Reflection and Unit Testing

This post is prompted by a couple of things: (1) a limitation in the Moq mocking framework, (2) a look back at a unit test I wrote nearly 3 years ago when I first arrived at my current company. While you can use Moq to create an instance of a concrete class, you can’t set expectations on class members that aren’t virtual. In the case of one of our domain entities, this made it impossible to implement automated tests one of our business rules–at least not without creating real versions of multiple dependencies (and thereby creating an integration test). Or so I (incorrectly) thought.

Our solution architect sent me an unit test example that used reflection to set the non-virtual properties in question so they could be used for testing. While the approach is a bit clunky when compared to the capabilities provided by Moq, it works. Here’s some pseudo-code of an XUnit test that follows his model by using reflection to set a non-virtual property:

public override void RuleIsTriggered()
  var sde = new SomeDomainEntity(ClientId, null );
  SetWorkflowStatus(sde, SomeDomainEntityStatus.PendingFirstReview);

  var context = GetBusinessRuleContext(sde);

private void SetWorkflowStatus(SomeDomainEntity someDomainEntity, WorkflowStatus workflowStatus)
  var workflowStatusProperty = typeof(SomeDomainEntity).GetProperty("WorkflowStatus");
  workflowStatusProperty.SetValue(someDomainEntity, workflowStatus, null);

With the code above, if the business rule returned by RuleUnderTest looks at WorkflowStatus to determine whether or not the instance of SomeDomainEntity is valid, the value set via reflection will be what is returned. As an aside, the “context” returned from GetBusinessRuleContext is a mock configured to return sde if the business rule looks for it as part of its execution.

After seeing the previous unit test example (and a failing unit test on another branch of code), I was reminded of a unit test I wrote back in 2012 when I was getting up to speed with a new system. Based on the information I was given at the time, our value objects all needed to implement the IEquatable interface. Since we identified value objects with IValueObject (which was just a marker interface), using reflection and a bit of LINQ resulted in a test that would fail if any types implementing IValueObject did not also implement IEquatable. The test class is available here. If you need similar functionality for your own purposes, changing the types reflected on is quite a simple matter.

Pseudo-random Sampling and .NET

One of the requirements I received for my current application was to select five percent of entities generated by another process for further review by an actual person. The requirement wasn’t quite a request for a simple random sample (since the process generates entities one at a time instead of in batches), so the code I had to write needed to give each entity generated a five percent chance of being selected for further review.  In .NET, anything involving percentage chances means using the Random class in some way.  Because the class doesn’t generate truly random numbers (it generates pseudo-random numbers), additional work is needed to make the outcomes more random.

The first part of my approach to making the outcomes more random was to simplify the five percent aspect of the requirement to a yes or no decision, where “yes” meant treat the entity normally and “no” meant select the entity for further review.  I modeled this as a collection of 100 boolean values with 95 true and five false.  I ended up using a for-loop to populate the boolean list with 95 true values.  Another option I considered was using Enumerable.Repeat (described in great detail in this post), but apparently that operation is quite a bit slower.  I could have used Enumerable.Range instead, and may investigate the possibility later to see what advantages or disadvantages there are in performance and code clarity.

Having created the list of decisions, I needed to randomize their order.  To accomplish this, I used LINQ to sort the list by the value of newly-generated GUIDs:

decisions.OrderBy(d => Guid.NewGuid()) //decisions is a list of bool

With a randomly-ordered list of decisions, the final step was to select a decision from a random location in the list.  For that, I turned to a Jon Skeet post that provided a provided a helper class (see the end of that post) for retrieving a thread-safe instance of Random to use for generating a pseudo-random value within the range of possible decisions.  The resulting code is as follows:

return decisions.OrderBy(d => Guid.NewGuid()).ToArray()[RandomProvider.GetThreadRandom().Next(100)]; //decisions is a list of bool

I used LINQPad to test my code and over multiple executions, I got between 3 and 6 “no” results.

RadioButtonListFor and jQuery

One requirement I received for a recent ASP.NET MVC form implementation was that particular radio buttons be checked on the basis of other radio buttons being checked. Because it’s a relatively simple form, I opted to fulfill the requirement with just jQuery instead of adding knockout.js as a dependency.

