some time.

I was visiting the family. It was a Sunday afternoon. It was pouring.

The system setup

There is a home network to which, through a wireless router, the three family laptops all running a Windows operating system are connected. Also using the wireless router are two tablets, one iPad 2 and another unidentified tablet running some version of the Android operating system. Also, one (Android) smart phone will probably use the wireless router for internet access. Not all devices are switched on at the same time. The wireless router is linked to a DSL modem. Both modem and router are of the Cisco brand, types of modem and router were not investigated.

First inference

Since all devices suffered from the same lags in loading and displaying videos I concluded that neither the operating systems nor the devices were the first candidates for investigation. In fact the speed of the internet connection would be a good starting point for the investigation.

The requirements

At that point two things needed to be established. One, the expected speed of the connection and two, the actual speed of the connection. The expected speed of the connection was established by looking up the service contract with the internet provider. The download speed should be 50 Mbps.

The first test: how fast is the internet connection?

With this in mind it was time to investigate the actual speed. This could be done with an online tool on the website of the provider. But since I wanted a second tool to verify the results of the tool offered by the provider I found a nifty little iPad app that also measured the speed of the internet connection.

Using these two applications, one on one of the laptops and one on the iPad it was established that the download speed of internet connection was at least fifteen times slower than the speed promised by the provider. According to the tool offered by the provider was about 3 Mbps (several measurements in a time frame of a couple of minutes) and the speed according to the iPad app was even less; 0,5 Mbps ( (again several measurements in a time frame of a couple of minutes).

Notice that there is quite a difference between the measurements with the respective tools. I decided not to bother with the difference since a) it would probably be impossible to find the cause of the difference and b) because both measurements clearly indicated that the connection speed was (very) low. The classification high, medium, low, very low contained enough granularity for me at that time.

Another thing that struck me was the the upload speed was actually higher than the download speed. Normally, in DSL lines, the download speed is significantly higher than the upload speed. I did not investigate into this further but it may have been a pointer towards the reason for the problems.

I was told that the wireless router had fallen at least once. The owner of the router thought that that may have caused a defect in the router.

The second test: is the wireless router the cause of the problem?

To eliminate the router as a cause I reset it a couple of times, without any effect on the download speed. Then I decided to take the router out of the equation. I connected the laptop directly to the modem. Should the router be the cause of the troubles then with this new setup I should have a good download speed. It was possible to connect the laptop to the modem but there was no internet connection. This was the point at which I was baffled a little bit. I thought connecting the laptop to the modem should be a plug-and-play thing. But apparently this was not the case.

To see if I could get the internet connection working through the modem I rebooted the laptop. Without success. Then I started paying attention to the suggestions that Windows offered by which I should be able to fix the connection. One of them was to adjust the DHCP settings. Since I am not a wizard at network settings, I decided to leave that alone at first. One other suggestions was to reset the moden.

The solution: beaten by Windows

Why hadn’t I thought of that before? I was beaten to the chase by Windows! In the early days of DSL I had tinkered with modems and routers for the better part of an afternoon to get the network going. I remembered having tried all sorts of boot and reset sequences. The one I still use infrequently is to shut down the router, shut down the modem, boot the modem, wait till it’s fully functional, then boot the router, wait till the router is online and then start the pc or laptop.

I skipped the ‘internet connection through the modem‘ test and applied the somewhat familiar boot sequence. Then I grabbed the iPad and hit the speed test application. It now showed a very nice and satisfying download speed of about 20 Mbps. I was happy to report that the internet connection speed had just increased by a factor 40.

The deliverables: happiness despite the rain

But was that 20 Mbps speed really satisfying? Our requirement said that the speed in fact ought to be 50 Mbps. But then I remembered that DSL connections seldom reach the specified speed. It has something to do with the distance from the switch, the number of houses using that connection, possible the quality of the ethernet cables and interference from other devices. Interference could not be ruled out because the modem and the router were close to a phone, a television, a dvd player and such.

I could have done many other tests in search for the cause of the issue. I could have finished the ‘internet connection through the modem‘ test. I could have upgraded the firmware of the router and the modem and test the effect of that. I could have tried a different ethernet cable. But the result of the session was reasonable and spending more time on testing seemed like a waste. I handed the iPad to the owner and he conjured up the video – a cooking instruction video – that had been plaguing him for quite some time. This time it ran without a flaw. As an extra test on the laptop I loaded a lengthy video in Youtube (Goranka Bjedov – Using open source tools for performance testing). Goranka came through clear and without a hitch.

Though the actual cause of the error had not been found I was happy to hand to the family two testing products that may help them in the future: 1) a nice little iPad tool to measure the internet connection speed and 2) a procedure to reboot the internet connection just in case.

I like to think I delivered exactly what was needed.