Our HTML helper for radio button lists is not much different than this one.  So the first task was to identify whether or not the radio button checked was the one that should trigger another action.  As has always been the case when grouping radio buttons in HTML, each radio button in the group shares the same name and differs by id and value.  The HTML looks kind of like this:

@Html.RadioButtonListFor(m => m.Choice.Id, Model.Choice.Id, Model.ChoiceListItems)

where ChoiceListItems is a list of System.Web.Mvc.SelectListItem and the ids are strings.  The jQuery to see if a radio button in the group has been checked looks like this:


Having determined that a radio button in the group has been checked, we must be more specific and see if the checked radio button is the one that should trigger additional action. To accomplish this, the code snippet above is changed to the following:

if($("input[name='Choice.Id']:checked").val() == '@Model.SpecialChoiceId'){

The SpecialChoiceId value is retrieved from the database. It’s one of the values used when building the ChoiceListItems collection mentioned earlier (so we know a match is possible). Now the only task that remains is to check the appropriate radio button in the second grouping. I used jQuery’s multiple attribute selector for this task.  Here’s the code:

 if($("input[name='Choice.Id']:checked").val() == '@Model.SpecialChoiceId'){

The first attribute filter selects the second radio button group, the second attribute filter selects the specific radio button, and prop(‘checked’,true) adds the ‘checked’ attribute to the HTML. Like SpecialChoiceId, Choice2TriggerId is retrieved from the database (RavenDB in our specific case).

Complex Object Model Binding in ASP.NET MVC

In the weeks since my last post, I’ve been doing more client-side work and re-acquainting myself with ASP.NET MVC model binding.  The default model binder in ASP.NET MVC works extremely well.  In the applications I’ve worked on over the past 2 1/2 years, there have been maybe a couple of instances where the default model binder didn’t work the way I needed. The problems I’ve encountered with model binding lately have had more to do with read-only scenarios where certain data still needs to be posted back.  In the Razor template, I’ll have something like the following:

@Html.LabelFor(m => m.Entity.Person, "Person: ")
@Html.DisplayFor(m => m.Entity.Person.Name)
@Html.HiddenFor(m => m.Entity.Person.Name)

Nothing is wrong with the approach above if Name is a primitive (i.e. string).  But if I forgot that Name was a complex type (as I did on one occasion), the end result was that no name was persisted to our datastore (RavenDB) which meant that there was no data to bring back when the entity was retrieved.  The correct approach for leveraging the default model binder in such cases is this:

@Html.LabelFor(m => m.Entity.Person, "Person: ")
@Html.DisplayFor(m => m.Entity.Person.Name)
@Html.HiddenFor(m => m.Entity.Person.Name.FirstName)
@Html.HiddenFor(m => m.Entity.Person.Name.LastName)
@Html.HiddenFor(m => m.Entity.Person.Name.Id)

Since FirstName, LastName and Id are all primitives, the default model binder handles them appropriately and the data is persisted on postback.

XUnit: Beyond the Fact Attribute (Part 2)

One thing I initially missed about NUnit compared to XUnit (besides built-in support for it in tools like TeamCity) is attributes like SetUp and TestFixtureSetUp that enable you to decorate a method with variables that need to be set (or any other logic that needs to run) before each test or before all the tests in a test fixture.  When I first adopted test-driven development as a work practice, I felt it made things easier.  But the authors of NUnit eventually came to a different conclusion about those attributes, and implemented XUnit without those attributes as a result.

Rather than define attributes for per-test and per-fixture setup, the authors of XUnit recommend using a no-arg constructor where you’d use SetUp and IUseFixture where you’d use TestFixtureSetUp or TestFixtureTearDown.  While this took me some time to get used to, leveraging the interface made it easier to handle the external dependencies of code I needed to implement and test.  One technique I’ve adopted to give myself additional flexibility in my test implementations is to add an extension point to the implementation of the SetFixture method

In this example, the extension point is a method named AdditionalFixtureConfiguration.  Calling it inside SetFixture ensures it will be called before each test class derived from UnitTestBase.  Making the method virtual and keeping it free of implementation means that I only need to override it if I need additional pre-test setup for particular test scenarios.  Because we use StructureMap as our IOC container, the equivalent of UnitTestFixture class in my example has a public attribute of type StructureMap.IContainer.  The AdditionalFixtureConfiguration method provides a natural home for any setup code needed to configure additional mappings between interfaces and implementations, set up mocks, and even inject mocks into the container if a concrete implementation isn’t available (or isn’t needed).

While this is the implementation I’ve chosen, there are certainly other ways to accomplish the same thing.  Instead of defining AdditionalFixtureConfiguration in UnitTestBase, I could define it in UnitTestFixture instead and call it in every SetFixture implementation (or not , if that customization wasn’t needed).  I prefer having the abstract class because it makes for simpler actual test code.

Everyone is Junior at Something–Even You

Hanselminutes #427  was an excellent interview with Jonathan Barronville, the author (perhaps the most intelligent and articulate 19-year-old I’ve ever heard) of this article on Medium.  The discussion covered a lot of ground, and posed a number of thought-provoking questions.  Three of the questions struck me as especially important.

What is senior?

In the podcast, Hanselman suggested three criteria: years in industry, years writing code in a language, and age.  Since I’ve been in the industry for 18 years, have been writing production C# code for about 10 of them, and turned 40 at the beginning of the year, those criteria argue in favor of me being considered senior.  Those numbers can also work against me in a way (and not just because of the field’s well-known problems with age discrimination).  Before I took my current job (over 2 years ago), I hadn’t written any production ASP.NET MVC, jQuery or knockout.js.  The last time I’d written any production JavaScript before then was before also jQuery and node.js even existed.  So from the perspective of those technologies, I was junior (and still am in some respects).

While industry today seems to have a fetish for young developers, there is merit to that interest in one respect.  Men and women entering the industry right now, whether they’re fresh out of college (or even younger) are too young to have any memory of a world where Google didn’t exist.  They’re too young to remember a world before web pages.  Some of them have been writing software for the majority of their lives.  It’s natural to them in a way it wasn’t for me because I had to go to college to get access to really good computers and high-speed Internet.

That said, the number of years (or lack of them) isn’t an advantage or disadvantage if you haven’t had the sort of experiences you can grow from as a developer (and learned the right lessons from them).  Regardless of what age you were when you had the experiences, if you’ve had to build software that solved enterprise-level problems, dealt with scaling, refactoring and enhancement of large systems, or integration of systems, both succeeding and failing at addressing those challenges are what really make a senior developer.  More time in industry may give someone more opportunities to have those experiences, but if they haven’t had them, they’ve just been writing software for a long time.

What is the rush to be senior?

Hanselman made a comparison between tradesmen like carpenters and plumbers (who have to work as apprentices for 3-5 years and pass an exam before they can become journeymen) and our industry, where someone can have senior in their title without much experience.  While some of it (if not most of it) has to do with pay, there are drawbacks.  Because our field is relatively young in the grand scheme of things, there aren’t universally accepted standards and practices (especially compared to some branches of engineering, which have hundreds of years of history).  We place too much of a premium on speed, and not enough on depth of experience (and the time it takes to earn it).  One of the end results of this is the sort of interviews I’ve experienced on a regular basis.  I’ve seen tons of resumes from people with senior titles who are stymied by interview exercises that ask fairly basic questions (on the level of the Fizz Buzz test).

I’d been working for less than four years when I first got “senior” added to my title.  It came with a nice raise (which I was certainly happy about) and more responsibilities (team leadership), but I certainly wasn’t senior in terms of software development experience after that short a period of time.  Not unlike what the classic Peter Norvig essay suggests about teaching yourself programming in ten years, that’s about how long it took for me to see myself as legitimately senior from an experience perspective.  Even now, having spent over 17 years in industry, I’m sure there are workplaces where I wouldn’t be considered senior because I haven’t architected super-large systems or led a team with dozens of people—and I’m alright with that.  I’ve matured enough after this amount of time to be more concerned with what kind of work I’m doing (and what I’m learning) than I am with what title an employer gives me.

Are we okay with not knowing something and then learning?

This question points in two directions:

  • are we okay with ourselves not knowing something and then having to learn it?
  • are we okay with others not knowing something and then having to learn it?

For me, the answer to the first question is yes.  In the case jQuery and knockout.js (and other unfamiliar technologies like RavenDB), I had to be okay with not knowing.  Doing my own research, and not being too proud to ask a lot of questions younger developers on the team who clearly had more experience with those technologies was necessary to advance to the point where I could do all that work myself.

The answer to the second question is the key to many of the problems with our industry, particularly when it comes to issues of gender and diversity.  Too many in our industry go beyond not being okay with someone not knowing something and cross the line to being condescending, rude, and even hostile.  I’ve been on the receiving end of that kind of treatment more often than I care to remember.  Too many workplaces allow people with superior technical skills to act like children instead of adults in interacting with their co-workers.  There is more and more being written about the sexism in the industry (pieces like this one, and this one), but not nearly enough on the negative impact that environment has on the ability and desires of others to learn and grow as professionals.  I think the persistently low numbers of minorities and women in the tech industry has as much to do with the perception (if not reality) that a lot of tech companies have high “a**hole thresholds” as it does with insufficient exposure to math and science in school.

The bottom line for me from the article and the podcast is not only that everyone in this industry starts out as junior level, but that technology changes so quickly that we will all be junior at something at multiple points throughout our careers in the tech industry.  We need to keep that knowledge in mind so that we can be more patient with ourselves as we learn and with those we work with as they learn.

XUnit: Beyond the Fact Attribute

After using XUnit for unit testing the past couple of years, I finally got a bit tired of the cut-and-paste-and-modify cycle of expanding test coverage for functionality I was building.  I’ve used “row test” functionality in NUnit and MbUnit in the past, but hadn’t gotten around to finding out how to use them in XUnit until quite recently.  The process I followed was relatively short: (1) Add the xunit.extensions dependency to my existing test assembly, (2) take a test that class that contained significant duplication and re-factor it to leverage one of the row testing attributes available.

In my case, the area I picked that would benefit from row testing was workflow testing.  We implement our workflows as state machines with transition tables that define a starting state, an ending state, and the event that triggers the state change.  I’d previously tested these workflows with one unit test per transition, with those three attributes (start state, end state, event) being the only thing different between them.  This meant a single test  that took these attributes as parameters should be sufficient to test transitions, and that the row testing attribute(s) would contain the actual values.

The Theory attribute supports a number of different data sources, including inline data, Excel data, and SQL Server data.  I started with inline data believing it would be the simplest.  But I wasn’t able to get that to work because the InlineData attribute of Theory only appears to support primitives (int, bool, string, etc).  Our workflow events and statuses are classes with attributes that are static and read-only.  From there, I moved on to try the PropertyData attribute.  This attribute takes one string argument that needs to match the name of a property added to the test class that returns a collection of object arrays with one set of test parameters per entry.

Following that approach worked much better.  The results display pretty nicely in ReSharper as well:


I’m still working on a code example of this to make available on GitHub, but here are a few links I found useful in getting up to speed with the additional capabilities of XUnit:

Learning New Programming Languages

Important advice from The Pragmatic Programmer (page 62):

“Learn at least one new language every year.”

It’s advice I’ve been trying to follow more seriously since I first started reading the book last month.  One site I’ve been using to learn more JavaScript that’s proven to be pretty cool for that is (thanks Dean).  The katas are small enough that it doesn’t take a ton of research to figure out how to do something in a language you’re learning.  Once you’ve developed a working solution, you can see how others have solved it (and compare your solution to theirs).  Since you write and test the katas in the browser, there’s none of the overhead of firing up an editor or uploading your solution somewhere.  Ideally I’d be writing a few katas per day, but a few katas a week are what I’ve been able to manage so far.

Since Apple introduced yet another language (Swift) at WWDC earlier this week, I’m starting to learn that language as well.  So far, the syntax is a lot easier to grasp than Objective-C.  The only real hassle with writing the code examples as I read the language guide is that XCode 6 Beta crashes every half hour.

With both languages (or any language really), the real leap forward comes from building something non-trivial with them. Figuring out what that non-trivial something will be is another challenge altogether. I wish there were a sites like (or Project Euler) that put out larger-scale problems intended to be solved with software. Being part of the developer interview loop at work pushed me to create a few problems of that sort for use in interviewing developer candidates, but none of those exercises require more than 20 minutes of work. More significant challenges should make it useful to explore features beyond basic control flow and data structures.

When Third-Party Dependencies Attack

Last week provided our office with an inconvenient lesson in what can happen when third-party dependencies break in unanticipated ways.  PostSharp is a key third-party dependency in the line of business web application we sell.  On the morning of May 20, our continuous integration server (we use TeamCity) began indicating a build failure with the following message:

  • PostSharp.3.1.34\tools\PostSharp.targets(313, 5): error MSB6006: “postsharp.4.0-x86.exe” exited with code -199.

The changed file was a Razor template file–nothing at all to do with PostSharp.  Only one person on our development team was experiencing this error on their local machine, but the end result–not being able to compile the solution locally–pretty much eliminated the possibility of that person being productive for the rest of the day.  As the day progressed, the CI server began showing exactly the same error in other branches–even with no changes to code.  It wasn’t until the next day that we received the explanation (and a resolution).

Reading the entire explanation is worthwhile, but the key reason for the failure is this:

“we … assumed that all failures would be in the form of a managed exceptions. We did not anticipate that the library would terminate the process.”

The fail-safe code that PostSharp implemented around a third-party licensing component assumed all failures would be managed exceptions (which they could catch and deal with accordingly).  Instead, this third-party component simply terminated the process.  The end result–any of their customers using the latest version of PostSharp couldn’t compile any solution that included it.  There’s no way of knowing for sure how many hours of productivity (and money) was lost as a result of this component, but the amounts were probably significant.  To his credit, the CEO apologized, his development team removed the offending dependency and sacrificed the feature which depended on it.

There are many lessons to be learned (or re-learned) from what we experienced with PostSharp, but I’ll talk about three. First, if a third-party dependency is critical to your application and has a licensing option that includes support, it is best to pay the money so that you have recourse if and when there’s an issue. On the Microsoft stack, this is proving increasing costly as more third-party .NET libraries and tools raise their prices (there are quite a few formerly free .NET tools that have been purchased by companies and re-released as rather costly closed-source software).

Second, whether or not there are licensing costs, it’s a good idea to have more than one option for critical third-party dependencies. In the case of aspect-oriented programming on .NET, there are a number of alternatives to PostSharp. The vendor is even confident enough to list them on their website. So if licensing costs are significant enough a concern, it may be better to choose an open-source option that is less-convenient but gives you the ability to customize it than a paid option which doesn’t (and yokes you to a specific vendor).

Third, It may make sense to avoid taking on a third-party dependency altogether. When it comes to the Microsoft stack, it’s likely that they offer a framework or API with at least some of the capabilities you need for your solution. In the case of AOP, Microsoft offers Unity to support those capabilities. Especially in the case where you’re only considering the free tier of capabilities for a third-party dependency where Microsoft offers a product, if that free tier functionality isn’t a significant improvement, it may be best to stick with the Microsoft option.


Another Tale of a Developer Interview Loop

There are literally millions of links on the internet about interviewing developers–interview questions, posts on why the way you do it is wrong, and even guides on making it through an interview process.  Here’s one more, about the interview process at my current employer.

Before I joined my current company (back in 2012), before I even set foot onsite, there was a phone screen.  At the time, there seemed to be one developer responsible for phone screening every prospective developer candidate for the entire company.  If memory serves, the phone screen lasted around 45 minutes.  The questions were challenging, but not impossible to answer.  When the in-person interviews were scheduled, I had no idea what I was in for.  Over the course of 5 hours, I spoke to 7 different people who had some role in 2 or 3 different projects or products within the company.  The first hour, they put a laptop with Visual Studio in front of me and asked me to write a console app that performed three different tasks (I won’t go into too much more detail as we still use the same exercise to this day).  I was able to complete the exercise with enough time to spare for my two interviewers to ask me questions about my solution (which while it worked correctly, was not the most elegant).  The rest of the interviews were all questions, some behavioral/team/”fit”-related, but mostly technical.  All the developer hires that came in after me presumably went through a similar process.

Fast-forward to the fall of 2013–we’ve won a contract that more than doubles the amount of work we need to deliver.  The pressure is on to find, hire and onboard as many good developers as we can find.  An interview process that works just fine when you only hire a new developer every month or two scales poorly when you have a month or two to hire a dozen developers.  So we involve more developers in the interview process and cast a wide net for prospective hires.  After spending many man-hours interviewing candidates who struggle with our programming exercises, we find a few external candidates to hire–but far less than the dozen we need.  We end up grabbing people from other teams within the company to compensate.

So when our company changed the process again to involve developers in the phone screen process, I did some Googling to find out what sort of questions make an effective phone screen.  By far, the most useful post I’ve found is Steve Yegge’s Five Essential Phone-Screen Questions.  Reading (and re-reading) the whole thing is definitely worth your time.  Our recruiters only allot 30 minutes of time for our phone screens (and I usually have code to design & write, or bugs to fix), so my phone screen generally only covers 3 of Yegge’s 5 areas–coding, OO design and data structures.  In the coding area, instead of giving the candidates homework (or having them read the code over the phone), I started sharing a Google document with them and watching them write their answer to the coding question.  This is a great way to get a sense of how quickly a prospective developer hire can come up with a solution on the fly.  A more involved (and somewhat more buggy) approach, is to use the .NET Fiddle online console along with its collaboration feature.  If it doesn’t crash on you during the interview, you’ll be able to see if the solution compiles and runs successfully on the spot.  Thirty minutes has proven to be enough to get in a coding exercise and enough questions about OO design and data structures to have a good feel for whether or not it would be worthwhile to move someone on to the in-person interview phase of our process.  Since in-person interviews are generally conducted in pairs, each 30-minute phone screen that properly rejects a candidate saves 2-4 man-hours of additional interview time.

If there is any revision I would make to the current interview process, it would be to push our simpler questions into the candidate “homework” idea Yegge mentions early in his post.  Then we could preserve our 30 minutes of phone screen time for candidates who we already know have handled our easiest exercises.

Proudly powered by WordPress
Theme: Esquire by Matthew Buchanan.